{"id":"env-001","query":"What database does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses db-staging.internal for its primary application database.","tags":["staging","database","infra"],"depth":1},{"id":"e2","text":"The production environment uses db-prod.internal for its primary application database.","tags":["production","database","infra"],"depth":1},{"id":"e3","text":"The development environment uses db-dev.internal for its primary application database.","tags":["development","database","infra"],"depth":2},{"id":"e4","text":"The staging environment still keeps db-staging-readonly.internal for replaying migration dry runs.","tags":["staging","database","readonly"],"depth":2},{"id":"e5","text":"Before the analytics split, staging temporarily queried db-staging.internal for reporting jobs as well.","tags":["staging","database","historical"],"depth":3,"timestamp":"2025-10-17"},{"id":"e6","text":"The staging cutover checklist warns that db-shadow-staging.internal is only for verification and is not the live application database.","tags":["staging","database","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded early environment case with a read-only neighbor, historical shared use, and a shadow host distractor."}
{"id":"env-002","query":"What database does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses db-prod.internal for its primary application database.","tags":["production","database","infra"],"depth":1},{"id":"e2","text":"The staging environment uses db-staging.internal for its primary application database.","tags":["staging","database","infra"],"depth":1},{"id":"e3","text":"The failover environment uses db-dr.internal for the replicated application database.","tags":["failover","database","infra"],"depth":2},{"id":"e4","text":"Production analysts can replay old reads from db-prod-readonly.internal, which is not the live write database.","tags":["production","database","readonly"],"depth":2},{"id":"e5","text":"An archived migration worksheet still refers to db-main.internal from before the production database host was renamed.","tags":["production","database","historical"],"depth":3,"timestamp":"2025-08-09"},{"id":"e6","text":"The production runbook notes that db-prod-shadow.internal is only used during verification windows and should not answer normal traffic.","tags":["production","database","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded early environment case with failover, read-only, and stale hostname distractors instead of a simple env list."}
{"id":"env-003","query":"What database does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses db-dev.internal for its database.","tags":["development","database","infra"],"depth":1},{"id":"e2","text":"The production environment uses db-staging.internal for its database.","tags":["production","database","infra"],"depth":1},{"id":"e3","text":"The staging environment uses db-prod.internal for its database.","tags":["staging","database","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the database.","tags":["development","database","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-004","query":"What redis cache does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses redis-staging.internal for its redis cache.","tags":["staging","redis-cache","infra"],"depth":1},{"id":"e2","text":"The production environment uses redis-prod.internal for its redis cache.","tags":["production","redis-cache","infra"],"depth":1},{"id":"e3","text":"The development environment uses redis-dev.internal for its redis cache.","tags":["development","redis-cache","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the redis cache.","tags":["staging","redis-cache","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-005","query":"What redis cache does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses redis-prod.internal for its redis cache.","tags":["production","redis-cache","infra"],"depth":1},{"id":"e2","text":"The staging environment uses redis-staging.internal for its redis cache.","tags":["staging","redis-cache","infra"],"depth":1},{"id":"e3","text":"The development environment uses redis-dev.internal for its redis cache.","tags":["development","redis-cache","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the redis cache.","tags":["production","redis-cache","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-006","query":"What redis cache does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses redis-dev.internal for its redis cache.","tags":["development","redis-cache","infra"],"depth":1},{"id":"e2","text":"The production environment uses redis-staging.internal for its redis cache.","tags":["production","redis-cache","infra"],"depth":1},{"id":"e3","text":"The staging environment uses redis-prod.internal for its redis cache.","tags":["staging","redis-cache","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the redis cache.","tags":["development","redis-cache","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-007","query":"What object storage bucket does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses assets-staging for its object storage bucket.","tags":["staging","object-storage-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses assets-prod for its object storage bucket.","tags":["production","object-storage-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses assets-dev for its object storage bucket.","tags":["development","object-storage-bucket","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the object storage bucket.","tags":["staging","object-storage-bucket","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-008","query":"What object storage bucket does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses assets-prod for its object storage bucket.","tags":["production","object-storage-bucket","infra"],"depth":1},{"id":"e2","text":"The staging environment uses assets-staging for its object storage bucket.","tags":["staging","object-storage-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses assets-dev for its object storage bucket.","tags":["development","object-storage-bucket","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the object storage bucket.","tags":["production","object-storage-bucket","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-009","query":"What object storage bucket does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses assets-dev for its object storage bucket.","tags":["development","object-storage-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses assets-staging for its object storage bucket.","tags":["production","object-storage-bucket","infra"],"depth":1},{"id":"e3","text":"The staging environment uses assets-prod for its object storage bucket.","tags":["staging","object-storage-bucket","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the object storage bucket.","tags":["development","object-storage-bucket","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-010","query":"What email sender does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses ses-staging for its email sender.","tags":["staging","email-sender","infra"],"depth":1},{"id":"e2","text":"The production environment uses ses-prod for its email sender.","tags":["production","email-sender","infra"],"depth":1},{"id":"e3","text":"The development environment uses ses-dev for its email sender.","tags":["development","email-sender","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the email sender.","tags":["staging","email-sender","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-011","query":"What email sender does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses ses-prod for its email sender.","tags":["production","email-sender","infra"],"depth":1},{"id":"e2","text":"The staging environment uses ses-staging for its email sender.","tags":["staging","email-sender","infra"],"depth":1},{"id":"e3","text":"The development environment uses ses-dev for its email sender.","tags":["development","email-sender","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the email sender.","tags":["production","email-sender","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-012","query":"What email sender does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses ses-dev for its email sender.","tags":["development","email-sender","infra"],"depth":1},{"id":"e2","text":"The production environment uses ses-staging for its email sender.","tags":["production","email-sender","infra"],"depth":1},{"id":"e3","text":"The staging environment uses ses-prod for its email sender.","tags":["staging","email-sender","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the email sender.","tags":["development","email-sender","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-013","query":"What nats cluster does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses nats-staging.internal for its nats cluster.","tags":["staging","nats-cluster","infra"],"depth":1},{"id":"e2","text":"The production environment uses nats-prod.internal for its nats cluster.","tags":["production","nats-cluster","infra"],"depth":1},{"id":"e3","text":"The development environment uses nats-dev.internal for its nats cluster.","tags":["development","nats-cluster","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the nats cluster.","tags":["staging","nats-cluster","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-014","query":"What nats cluster does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses nats-prod.internal for its nats cluster.","tags":["production","nats-cluster","infra"],"depth":1},{"id":"e2","text":"The staging environment uses nats-staging.internal for its nats cluster.","tags":["staging","nats-cluster","infra"],"depth":1},{"id":"e3","text":"The development environment uses nats-dev.internal for its nats cluster.","tags":["development","nats-cluster","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the nats cluster.","tags":["production","nats-cluster","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-015","query":"What nats cluster does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses nats-dev.internal for its nats cluster.","tags":["development","nats-cluster","infra"],"depth":1},{"id":"e2","text":"The production environment uses nats-staging.internal for its nats cluster.","tags":["production","nats-cluster","infra"],"depth":1},{"id":"e3","text":"The staging environment uses nats-prod.internal for its nats cluster.","tags":["staging","nats-cluster","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the nats cluster.","tags":["development","nats-cluster","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-016","query":"What API base URL does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses https://api-staging.ecue.ai for its API base URL.","tags":["staging","api-base-url","infra"],"depth":1},{"id":"e2","text":"The production environment uses https://api.ecue.ai for its API base URL.","tags":["production","api-base-url","infra"],"depth":1},{"id":"e3","text":"The development environment uses https://api-dev.ecue.ai for its API base URL.","tags":["development","api-base-url","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the API base URL.","tags":["staging","api-base-url","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-017","query":"What API base URL does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses https://api.ecue.ai for its API base URL.","tags":["production","api-base-url","infra"],"depth":1},{"id":"e2","text":"The staging environment uses https://api-staging.ecue.ai for its API base URL.","tags":["staging","api-base-url","infra"],"depth":1},{"id":"e3","text":"The development environment uses https://api-dev.ecue.ai for its API base URL.","tags":["development","api-base-url","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the API base URL.","tags":["production","api-base-url","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-018","query":"What API base URL does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses https://api-dev.ecue.ai for its API base URL.","tags":["development","api-base-url","infra"],"depth":1},{"id":"e2","text":"The production environment uses https://api-staging.ecue.ai for its API base URL.","tags":["production","api-base-url","infra"],"depth":1},{"id":"e3","text":"The staging environment uses https://api.ecue.ai for its API base URL.","tags":["staging","api-base-url","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the API base URL.","tags":["development","api-base-url","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-019","query":"What metrics backend does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses vm-staging.internal for its metrics backend.","tags":["staging","metrics-backend","infra"],"depth":1},{"id":"e2","text":"The production environment uses vm-prod.internal for its metrics backend.","tags":["production","metrics-backend","infra"],"depth":1},{"id":"e3","text":"The development environment uses vm-dev.internal for its metrics backend.","tags":["development","metrics-backend","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the metrics backend.","tags":["staging","metrics-backend","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-020","query":"What metrics backend does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses vm-prod.internal for its metrics backend.","tags":["production","metrics-backend","infra"],"depth":1},{"id":"e2","text":"The staging environment uses vm-staging.internal for its metrics backend.","tags":["staging","metrics-backend","infra"],"depth":1},{"id":"e3","text":"The development environment uses vm-dev.internal for its metrics backend.","tags":["development","metrics-backend","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the metrics backend.","tags":["production","metrics-backend","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-021","query":"What metrics backend does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses vm-dev.internal for its metrics backend.","tags":["development","metrics-backend","infra"],"depth":1},{"id":"e2","text":"The production environment uses vm-staging.internal for its metrics backend.","tags":["production","metrics-backend","infra"],"depth":1},{"id":"e3","text":"The staging environment uses vm-prod.internal for its metrics backend.","tags":["staging","metrics-backend","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the metrics backend.","tags":["development","metrics-backend","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-022","query":"What auth issuer does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses https://auth-staging.ecue.ai for its auth issuer.","tags":["staging","auth-issuer","infra"],"depth":1},{"id":"e2","text":"The production environment uses https://auth.ecue.ai for its auth issuer.","tags":["production","auth-issuer","infra"],"depth":1},{"id":"e3","text":"The development environment uses https://auth-dev.ecue.ai for its auth issuer.","tags":["development","auth-issuer","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the auth issuer.","tags":["staging","auth-issuer","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-023","query":"What auth issuer does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses https://auth.ecue.ai for its auth issuer.","tags":["production","auth-issuer","infra"],"depth":1},{"id":"e2","text":"The staging environment uses https://auth-staging.ecue.ai for its auth issuer.","tags":["staging","auth-issuer","infra"],"depth":1},{"id":"e3","text":"The development environment uses https://auth-dev.ecue.ai for its auth issuer.","tags":["development","auth-issuer","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the auth issuer.","tags":["production","auth-issuer","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-024","query":"What auth issuer does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses https://auth-dev.ecue.ai for its auth issuer.","tags":["development","auth-issuer","infra"],"depth":1},{"id":"e2","text":"The production environment uses https://auth-staging.ecue.ai for its auth issuer.","tags":["production","auth-issuer","infra"],"depth":1},{"id":"e3","text":"The staging environment uses https://auth.ecue.ai for its auth issuer.","tags":["staging","auth-issuer","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the auth issuer.","tags":["development","auth-issuer","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-025","query":"What artifact bucket does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses artifacts-staging for its artifact bucket.","tags":["staging","artifact-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses artifacts-prod for its artifact bucket.","tags":["production","artifact-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses artifacts-dev for its artifact bucket.","tags":["development","artifact-bucket","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the artifact bucket.","tags":["staging","artifact-bucket","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-026","query":"What artifact bucket does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses artifacts-prod for its artifact bucket.","tags":["production","artifact-bucket","infra"],"depth":1},{"id":"e2","text":"The staging environment uses artifacts-staging for its artifact bucket.","tags":["staging","artifact-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses artifacts-dev for its artifact bucket.","tags":["development","artifact-bucket","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the artifact bucket.","tags":["production","artifact-bucket","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-027","query":"What artifact bucket does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses artifacts-dev for its artifact bucket.","tags":["development","artifact-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses artifacts-staging for its artifact bucket.","tags":["production","artifact-bucket","infra"],"depth":1},{"id":"e3","text":"The staging environment uses artifacts-prod for its artifact bucket.","tags":["staging","artifact-bucket","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the artifact bucket.","tags":["development","artifact-bucket","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-028","query":"What search index does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses meili-staging.internal for its search index.","tags":["staging","search-index","infra"],"depth":1},{"id":"e2","text":"The production environment uses meili-prod.internal for its search index.","tags":["production","search-index","infra"],"depth":1},{"id":"e3","text":"The development environment uses meili-dev.internal for its search index.","tags":["development","search-index","infra"],"depth":2},{"id":"e4","text":"The staging environment rotates credentials weekly for the search index.","tags":["staging","search-index","ops"],"depth":2},{"id":"e5","text":"The staging email sender is ses-staging for notification traffic.","tags":["staging","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-029","query":"What search index does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses meili-prod.internal for its search index.","tags":["production","search-index","infra"],"depth":1},{"id":"e2","text":"The staging environment uses meili-staging.internal for its search index.","tags":["staging","search-index","infra"],"depth":1},{"id":"e3","text":"The development environment uses meili-dev.internal for its search index.","tags":["development","search-index","infra"],"depth":2},{"id":"e4","text":"The production environment rotates credentials weekly for the search index.","tags":["production","search-index","ops"],"depth":2},{"id":"e5","text":"The production email sender is ses-production for notification traffic.","tags":["production","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"env-030","query":"What search index does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses meili-dev.internal for its search index.","tags":["development","search-index","infra"],"depth":1},{"id":"e2","text":"The production environment uses meili-staging.internal for its search index.","tags":["production","search-index","infra"],"depth":1},{"id":"e3","text":"The staging environment uses meili-prod.internal for its search index.","tags":["staging","search-index","infra"],"depth":2},{"id":"e4","text":"The development environment rotates credentials weekly for the search index.","tags":["development","search-index","ops"],"depth":2},{"id":"e5","text":"The development email sender is ses-development for notification traffic.","tags":["development","email","infra"],"depth":3}],"relevant_ids":["e1"],"notes":"Environment swap distractors with same service family."}
{"id":"envx-001","query":"What feature flag service does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses flags-staging.internal for the feature flag service.","tags":["staging","feature-flag-service","infra"],"depth":1},{"id":"e2","text":"The production environment uses flags-prod.internal for the feature flag service.","tags":["production","feature-flag-service","infra"],"depth":1},{"id":"e3","text":"The development environment uses flags-dev.internal for the feature flag service.","tags":["development","feature-flag-service","infra"],"depth":2},{"id":"e4","text":"Staging still mirrors historical flag snapshots through flags-staging-readonly.internal, which is not the live flag service.","tags":["staging","feature-flag-service","readonly"],"depth":2},{"id":"e5","text":"A rollout worksheet still refers to featureflags-staging.internal from before the host family was standardized.","tags":["staging","feature-flag-service","historical"],"depth":3,"timestamp":"2025-09-13"},{"id":"e6","text":"The staging runbook warns that flags-shadow-staging.internal is verification-only and should not serve normal flag reads.","tags":["staging","feature-flag-service","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra environment case with read-only and shadow-host distractors plus a stale alias."}
{"id":"envx-002","query":"What feature flag service does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses flags-prod.internal for the feature flag service.","tags":["production","feature-flag-service","infra"],"depth":1},{"id":"e2","text":"The staging environment uses flags-staging.internal for the feature flag service.","tags":["staging","feature-flag-service","infra"],"depth":1},{"id":"e3","text":"The failover environment uses flags-dr.internal for replicated feature flag reads.","tags":["failover","feature-flag-service","infra"],"depth":2},{"id":"e4","text":"Production can replay old flag states through flags-prod-readonly.internal, which is not the live flag service.","tags":["production","feature-flag-service","readonly"],"depth":2},{"id":"e5","text":"An archived cutover note still names flags-main.internal from before the production host rename.","tags":["production","feature-flag-service","historical"],"depth":3,"timestamp":"2025-08-06"},{"id":"e6","text":"The production runbook notes that flags-prod-shadow.internal is used only during validation windows.","tags":["production","feature-flag-service","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra environment case with failover, read-only, and stale-name distractors."}
{"id":"envx-003","query":"What artifact registry does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses registry-staging.internal for the artifact registry.","tags":["staging","artifact-registry","infra"],"depth":1},{"id":"e2","text":"The production environment uses registry-prod.internal for the artifact registry.","tags":["production","artifact-registry","infra"],"depth":1},{"id":"e3","text":"The development environment uses registry-dev.internal for the artifact registry.","tags":["development","artifact-registry","infra"],"depth":2},{"id":"e4","text":"Staging still exposes registry-staging-readonly.internal for replaying older package pulls, which is not the live artifact registry.","tags":["staging","artifact-registry","readonly"],"depth":2},{"id":"e5","text":"A release worksheet still lists artifacts-registry-staging.internal from before the host family rename.","tags":["staging","artifact-registry","historical"],"depth":3,"timestamp":"2025-09-04"},{"id":"e6","text":"The staging publish runbook notes that registry-shadow-staging.internal is validation-only and should not be used for normal pulls.","tags":["staging","artifact-registry","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra environment case with read-only, shadow-host, and stale-alias distractors."}
{"id":"envx-004","query":"What artifact registry does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses registry-prod.internal for the artifact registry.","tags":["production","artifact-registry","infra"],"depth":1},{"id":"e2","text":"The staging environment uses registry-staging.internal for the artifact registry.","tags":["staging","artifact-registry","infra"],"depth":1},{"id":"e3","text":"The failover environment uses registry-dr.internal for replicated package availability checks.","tags":["failover","artifact-registry","infra"],"depth":2},{"id":"e4","text":"Production can replay historical pulls through registry-prod-readonly.internal, which is not the live artifact registry.","tags":["production","artifact-registry","readonly"],"depth":2},{"id":"e5","text":"An archived deployment note still refers to registry-main.internal from before the production host rename.","tags":["production","artifact-registry","historical"],"depth":3,"timestamp":"2025-08-01"},{"id":"e6","text":"The production runbook says registry-shadow-prod.internal is used only for validation windows and should not serve ordinary package traffic.","tags":["production","artifact-registry","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra environment case with failover, read-only, and stale-name distractors."}
{"id":"envx-005","query":"What artifact registry does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses registry-dev.internal for the artifact registry.","tags":["development","artifact-registry","infra"],"depth":1},{"id":"e2","text":"The production environment uses registry-prod.internal for the artifact registry.","tags":["production","artifact-registry","infra"],"depth":1},{"id":"e3","text":"The staging environment uses registry-staging.internal for the artifact registry.","tags":["staging","artifact-registry","infra"],"depth":2},{"id":"e4","text":"Development still serves historical test pulls through registry-dev-readonly.internal, which is not the live artifact registry.","tags":["development","artifact-registry","readonly"],"depth":2},{"id":"e5","text":"A local bootstrap note still refers to dev-registry.internal from before the host family rename.","tags":["development","artifact-registry","historical"],"depth":3,"timestamp":"2025-08-25"},{"id":"e6","text":"The development runbook notes that registry-shadow-dev.internal is validation-only and should not be wired into normal package pulls.","tags":["development","artifact-registry","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded environment case with development-specific read-only, stale alias, and shadow-host clutter."}
{"id":"envx-006","query":"What workflow queue does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses queue-staging.internal for the workflow queue.","tags":["staging","workflow-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-prod.internal for the workflow queue.","tags":["production","workflow-queue","infra"],"depth":1},{"id":"e3","text":"The development environment uses queue-dev.internal for the workflow queue.","tags":["development","workflow-queue","infra"],"depth":2},{"id":"e4","text":"Staging drains replayed tasks through queue-staging-hold.internal, which is not the primary workflow queue.","tags":["staging","workflow-queue","hold-queue"],"depth":2},{"id":"e5","text":"A rollout worksheet still lists workflow-staging.internal from before the queue host rename.","tags":["staging","workflow-queue","historical"],"depth":3,"timestamp":"2025-09-02"},{"id":"e6","text":"The staging runbook warns that queue-shadow-staging.internal is only for verification and should not receive live task traffic.","tags":["staging","workflow-queue","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded environment case with hold-queue, shadow-host, and stale-name distractors."}
{"id":"envx-007","query":"What workflow queue does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses queue-prod.internal for the workflow queue.","tags":["production","workflow-queue","infra"],"depth":1},{"id":"e2","text":"The staging environment uses queue-staging.internal for the workflow queue.","tags":["staging","workflow-queue","infra"],"depth":1},{"id":"e3","text":"The development environment uses queue-dev.internal for the workflow queue.","tags":["development","workflow-queue","infra"],"depth":2},{"id":"e4","text":"The production runbook mentions how to rotate secrets for the workflow queue.","tags":["production","workflow-queue","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-008","query":"What workflow queue does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses queue-dev.internal for the workflow queue.","tags":["development","workflow-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-prod.internal for the workflow queue.","tags":["production","workflow-queue","infra"],"depth":1},{"id":"e3","text":"The staging environment uses queue-staging.internal for the workflow queue.","tags":["staging","workflow-queue","infra"],"depth":2},{"id":"e4","text":"The development runbook mentions how to rotate secrets for the workflow queue.","tags":["development","workflow-queue","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-009","query":"What analytics endpoint does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses analytics-staging.internal for the analytics endpoint.","tags":["staging","analytics-endpoint","infra"],"depth":1},{"id":"e2","text":"The production environment uses analytics-prod.internal for the analytics endpoint.","tags":["production","analytics-endpoint","infra"],"depth":1},{"id":"e3","text":"The development environment uses analytics-dev.internal for the analytics endpoint.","tags":["development","analytics-endpoint","infra"],"depth":2},{"id":"e4","text":"The staging runbook mentions how to rotate secrets for the analytics endpoint.","tags":["staging","analytics-endpoint","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-010","query":"What analytics endpoint does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses analytics-prod.internal for the analytics endpoint.","tags":["production","analytics-endpoint","infra"],"depth":1},{"id":"e2","text":"The staging environment uses analytics-staging.internal for the analytics endpoint.","tags":["staging","analytics-endpoint","infra"],"depth":1},{"id":"e3","text":"The development environment uses analytics-dev.internal for the analytics endpoint.","tags":["development","analytics-endpoint","infra"],"depth":2},{"id":"e4","text":"The production runbook mentions how to rotate secrets for the analytics endpoint.","tags":["production","analytics-endpoint","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-011","query":"What analytics endpoint does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses analytics-dev.internal for the analytics endpoint.","tags":["development","analytics-endpoint","infra"],"depth":1},{"id":"e2","text":"The production environment uses analytics-prod.internal for the analytics endpoint.","tags":["production","analytics-endpoint","infra"],"depth":1},{"id":"e3","text":"The staging environment uses analytics-staging.internal for the analytics endpoint.","tags":["staging","analytics-endpoint","infra"],"depth":2},{"id":"e4","text":"The development runbook mentions how to rotate secrets for the analytics endpoint.","tags":["development","analytics-endpoint","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-012","query":"What auth callback domain does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses callback-staging.ecue.ai for the auth callback domain.","tags":["staging","auth-callback-domain","infra"],"depth":1},{"id":"e2","text":"The production environment uses callback.ecue.ai for the auth callback domain.","tags":["production","auth-callback-domain","infra"],"depth":1},{"id":"e3","text":"The development environment uses callback-dev.ecue.ai for the auth callback domain.","tags":["development","auth-callback-domain","infra"],"depth":2},{"id":"e4","text":"The staging runbook mentions how to rotate secrets for the auth callback domain.","tags":["staging","auth-callback-domain","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-013","query":"What auth callback domain does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses callback.ecue.ai for the auth callback domain.","tags":["production","auth-callback-domain","infra"],"depth":1},{"id":"e2","text":"The staging environment uses callback-staging.ecue.ai for the auth callback domain.","tags":["staging","auth-callback-domain","infra"],"depth":1},{"id":"e3","text":"The development environment uses callback-dev.ecue.ai for the auth callback domain.","tags":["development","auth-callback-domain","infra"],"depth":2},{"id":"e4","text":"Production still keeps callback-readonly.ecue.ai for replaying historical auth handshakes, which is not the live callback domain.","tags":["production","auth-callback-domain","readonly"],"depth":2},{"id":"e5","text":"An old tenant migration note still refers to auth-callback.ecue.ai from before the production hostname was shortened.","tags":["production","auth-callback-domain","historical"],"depth":3,"timestamp":"2025-08-17"},{"id":"e6","text":"The production runbook warns that callback-shadow.ecue.ai is used only during verification windows and should not be handed to normal clients.","tags":["production","auth-callback-domain","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded environment case with read-only, stale hostname, and shadow-domain distractors."}
{"id":"envx-014","query":"What auth callback domain does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses callback-dev.ecue.ai for the auth callback domain.","tags":["development","auth-callback-domain","infra"],"depth":1},{"id":"e2","text":"The production environment uses callback.ecue.ai for the auth callback domain.","tags":["production","auth-callback-domain","infra"],"depth":1},{"id":"e3","text":"The staging environment uses callback-staging.ecue.ai for the auth callback domain.","tags":["staging","auth-callback-domain","infra"],"depth":2},{"id":"e4","text":"Development still routes old fixture traffic through callback-dev-readonly.ecue.ai, which is not the live callback domain used by normal local auth flows.","tags":["development","auth-callback-domain","readonly"],"depth":2},{"id":"e5","text":"A bootstrap note still mentions dev-callback.ecue.ai from before environment hostnames were aligned.","tags":["development","auth-callback-domain","historical"],"depth":3,"timestamp":"2025-09-08"},{"id":"e6","text":"The development runbook notes that callback-shadow-dev.ecue.ai is intended for verification runs only and should not be given to test clients by default.","tags":["development","auth-callback-domain","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded environment case with read-only, stale alias, and verification-only distractors."}
{"id":"envx-015","query":"What payments webhook host does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses payments-staging.internal for the payments webhook host.","tags":["staging","payments-webhook-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses payments-prod.internal for the payments webhook host.","tags":["production","payments-webhook-host","infra"],"depth":1},{"id":"e3","text":"The development environment uses payments-dev.internal for the payments webhook host.","tags":["development","payments-webhook-host","infra"],"depth":2},{"id":"e4","text":"The staging runbook mentions how to rotate secrets for the payments webhook host.","tags":["staging","payments-webhook-host","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-016","query":"What payments webhook host does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses payments-prod.internal for the payments webhook host.","tags":["production","payments-webhook-host","infra"],"depth":1},{"id":"e2","text":"The staging environment uses payments-staging.internal for the payments webhook host.","tags":["staging","payments-webhook-host","infra"],"depth":1},{"id":"e3","text":"The development environment uses payments-dev.internal for the payments webhook host.","tags":["development","payments-webhook-host","infra"],"depth":2},{"id":"e4","text":"The production runbook mentions how to rotate secrets for the payments webhook host.","tags":["production","payments-webhook-host","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-017","query":"What payments webhook host does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses payments-dev.internal for the payments webhook host.","tags":["development","payments-webhook-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses payments-prod.internal for the payments webhook host.","tags":["production","payments-webhook-host","infra"],"depth":1},{"id":"e3","text":"The staging environment uses payments-staging.internal for the payments webhook host.","tags":["staging","payments-webhook-host","infra"],"depth":2},{"id":"e4","text":"The development runbook mentions how to rotate secrets for the payments webhook host.","tags":["development","payments-webhook-host","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-018","query":"What logs bucket does staging use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The staging environment uses logs-staging for the logs bucket.","tags":["staging","logs-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses logs-prod for the logs bucket.","tags":["production","logs-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses logs-dev for the logs bucket.","tags":["development","logs-bucket","infra"],"depth":2},{"id":"e4","text":"The staging runbook mentions how to rotate secrets for the logs bucket.","tags":["staging","logs-bucket","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-019","query":"What logs bucket does production use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The production environment uses logs-prod for the logs bucket.","tags":["production","logs-bucket","infra"],"depth":1},{"id":"e2","text":"The staging environment uses logs-staging for the logs bucket.","tags":["staging","logs-bucket","infra"],"depth":1},{"id":"e3","text":"The development environment uses logs-dev for the logs bucket.","tags":["development","logs-bucket","infra"],"depth":2},{"id":"e4","text":"The production runbook mentions how to rotate secrets for the logs bucket.","tags":["production","logs-bucket","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"envx-020","query":"What logs bucket does development use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The development environment uses logs-dev for the logs bucket.","tags":["development","logs-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses logs-prod for the logs bucket.","tags":["production","logs-bucket","infra"],"depth":1},{"id":"e3","text":"The staging environment uses logs-staging for the logs bucket.","tags":["staging","logs-bucket","infra"],"depth":2},{"id":"e4","text":"The development runbook mentions how to rotate secrets for the logs bucket.","tags":["development","logs-bucket","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra environment confusion case."}
{"id":"entity-001","query":"What milk does Caroline prefer during release checklist mornings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Caroline prefers oat milk during release checklist mornings.","tags":["caroline","release-checklists","preference"],"depth":1},{"id":"e2","text":"Catherine prefers oat milk during release checklist mornings.","tags":["catherine","release-checklists","preference"],"depth":1},{"id":"e3","text":"Caroline switches to black tea on migration night watches instead of release checklist mornings.","tags":["caroline","migration-night-watches","preference"],"depth":2},{"id":"e4","text":"Coraline prefers oat milk during architecture note reviews, not release checklist mornings.","tags":["coraline","architecture-notes","preference"],"depth":2},{"id":"e5","text":"A kitchen order note said Caroline and Catherine both avoided dairy on long release days.","tags":["caroline","catherine","kitchen-note"],"depth":3},{"id":"e6","text":"Caroline leaves the oat milk carton beside the printed release checklist before anyone else arrives.","tags":["caroline","release-checklists","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded early entity case with a more specific context, overlapping same-value distractors, and a neighboring routine."}
{"id":"entity-002","query":"What desk setup does Jordan prefer during incident review weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Jordan prefers a standing desk during incident review weeks.","tags":["jordan","incident-review-weeks","preference"],"depth":1},{"id":"e2","text":"Jordyn prefers a standing desk during incident review weeks.","tags":["jordyn","incident-review-weeks","preference"],"depth":1},{"id":"e3","text":"Jordan switches to an ergonomic chair during documentation days instead of incident review weeks.","tags":["jordan","documentation-days","preference"],"depth":2},{"id":"e4","text":"Jorian prefers a standing desk during postmortem workshops, not incident review weeks.","tags":["jorian","postmortem-workshops","preference"],"depth":2},{"id":"e5","text":"An office supply thread said Jordan and Jordyn both requested anti-fatigue mats for long review sessions.","tags":["jordan","jordyn","office-supplies"],"depth":3},{"id":"e6","text":"Jordan keeps the incident review binder propped on the standing desk rail during escalation meetings.","tags":["jordan","incident-review-weeks","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded early entity case with realistic work-mode switching and a near-name same-value distractor."}
{"id":"entity-003","query":"What database does Alicia prefer during schema repair sessions?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Alicia prefers Postgres during schema repair sessions.","tags":["alicia","schema-repair","preference"],"depth":1},{"id":"e2","text":"Alice prefers Postgres during schema repair sessions.","tags":["alice","schema-repair","preference"],"depth":1},{"id":"e3","text":"Alicia switches to MySQL during customer import debugging instead of schema repair sessions.","tags":["alicia","import-debugging","preference"],"depth":2},{"id":"e4","text":"Alina prefers Postgres during migration retros, not schema repair sessions.","tags":["alina","migration-retros","preference"],"depth":2},{"id":"e5","text":"A database tooling note said Alicia and Alice both wanted Postgres shortcuts synchronized across their editors.","tags":["alicia","alice","tooling-note"],"depth":3},{"id":"e6","text":"Alicia keeps a printed Postgres recovery checklist in the front sleeve of the schema repair binder.","tags":["alicia","schema-repair","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a concrete work context and same-value distractors tied to adjacent database tasks."}
{"id":"entity-004","query":"What editor does Mika prefer during terminal-only incident shifts?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mika prefers vim during terminal-only incident shifts.","tags":["mika","incident-shifts","preference"],"depth":1},{"id":"e2","text":"Mila prefers vim during terminal-only incident shifts.","tags":["mila","incident-shifts","preference"],"depth":1},{"id":"e3","text":"Mika switches to emacs during doc-heavy review days instead of terminal-only incident shifts.","tags":["mika","review-days","preference"],"depth":2},{"id":"e4","text":"Mina prefers vim during overnight deploy watches, not terminal-only incident shifts.","tags":["mina","deploy-watches","preference"],"depth":2},{"id":"e5","text":"A tooling thread said Mika and Mila both kept matching vim keybind sheets on their desks.","tags":["mika","mila","tooling-thread"],"depth":3},{"id":"e6","text":"Mika keeps a printed vim command card clipped to the incident terminal cart.","tags":["mika","incident-shifts","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a stronger situational cue and same-value near-name editor confusion."}
{"id":"entity-005","query":"What theme does Darren prefer during overnight monitoring weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Darren prefers dark mode during overnight monitoring weeks.","tags":["darren","overnight-monitoring","preference"],"depth":1},{"id":"e2","text":"Dorian prefers dark mode during overnight monitoring weeks.","tags":["dorian","overnight-monitoring","preference"],"depth":1},{"id":"e3","text":"Darren switches to light mode during daytime dashboard workshops instead of overnight monitoring weeks.","tags":["darren","dashboard-workshops","preference"],"depth":2},{"id":"e4","text":"Darian prefers dark mode during incident review drills, not overnight monitoring weeks.","tags":["darian","incident-drills","preference"],"depth":2},{"id":"e5","text":"A workspace note said Darren and Dorian both dim their monitors aggressively during long night shifts.","tags":["darren","dorian","workspace-note"],"depth":3},{"id":"e6","text":"Darren keeps the monitoring wall on dark mode even when the rest of the room is brighter.","tags":["darren","overnight-monitoring","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with overnight-context grounding and same-value near-name distractors."}
{"id":"entity-006","query":"What API style does Riley prefer during frontend integration weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Riley prefers GraphQL during frontend integration weeks.","tags":["riley","frontend-integration","preference"],"depth":1},{"id":"e2","text":"Ryan prefers GraphQL during frontend integration weeks.","tags":["ryan","frontend-integration","preference"],"depth":1},{"id":"e3","text":"Riley switches to REST during public webhook rollouts instead of frontend integration weeks.","tags":["riley","webhook-rollouts","preference"],"depth":2},{"id":"e4","text":"Rylan prefers GraphQL during mobile schema syncs, not frontend integration weeks.","tags":["rylan","mobile-syncs","preference"],"depth":2},{"id":"e5","text":"An API guild note said Riley and Ryan both wanted GraphQL fragments standardized across their review examples.","tags":["riley","ryan","api-guild"],"depth":3},{"id":"e6","text":"Riley keeps a printed GraphQL schema snapshot open during every integration review.","tags":["riley","frontend-integration","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with concrete API-workflow context and same-value near-name distractors."}
{"id":"entity-007","query":"What check-in rhythm does Marina prefer during launch weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Marina prefers daily standups during launch weeks.","tags":["marina","launch-weeks","preference"],"depth":1},{"id":"e2","text":"Maria prefers daily standups during launch weeks.","tags":["maria","launch-weeks","preference"],"depth":1},{"id":"e3","text":"Marina switches to weekly status docs during maintenance stretches instead of launch weeks.","tags":["marina","maintenance-stretches","preference"],"depth":2},{"id":"e4","text":"Maren prefers daily standups during hiring sprints, not launch weeks.","tags":["maren","hiring-sprints","preference"],"depth":2},{"id":"e5","text":"A planning note said Marina and Maria both preferred short standing check-ins when deadlines were near.","tags":["marina","maria","planning-note"],"depth":3},{"id":"e6","text":"Marina pins the launch board agenda to the top of the standup notes every morning.","tags":["marina","launch-weeks","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a more specific project rhythm and same-value near-name overlap."}
{"id":"entity-008","query":"What build setup does Talia prefer during service bootstrap work?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Talia prefers Go modules during service bootstrap work.","tags":["talia","service-bootstrap","preference"],"depth":1},{"id":"e2","text":"Tanya prefers Go modules during service bootstrap work.","tags":["tanya","service-bootstrap","preference"],"depth":1},{"id":"e3","text":"Talia switches to Bazel workspaces during monorepo integration work instead of service bootstrap work.","tags":["talia","monorepo-integration","preference"],"depth":2},{"id":"e4","text":"Tahlia prefers Go modules during CLI packaging reviews, not service bootstrap work.","tags":["tahlia","cli-packaging","preference"],"depth":2},{"id":"e5","text":"A tooling note said Talia and Tanya both preferred module-based examples when onboarding new service owners.","tags":["talia","tanya","tooling-note"],"depth":3},{"id":"e6","text":"Talia keeps a short Go module init checklist in the front pocket of the bootstrap folder.","tags":["talia","service-bootstrap","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with concrete build-context switching and same-value near-name confusion."}
{"id":"entity-009","query":"What does Nolan prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Nolan prefers CUDA containers for daily work.","tags":["nolan","preference"],"depth":1},{"id":"e2","text":"Noah prefers CUDA containers for daily work.","tags":["noah","preference"],"depth":1},{"id":"e3","text":"Nolan also likes CPU-only builds in some situations.","tags":["nolan","preference"],"depth":2},{"id":"e4","text":"The team discussed CUDA containers and CPU-only builds during planning.","tags":["team","planning"],"depth":3},{"id":"e5","text":"Nolan wrote a migration note about deployment safety.","tags":["nolan","deployment"],"depth":3}],"relevant_ids":["e1"],"notes":"Entity swap with near-name collision and shared preference value."}
{"id":"entity-010","query":"What does Elena prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Elena prefers Linear for daily work.","tags":["elena","preference"],"depth":1},{"id":"e2","text":"Alina prefers Linear for daily work.","tags":["alina","preference"],"depth":1},{"id":"e3","text":"Elena also likes Jira in some situations.","tags":["elena","preference"],"depth":2},{"id":"e4","text":"The team discussed Linear and Jira during planning.","tags":["team","planning"],"depth":3},{"id":"e5","text":"Elena wrote a migration note about deployment safety.","tags":["elena","deployment"],"depth":3}],"relevant_ids":["e1"],"notes":"Entity swap with near-name collision and shared preference value."}
{"id":"entity-011","query":"What does Sonia prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Sonia prefers dark roast for daily work.","tags":["sonia","preference"],"depth":1},{"id":"e2","text":"Sonya prefers dark roast for daily work.","tags":["sonya","preference"],"depth":1},{"id":"e3","text":"Sonia also likes green tea in some situations.","tags":["sonia","preference"],"depth":2},{"id":"e4","text":"The team discussed dark roast and green tea during planning.","tags":["team","planning"],"depth":3},{"id":"e5","text":"Sonia wrote a migration note about deployment safety.","tags":["sonia","deployment"],"depth":3}],"relevant_ids":["e1"],"notes":"Entity swap with near-name collision and shared preference value."}
{"id":"entity-012","query":"What does Maren prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Maren prefers Nix for daily work.","tags":["maren","preference"],"depth":1},{"id":"e2","text":"Karen prefers Nix for daily work.","tags":["karen","preference"],"depth":1},{"id":"e3","text":"Maren also likes Docker Compose in some situations.","tags":["maren","preference"],"depth":2},{"id":"e4","text":"The team discussed Nix and Docker Compose during planning.","tags":["team","planning"],"depth":3},{"id":"e5","text":"Maren wrote a migration note about deployment safety.","tags":["maren","deployment"],"depth":3}],"relevant_ids":["e1"],"notes":"Entity swap with near-name collision and shared preference value."}
{"id":"entityx-001","query":"What lunch does Colin prefer during migration retro days?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Colin prefers sourdough bread during migration retro days.","tags":["colin","migration-retros","preference"],"depth":1},{"id":"e2","text":"Collin prefers sourdough bread during migration retro days.","tags":["collin","migration-retros","preference"],"depth":1},{"id":"e3","text":"Colin switches to rice noodles during late-night import repairs instead of migration retro days.","tags":["colin","import-repairs","preference"],"depth":2},{"id":"e4","text":"Colan prefers sourdough bread during postmortem lunches, not migration retro days.","tags":["colan","postmortems","preference"],"depth":2},{"id":"e5","text":"A team lunch note said Colin and Collin both preferred plain breads that did not crumble onto printouts.","tags":["colin","collin","lunch-note"],"depth":3},{"id":"e6","text":"Colin keeps the sourdough wrapped separately from the printed retro agenda.","tags":["colin","migration-retros","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra entity case with specific work context and same-item near-name distractors."}
{"id":"entityx-002","query":"What tracker does Amira prefer during release coordination weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Amira prefers Linear during release coordination weeks.","tags":["amira","release-coordination","preference"],"depth":1},{"id":"e2","text":"Amara prefers Linear during release coordination weeks.","tags":["amara","release-coordination","preference"],"depth":1},{"id":"e3","text":"Amira switches to a spreadsheet backlog during vendor import cleanup instead of release coordination weeks.","tags":["amira","vendor-imports","preference"],"depth":2},{"id":"e4","text":"Samira prefers Linear during launch readiness reviews, not release coordination weeks.","tags":["samira","launch-readiness","preference"],"depth":2},{"id":"e5","text":"A project ops note said Amira and Amara both preferred Linear views with due dates visible during high-tempo weeks.","tags":["amira","amara","project-ops"],"depth":3},{"id":"e6","text":"Amira keeps the release blocker board pinned in Linear throughout coordination week.","tags":["amira","release-coordination","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra entity case with a clearer operational context and same-tool near-name overlap."}
{"id":"entityx-003","query":"What orchestration platform does Devon prefer during cluster recovery drills?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Devon prefers Kubernetes during cluster recovery drills.","tags":["devon","cluster-recovery","preference"],"depth":1},{"id":"e2","text":"Devin prefers Kubernetes during cluster recovery drills.","tags":["devin","cluster-recovery","preference"],"depth":1},{"id":"e3","text":"Devon switches to Nomad during lightweight sandbox experiments instead of cluster recovery drills.","tags":["devon","sandbox-experiments","preference"],"depth":2},{"id":"e4","text":"Davin prefers Kubernetes during ingress migration reviews, not cluster recovery drills.","tags":["davin","ingress-migrations","preference"],"depth":2},{"id":"e5","text":"A platform note said Devon and Devin both preferred Kubernetes dashboards with pod restarts visible during incident drills.","tags":["devon","devin","platform-note"],"depth":3},{"id":"e6","text":"Devon keeps the cluster recovery runbook open beside the Kubernetes node view during every drill.","tags":["devon","cluster-recovery","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra entity case with concrete ops context and same-platform near-name distractors."}
{"id":"entityx-004","query":"What frontend stack does Selena prefer during reliability dashboard rewrites?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Selena prefers Elm during reliability dashboard rewrites.","tags":["selena","dashboard-rewrites","preference"],"depth":1},{"id":"e2","text":"Sienna prefers Elm during reliability dashboard rewrites.","tags":["sienna","dashboard-rewrites","preference"],"depth":1},{"id":"e3","text":"Selena switches to React during marketing microsite work instead of reliability dashboard rewrites.","tags":["selena","marketing-microsites","preference"],"depth":2},{"id":"e4","text":"Celena prefers Elm during design system audits, not reliability dashboard rewrites.","tags":["celena","design-system-audits","preference"],"depth":2},{"id":"e5","text":"A frontend guild note said Selena and Sienna both preferred Elm examples when discussing state-machine-heavy screens.","tags":["selena","sienna","frontend-guild"],"depth":3},{"id":"e6","text":"Selena keeps an Elm architecture sketch clipped to the dashboard rewrite checklist.","tags":["selena","dashboard-rewrites","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded extra entity case with a concrete UI context and same-stack near-name distractors."}
{"id":"entityx-005","query":"What meeting style does Tristan prefer during incident coordination weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Tristan prefers standing meetings during incident coordination weeks.","tags":["tristan","incident-coordination","preference"],"depth":1},{"id":"e2","text":"Tristen prefers standing meetings during incident coordination weeks.","tags":["tristen","incident-coordination","preference"],"depth":1},{"id":"e3","text":"Tristan switches to async updates during quiet maintenance windows instead of incident coordination weeks.","tags":["tristan","maintenance-windows","preference"],"depth":2},{"id":"e4","text":"Tristam prefers standing meetings during launch war rooms, not incident coordination weeks.","tags":["tristam","launch-war-rooms","preference"],"depth":2},{"id":"e5","text":"A planning note said Tristan and Tristen both preferred short standups when incidents created too many parallel threads.","tags":["tristan","tristen","planning-note"],"depth":3},{"id":"e6","text":"Tristan keeps the incident bridge agenda printed because standing meetings tend to drift without it.","tags":["tristan","incident-coordination","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a concrete coordination context and same-style near-name distractors."}
{"id":"entityx-006","query":"What drink does Nadia prefer during early benchmark triage mornings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Nadia prefers matcha during early benchmark triage mornings.","tags":["nadia","benchmark-triage","preference"],"depth":1},{"id":"e2","text":"Naomi prefers matcha during early benchmark triage mornings.","tags":["naomi","benchmark-triage","preference"],"depth":1},{"id":"e3","text":"Nadia switches to espresso during late-night release validation instead of early benchmark triage mornings.","tags":["nadia","release-validation","preference"],"depth":2},{"id":"e4","text":"Nadine prefers matcha during docs review sprints, not benchmark triage mornings.","tags":["nadine","docs-sprints","preference"],"depth":2},{"id":"e5","text":"A kitchen restock note said Nadia and Naomi both preferred matcha packets that dissolved quickly during rushed mornings.","tags":["nadia","naomi","kitchen-note"],"depth":3},{"id":"e6","text":"Nadia keeps a matcha tin beside the printed retrieval error buckets before triage starts.","tags":["nadia","benchmark-triage","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a more believable workplace routine and same-drink near-name overlap."}
{"id":"entityx-007","query":"What network tool does Harper prefer during remote access cleanup weeks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Harper prefers Tailscale during remote access cleanup weeks.","tags":["harper","remote-access-cleanup","preference"],"depth":1},{"id":"e2","text":"Harold prefers Tailscale during remote access cleanup weeks.","tags":["harold","remote-access-cleanup","preference"],"depth":1},{"id":"e3","text":"Harper switches to WireGuard during low-level appliance debugging instead of remote access cleanup weeks.","tags":["harper","appliance-debugging","preference"],"depth":2},{"id":"e4","text":"Harlin prefers Tailscale during contractor laptop rollouts, not remote access cleanup weeks.","tags":["harlin","laptop-rollouts","preference"],"depth":2},{"id":"e5","text":"A network tooling note said Harper and Harold both preferred Tailscale because it reduced manual peer config on temporary machines.","tags":["harper","harold","network-tooling"],"depth":3},{"id":"e6","text":"Harper keeps the remote access revocation checklist open beside the Tailscale admin screen during cleanup week.","tags":["harper","remote-access-cleanup","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with realistic network-ops context and same-tool near-name distractors."}
{"id":"entityx-008","query":"What database does Leona prefer during reporting pipeline repairs?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Leona prefers PostgreSQL during reporting pipeline repairs.","tags":["leona","reporting-pipelines","preference"],"depth":1},{"id":"e2","text":"Lenora prefers PostgreSQL during reporting pipeline repairs.","tags":["lenora","reporting-pipelines","preference"],"depth":1},{"id":"e3","text":"Leona switches to SQLite during small local fixture reproductions instead of reporting pipeline repairs.","tags":["leona","local-fixtures","preference"],"depth":2},{"id":"e4","text":"Leora prefers PostgreSQL during analytics backfill reviews, not reporting pipeline repairs.","tags":["leora","analytics-backfills","preference"],"depth":2},{"id":"e5","text":"A data tooling note said Leona and Lenora both preferred PostgreSQL when the work involved lock behavior, query plans, or staging restores.","tags":["leona","lenora","data-tooling"],"depth":3},{"id":"e6","text":"Leona keeps a printed checklist for broken reporting indexes tucked into the PostgreSQL incident binder.","tags":["leona","reporting-pipelines","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with realistic database-workflow context and same-value near-name distractors."}
{"id":"entityx-009","query":"What demo rhythm does Mason prefer during launch stabilization months?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mason prefers weekly demos during launch stabilization months.","tags":["mason","launch-stabilization","preference"],"depth":1},{"id":"e2","text":"Marlon prefers weekly demos during launch stabilization months.","tags":["marlon","launch-stabilization","preference"],"depth":1},{"id":"e3","text":"Mason switches to monthly demos during quieter platform maintenance stretches instead of launch stabilization months.","tags":["mason","platform-maintenance","preference"],"depth":2},{"id":"e4","text":"Masen prefers weekly demos during customer pilot reviews, not launch stabilization months.","tags":["masen","customer-pilots","preference"],"depth":2},{"id":"e5","text":"A release planning note said Mason and Marlon both preferred weekly demos when new bugs were still being surfaced in front of stakeholders.","tags":["mason","marlon","release-planning"],"depth":3},{"id":"e6","text":"Mason keeps the demo issue ledger open next to the weekly review deck throughout stabilization season.","tags":["mason","launch-stabilization","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with realistic launch rhythm and same-value near-name distractors."}
{"id":"entityx-010","query":"What editor theme does Tessa prefer during overnight dashboard review?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Tessa prefers dark theme during overnight dashboard review.","tags":["tessa","dashboard-review","preference"],"depth":1},{"id":"e2","text":"Teresa prefers dark theme during overnight dashboard review.","tags":["teresa","dashboard-review","preference"],"depth":1},{"id":"e3","text":"Tessa switches to Solarized Light during daylight docs polish instead of overnight dashboard review.","tags":["tessa","docs-polish","preference"],"depth":2},{"id":"e4","text":"Tessia prefers dark theme during release note assembly, not overnight dashboard review.","tags":["tessia","release-notes","preference"],"depth":2},{"id":"e5","text":"A UI tooling note said Tessa and Teresa both preferred darker color schemes when triaging bright monitoring screens at night.","tags":["tessa","teresa","ui-tooling"],"depth":3},{"id":"e6","text":"Tessa keeps the dashboard color contrast checklist clipped to the dark-theme review screenshots.","tags":["tessa","dashboard-review","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Upgraded entity case with a realistic night-review context and same-theme near-name distractors."}
{"id":"entityx-011","query":"What does Gideon prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Gideon prefers helm charts for everyday use.","tags":["gideon","preference"],"depth":1},{"id":"e2","text":"Gillian prefers helm charts for everyday use.","tags":["gillian","preference"],"depth":1},{"id":"e3","text":"Gideon sometimes uses terraform modules instead.","tags":["gideon","preference"],"depth":2},{"id":"e4","text":"The team discussed helm charts and terraform modules during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-012","query":"What does Priya prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Priya prefers quiet keyboards for everyday use.","tags":["priya","preference"],"depth":1},{"id":"e2","text":"Priyanka prefers quiet keyboards for everyday use.","tags":["priyanka","preference"],"depth":1},{"id":"e3","text":"Priya sometimes uses clicky keyboards instead.","tags":["priya","preference"],"depth":2},{"id":"e4","text":"The team discussed quiet keyboards and clicky keyboards during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-013","query":"What does Evan prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Evan prefers gpu builds for everyday use.","tags":["evan","preference"],"depth":1},{"id":"e2","text":"Ivan prefers gpu builds for everyday use.","tags":["ivan","preference"],"depth":1},{"id":"e3","text":"Evan sometimes uses cpu builds instead.","tags":["evan","preference"],"depth":2},{"id":"e4","text":"The team discussed gpu builds and cpu builds during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-014","query":"What does Lena prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Lena prefers shortbread cookies for everyday use.","tags":["lena","preference"],"depth":1},{"id":"e2","text":"Lina prefers shortbread cookies for everyday use.","tags":["lina","preference"],"depth":1},{"id":"e3","text":"Lena sometimes uses ginger cookies instead.","tags":["lena","preference"],"depth":2},{"id":"e4","text":"The team discussed shortbread cookies and ginger cookies during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-015","query":"What does Marco prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Marco prefers clickhouse for everyday use.","tags":["marco","preference"],"depth":1},{"id":"e2","text":"Marek prefers clickhouse for everyday use.","tags":["marek","preference"],"depth":1},{"id":"e3","text":"Marco sometimes uses bigquery instead.","tags":["marco","preference"],"depth":2},{"id":"e4","text":"The team discussed clickhouse and bigquery during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-016","query":"What does Keira prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Keira prefers rss feeds for everyday use.","tags":["keira","preference"],"depth":1},{"id":"e2","text":"Kiera prefers rss feeds for everyday use.","tags":["kiera","preference"],"depth":1},{"id":"e3","text":"Keira sometimes uses email newsletters instead.","tags":["keira","preference"],"depth":2},{"id":"e4","text":"The team discussed rss feeds and email newsletters during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-017","query":"What does Brennan prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Brennan prefers obsidian for everyday use.","tags":["brennan","preference"],"depth":1},{"id":"e2","text":"Brendan prefers obsidian for everyday use.","tags":["brendan","preference"],"depth":1},{"id":"e3","text":"Brennan sometimes uses logseq instead.","tags":["brennan","preference"],"depth":2},{"id":"e4","text":"The team discussed obsidian and logseq during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-018","query":"What does Farah prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Farah prefers blueberry yogurt for everyday use.","tags":["farah","preference"],"depth":1},{"id":"e2","text":"Fiona prefers blueberry yogurt for everyday use.","tags":["fiona","preference"],"depth":1},{"id":"e3","text":"Farah sometimes uses plain yogurt instead.","tags":["farah","preference"],"depth":2},{"id":"e4","text":"The team discussed blueberry yogurt and plain yogurt during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-019","query":"What does Noelle prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Noelle prefers go test for everyday use.","tags":["noelle","preference"],"depth":1},{"id":"e2","text":"Noel prefers go test for everyday use.","tags":["noel","preference"],"depth":1},{"id":"e3","text":"Noelle sometimes uses pytest instead.","tags":["noelle","preference"],"depth":2},{"id":"e4","text":"The team discussed go test and pytest during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"entityx-020","query":"What does Soren prefer?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Soren prefers daily walks for everyday use.","tags":["soren","preference"],"depth":1},{"id":"e2","text":"Sorin prefers daily walks for everyday use.","tags":["sorin","preference"],"depth":1},{"id":"e3","text":"Soren sometimes uses stationary bike instead.","tags":["soren","preference"],"depth":2},{"id":"e4","text":"The team discussed daily walks and stationary bike during planning.","tags":["team","planning"],"depth":3}],"relevant_ids":["e1"],"notes":"Extra entity confusion case."}
{"id":"time-001","query":"When did we migrate auth to bearer tokens?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We migrate auth to bearer tokens in March.","tags":["timeline","migrate-auth-to-bearer-tokens"],"depth":1,"timestamp":"2026-03-03"},{"id":"e2","text":"We migrate auth to bearer tokens in May.","tags":["timeline","migrate-auth-to-bearer-tokens"],"depth":1,"timestamp":"2026-05-04"},{"id":"e3","text":"We discussed how to migrate auth to bearer tokens throughout March planning.","tags":["planning","migrate-auth-to-bearer-tokens"],"depth":2},{"id":"e4","text":"We completed documentation for that change in May.","tags":["docs","migrate-auth-to-bearer-tokens"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-002","query":"When did we split staging and production domains?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We split staging and production domains in April.","tags":["timeline","split-staging-and-production-domains"],"depth":1,"timestamp":"2026-04-03"},{"id":"e2","text":"We split staging and production domains in June.","tags":["timeline","split-staging-and-production-domains"],"depth":1,"timestamp":"2026-06-04"},{"id":"e3","text":"We discussed how to split staging and production domains throughout April planning.","tags":["planning","split-staging-and-production-domains"],"depth":2},{"id":"e4","text":"We completed documentation for that change in June.","tags":["docs","split-staging-and-production-domains"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-003","query":"When did we move Redis into a private subnet?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We move Redis into a private subnet in January.","tags":["timeline","move-redis-into-a-private-subnet"],"depth":1,"timestamp":"2026-01-03"},{"id":"e2","text":"We move Redis into a private subnet in February.","tags":["timeline","move-redis-into-a-private-subnet"],"depth":1,"timestamp":"2026-02-04"},{"id":"e3","text":"We discussed how to move Redis into a private subnet throughout January planning.","tags":["planning","move-redis-into-a-private-subnet"],"depth":2},{"id":"e4","text":"We completed documentation for that change in February.","tags":["docs","move-redis-into-a-private-subnet"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-004","query":"When did we ship the reporting dashboard?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We ship the reporting dashboard in July.","tags":["timeline","ship-the-reporting-dashboard"],"depth":1,"timestamp":"2026-07-03"},{"id":"e2","text":"We ship the reporting dashboard in August.","tags":["timeline","ship-the-reporting-dashboard"],"depth":1,"timestamp":"2026-08-04"},{"id":"e3","text":"We discussed how to ship the reporting dashboard throughout July planning.","tags":["planning","ship-the-reporting-dashboard"],"depth":2},{"id":"e4","text":"We completed documentation for that change in August.","tags":["docs","ship-the-reporting-dashboard"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-005","query":"When did we enable GPU inference in Docker?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We enable GPU inference in Docker in September.","tags":["timeline","enable-gpu-inference-in-docker"],"depth":1,"timestamp":"2026-09-03"},{"id":"e2","text":"We enable GPU inference in Docker in October.","tags":["timeline","enable-gpu-inference-in-docker"],"depth":1,"timestamp":"2026-10-04"},{"id":"e3","text":"We discussed how to enable GPU inference in Docker throughout September planning.","tags":["planning","enable-gpu-inference-in-docker"],"depth":2},{"id":"e4","text":"We completed documentation for that change in October.","tags":["docs","enable-gpu-inference-in-docker"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-006","query":"When did we rotate the SES credentials?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rotate the SES credentials in November.","tags":["timeline","rotate-the-ses-credentials"],"depth":1,"timestamp":"2026-11-03"},{"id":"e2","text":"We rotate the SES credentials in December.","tags":["timeline","rotate-the-ses-credentials"],"depth":1,"timestamp":"2026-12-04"},{"id":"e3","text":"We discussed how to rotate the SES credentials throughout November planning.","tags":["planning","rotate-the-ses-credentials"],"depth":2},{"id":"e4","text":"We completed documentation for that change in December.","tags":["docs","rotate-the-ses-credentials"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-007","query":"When did we switch the default embedded model to bge-small?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We switch the default embedded model to bge-small in February.","tags":["timeline","switch-the-default-embedded-model-to-bge-small"],"depth":1,"timestamp":"2026-02-03"},{"id":"e2","text":"We switch the default embedded model to bge-small in April.","tags":["timeline","switch-the-default-embedded-model-to-bge-small"],"depth":1,"timestamp":"2026-04-04"},{"id":"e3","text":"We discussed how to switch the default embedded model to bge-small throughout February planning.","tags":["planning","switch-the-default-embedded-model-to-bge-small"],"depth":2},{"id":"e4","text":"We completed documentation for that change in April.","tags":["docs","switch-the-default-embedded-model-to-bge-small"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-008","query":"When did we publish the public GHCR image?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We publish the public GHCR image in May.","tags":["timeline","publish-the-public-ghcr-image"],"depth":1,"timestamp":"2026-05-03"},{"id":"e2","text":"We publish the public GHCR image in July.","tags":["timeline","publish-the-public-ghcr-image"],"depth":1,"timestamp":"2026-07-04"},{"id":"e3","text":"We discussed how to publish the public GHCR image throughout May planning.","tags":["planning","publish-the-public-ghcr-image"],"depth":2},{"id":"e4","text":"We completed documentation for that change in July.","tags":["docs","publish-the-public-ghcr-image"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-009","query":"When did we replace the handwritten MCP server?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We replace the handwritten MCP server in June.","tags":["timeline","replace-the-handwritten-mcp-server"],"depth":1,"timestamp":"2026-06-03"},{"id":"e2","text":"We replace the handwritten MCP server in August.","tags":["timeline","replace-the-handwritten-mcp-server"],"depth":1,"timestamp":"2026-08-04"},{"id":"e3","text":"We discussed how to replace the handwritten MCP server throughout June planning.","tags":["planning","replace-the-handwritten-mcp-server"],"depth":2},{"id":"e4","text":"We completed documentation for that change in August.","tags":["docs","replace-the-handwritten-mcp-server"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-010","query":"When did we introduce depth-aware ranking?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We introduce depth-aware ranking in October.","tags":["timeline","introduce-depth-aware-ranking"],"depth":1,"timestamp":"2026-10-03"},{"id":"e2","text":"We introduce depth-aware ranking in December.","tags":["timeline","introduce-depth-aware-ranking"],"depth":1,"timestamp":"2026-12-04"},{"id":"e3","text":"We discussed how to introduce depth-aware ranking throughout October planning.","tags":["planning","introduce-depth-aware-ranking"],"depth":2},{"id":"e4","text":"We completed documentation for that change in December.","tags":["docs","introduce-depth-aware-ranking"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-011","query":"When did we remove the TUI surface?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We remove the TUI surface in January.","tags":["timeline","remove-the-tui-surface"],"depth":1,"timestamp":"2026-01-03"},{"id":"e2","text":"We remove the TUI surface in March.","tags":["timeline","remove-the-tui-surface"],"depth":1,"timestamp":"2026-03-04"},{"id":"e3","text":"We discussed how to remove the TUI surface throughout January planning.","tags":["planning","remove-the-tui-surface"],"depth":2},{"id":"e4","text":"We completed documentation for that change in March.","tags":["docs","remove-the-tui-surface"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"time-012","query":"When did we generalize the tagger pipeline?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We generalize the tagger pipeline in April.","tags":["timeline","generalize-the-tagger-pipeline"],"depth":1,"timestamp":"2026-04-03"},{"id":"e2","text":"We generalize the tagger pipeline in September.","tags":["timeline","generalize-the-tagger-pipeline"],"depth":1,"timestamp":"2026-09-04"},{"id":"e3","text":"We discussed how to generalize the tagger pipeline throughout April planning.","tags":["planning","generalize-the-tagger-pipeline"],"depth":2},{"id":"e4","text":"We completed documentation for that change in September.","tags":["docs","generalize-the-tagger-pipeline"],"depth":2}],"relevant_ids":["e1"],"notes":"Same event with swapped month distractor."}
{"id":"timex-001","query":"When did we publish the first public image?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We publish the first public image in January.","tags":["publish-the-first-public-image","timeline"],"depth":1,"timestamp":"2026-01-03"},{"id":"e2","text":"We publish the first public image in March.","tags":["publish-the-first-public-image","timeline"],"depth":1,"timestamp":"2026-03-04"},{"id":"e3","text":"We planned to publish the first public image throughout January.","tags":["publish-the-first-public-image","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-002","query":"When did we rename the graph package to taggraph?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the graph package to taggraph in February.","tags":["rename-the-graph-package-to-taggraph","timeline"],"depth":1,"timestamp":"2026-02-03"},{"id":"e2","text":"We rename the graph package to taggraph in April.","tags":["rename-the-graph-package-to-taggraph","timeline"],"depth":1,"timestamp":"2026-04-04"},{"id":"e3","text":"We planned to rename the graph package to taggraph throughout February.","tags":["rename-the-graph-package-to-taggraph","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-003","query":"When did we switch OpenCode to the official MCP SDK?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We switch OpenCode to the official MCP SDK in May.","tags":["switch-opencode-to-the-official-mcp-sdk","timeline"],"depth":1,"timestamp":"2026-05-03"},{"id":"e2","text":"We switch OpenCode to the official MCP SDK in July.","tags":["switch-opencode-to-the-official-mcp-sdk","timeline"],"depth":1,"timestamp":"2026-07-04"},{"id":"e3","text":"We planned to switch OpenCode to the official MCP SDK throughout May.","tags":["switch-opencode-to-the-official-mcp-sdk","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-004","query":"When did we remove the terminal UI?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We remove the terminal UI in June.","tags":["remove-the-terminal-ui","timeline"],"depth":1,"timestamp":"2026-06-03"},{"id":"e2","text":"We remove the terminal UI in August.","tags":["remove-the-terminal-ui","timeline"],"depth":1,"timestamp":"2026-08-04"},{"id":"e3","text":"We planned to remove the terminal UI throughout June.","tags":["remove-the-terminal-ui","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-005","query":"When did we set bge-small as the default embedded model?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We set bge-small as the default embedded model in September.","tags":["set-bge-small-as-the-default-embedded-model","timeline"],"depth":1,"timestamp":"2026-09-03"},{"id":"e2","text":"We set bge-small as the default embedded model in November.","tags":["set-bge-small-as-the-default-embedded-model","timeline"],"depth":1,"timestamp":"2026-11-04"},{"id":"e3","text":"We planned to set bge-small as the default embedded model throughout September.","tags":["set-bge-small-as-the-default-embedded-model","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-006","query":"When did we introduce proper-noun tag extraction?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We introduce proper-noun tag extraction in October.","tags":["introduce-proper-noun-tag-extraction","timeline"],"depth":1,"timestamp":"2026-10-03"},{"id":"e2","text":"We introduce proper-noun tag extraction in December.","tags":["introduce-proper-noun-tag-extraction","timeline"],"depth":1,"timestamp":"2026-12-04"},{"id":"e3","text":"We planned to introduce proper-noun tag extraction throughout October.","tags":["introduce-proper-noun-tag-extraction","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-007","query":"When did we generalize the Docker runtime image?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We generalize the Docker runtime image in March.","tags":["generalize-the-docker-runtime-image","timeline"],"depth":1,"timestamp":"2026-03-03"},{"id":"e2","text":"We generalize the Docker runtime image in May.","tags":["generalize-the-docker-runtime-image","timeline"],"depth":1,"timestamp":"2026-05-04"},{"id":"e3","text":"We planned to generalize the Docker runtime image throughout March.","tags":["generalize-the-docker-runtime-image","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-008","query":"When did we fix anonymous GHCR pulls?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We fix anonymous GHCR pulls in April.","tags":["fix-anonymous-ghcr-pulls","timeline"],"depth":1,"timestamp":"2026-04-03"},{"id":"e2","text":"We fix anonymous GHCR pulls in June.","tags":["fix-anonymous-ghcr-pulls","timeline"],"depth":1,"timestamp":"2026-06-04"},{"id":"e3","text":"We planned to fix anonymous GHCR pulls throughout April.","tags":["fix-anonymous-ghcr-pulls","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-009","query":"When did we drop tier terminology from the interface?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We drop tier terminology from the interface in July.","tags":["drop-tier-terminology-from-the-interface","timeline"],"depth":1,"timestamp":"2026-07-03"},{"id":"e2","text":"We drop tier terminology from the interface in September.","tags":["drop-tier-terminology-from-the-interface","timeline"],"depth":1,"timestamp":"2026-09-04"},{"id":"e3","text":"We planned to drop tier terminology from the interface throughout July.","tags":["drop-tier-terminology-from-the-interface","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-010","query":"When did we move from topicgraph to taggraph?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We move from topicgraph to taggraph in August.","tags":["move-from-topicgraph-to-taggraph","timeline"],"depth":1,"timestamp":"2026-08-03"},{"id":"e2","text":"We move from topicgraph to taggraph in October.","tags":["move-from-topicgraph-to-taggraph","timeline"],"depth":1,"timestamp":"2026-10-04"},{"id":"e3","text":"We planned to move from topicgraph to taggraph throughout August.","tags":["move-from-topicgraph-to-taggraph","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-011","query":"When did we add benchmark charts?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We add benchmark charts in November.","tags":["add-benchmark-charts","timeline"],"depth":1,"timestamp":"2026-11-03"},{"id":"e2","text":"We add benchmark charts in January.","tags":["add-benchmark-charts","timeline"],"depth":1,"timestamp":"2026-01-04"},{"id":"e3","text":"We planned to add benchmark charts throughout November.","tags":["add-benchmark-charts","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-012","query":"When did we publish the adversarial benchmark skeleton?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We publish the adversarial benchmark skeleton in December.","tags":["publish-the-adversarial-benchmark-skeleton","timeline"],"depth":1,"timestamp":"2026-12-03"},{"id":"e2","text":"We publish the adversarial benchmark skeleton in February.","tags":["publish-the-adversarial-benchmark-skeleton","timeline"],"depth":1,"timestamp":"2026-02-04"},{"id":"e3","text":"We planned to publish the adversarial benchmark skeleton throughout December.","tags":["publish-the-adversarial-benchmark-skeleton","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-013","query":"When did we introduce depth-aware ranking bias?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We introduce depth-aware ranking bias in January.","tags":["introduce-depth-aware-ranking-bias","timeline"],"depth":1,"timestamp":"2026-01-03"},{"id":"e2","text":"We introduce depth-aware ranking bias in March.","tags":["introduce-depth-aware-ranking-bias","timeline"],"depth":1,"timestamp":"2026-03-04"},{"id":"e3","text":"We planned to introduce depth-aware ranking bias throughout January.","tags":["introduce-depth-aware-ranking-bias","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-014","query":"When did we refocus the README benchmark section?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We refocus the README benchmark section in February.","tags":["refocus-the-readme-benchmark-section","timeline"],"depth":1,"timestamp":"2026-02-03"},{"id":"e2","text":"We refocus the README benchmark section in April.","tags":["refocus-the-readme-benchmark-section","timeline"],"depth":1,"timestamp":"2026-04-04"},{"id":"e3","text":"We planned to refocus the README benchmark section throughout February.","tags":["refocus-the-readme-benchmark-section","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-015","query":"When did we switch the package image name to generic tagmem?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We switch the package image name to generic tagmem in May.","tags":["switch-the-package-image-name-to-generic-tagmem","timeline"],"depth":1,"timestamp":"2026-05-03"},{"id":"e2","text":"We switch the package image name to generic tagmem in July.","tags":["switch-the-package-image-name-to-generic-tagmem","timeline"],"depth":1,"timestamp":"2026-07-04"},{"id":"e3","text":"We planned to switch the package image name to generic tagmem throughout May.","tags":["switch-the-package-image-name-to-generic-tagmem","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-016","query":"When did we replace the shell-wrapper MCP path?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We replace the shell-wrapper MCP path in June.","tags":["replace-the-shell-wrapper-mcp-path","timeline"],"depth":1,"timestamp":"2026-06-03"},{"id":"e2","text":"We replace the shell-wrapper MCP path in August.","tags":["replace-the-shell-wrapper-mcp-path","timeline"],"depth":1,"timestamp":"2026-08-04"},{"id":"e3","text":"We planned to replace the shell-wrapper MCP path throughout June.","tags":["replace-the-shell-wrapper-mcp-path","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-017","query":"When did we add scripted ingest integration tests?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We add scripted ingest integration tests in September.","tags":["add-scripted-ingest-integration-tests","timeline"],"depth":1,"timestamp":"2026-09-03"},{"id":"e2","text":"We add scripted ingest integration tests in November.","tags":["add-scripted-ingest-integration-tests","timeline"],"depth":1,"timestamp":"2026-11-04"},{"id":"e3","text":"We planned to add scripted ingest integration tests throughout September.","tags":["add-scripted-ingest-integration-tests","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-018","query":"When did we add benchmark raw artifact publishing?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We add benchmark raw artifact publishing in October.","tags":["add-benchmark-raw-artifact-publishing","timeline"],"depth":1,"timestamp":"2026-10-03"},{"id":"e2","text":"We add benchmark raw artifact publishing in December.","tags":["add-benchmark-raw-artifact-publishing","timeline"],"depth":1,"timestamp":"2026-12-04"},{"id":"e3","text":"We planned to add benchmark raw artifact publishing throughout October.","tags":["add-benchmark-raw-artifact-publishing","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-019","query":"When did we make the public repo visible?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We make the public repo visible in March.","tags":["make-the-public-repo-visible","timeline"],"depth":1,"timestamp":"2026-03-03"},{"id":"e2","text":"We make the public repo visible in May.","tags":["make-the-public-repo-visible","timeline"],"depth":1,"timestamp":"2026-05-04"},{"id":"e3","text":"We planned to make the public repo visible throughout March.","tags":["make-the-public-repo-visible","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"timex-020","query":"When did we publish the refreshed runtime image?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We publish the refreshed runtime image in April.","tags":["publish-the-refreshed-runtime-image","timeline"],"depth":1,"timestamp":"2026-04-03"},{"id":"e2","text":"We publish the refreshed runtime image in June.","tags":["publish-the-refreshed-runtime-image","timeline"],"depth":1,"timestamp":"2026-06-04"},{"id":"e3","text":"We planned to publish the refreshed runtime image throughout April.","tags":["publish-the-refreshed-runtime-image","planning"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra time-swap case."}
{"id":"state-001","query":"What is the current production domain?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current production domain is ecue.ai.","tags":["current-production-domain","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current production domain used to be hifidelityai.com.","tags":["current-production-domain","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is preview.ecue.ai.","tags":["current-production-domain","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current production domain.","tags":["current-production-domain","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-002","query":"What is the default embed model?","adversary_type":"state_update","entries":[{"id":"e1","text":"The default embed model is bge-small-en-v1.5.","tags":["default-embed-model","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The default embed model used to be all-MiniLM-L6-v2.","tags":["default-embed-model","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is bge-base-en-v1.5.","tags":["default-embed-model","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the default embed model.","tags":["default-embed-model","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-003","query":"What is the current staging database?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current staging database is db-staging.internal.","tags":["current-staging-database","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current staging database used to be db-old-staging.internal.","tags":["current-staging-database","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is db-preview.internal.","tags":["current-staging-database","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current staging database.","tags":["current-staging-database","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-004","query":"What is the current metrics backend?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current metrics backend is VictoriaMetrics.","tags":["current-metrics-backend","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current metrics backend used to be Prometheus.","tags":["current-metrics-backend","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is InfluxDB.","tags":["current-metrics-backend","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current metrics backend.","tags":["current-metrics-backend","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-005","query":"What is the current mail sender?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current mail sender is ses-prod.","tags":["current-mail-sender","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current mail sender used to be ses-old.","tags":["current-mail-sender","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is ses-preview.","tags":["current-mail-sender","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current mail sender.","tags":["current-mail-sender","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-006","query":"What is the current OpenCode MCP name?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current OpenCode MCP name is tagmem.","tags":["current-opencode-mcp-name","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current OpenCode MCP name used to be tagmem_active.","tags":["current-opencode-mcp-name","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is mempalace_active.","tags":["current-opencode-mcp-name","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current OpenCode MCP name.","tags":["current-opencode-mcp-name","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-007","query":"What is the current published image?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current published image is ghcr.io/codysnider/tagmem.","tags":["current-published-image","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current published image used to be ghcr.io/codysnider/tagmem-opencode.","tags":["current-published-image","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is ghcr.io/codysnider/tagmem-preview.","tags":["current-published-image","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current published image.","tags":["current-published-image","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-008","query":"What is the current local data root?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current local data root is $HOME/.local/share/tagmem.","tags":["current-local-data-root","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current local data root used to be /data/tagmem.","tags":["current-local-data-root","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is /srv/tagmem.","tags":["current-local-data-root","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current local data root.","tags":["current-local-data-root","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-009","query":"What is the current vector backend?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current vector backend is chromem-go.","tags":["current-vector-backend","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current vector backend used to be ChromaDB.","tags":["current-vector-backend","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is SQLite FTS.","tags":["current-vector-backend","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current vector backend.","tags":["current-vector-backend","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"state-010","query":"What is the current graph package name?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current graph package name is taggraph.","tags":["current-graph-package-name","current"],"depth":0,"timestamp":"2026-04-01"},{"id":"e2","text":"The current graph package name used to be topicgraph.","tags":["current-graph-package-name","historical"],"depth":2,"timestamp":"2025-10-01"},{"id":"e3","text":"The staging or preview equivalent is memorygraph.","tags":["current-graph-package-name","preview"],"depth":2,"timestamp":"2026-04-01"},{"id":"e4","text":"We updated the runbook after changing the current graph package name.","tags":["current-graph-package-name","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Current fact vs stale fact and sibling environment distractor."}
{"id":"statex-001","query":"What is the current benchmark default model?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current benchmark default model is bge-small-en-v1.5.","tags":["current-benchmark-default-model","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current benchmark default model used to be all-MiniLM-L6-v2.","tags":["current-benchmark-default-model","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is bge-base-en-v1.5.","tags":["current-benchmark-default-model","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-002","query":"What is the current OpenCode MCP name?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current OpenCode MCP name is tagmem.","tags":["current-opencode-mcp-name","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current OpenCode MCP name used to be tagmem_active.","tags":["current-opencode-mcp-name","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is mempalace_active.","tags":["current-opencode-mcp-name","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-003","query":"What is the current Dockerfile name?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current Dockerfile name is Dockerfile.runtime.","tags":["current-dockerfile-name","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current Dockerfile name used to be Dockerfile.opencode.","tags":["current-dockerfile-name","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is Dockerfile.dev.","tags":["current-dockerfile-name","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-004","query":"What is the current GHCR image?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current GHCR image is ghcr.io/codysnider/tagmem.","tags":["current-ghcr-image","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current GHCR image used to be ghcr.io/codysnider/tagmem-opencode.","tags":["current-ghcr-image","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is ghcr.io/codysnider/tagmem-preview.","tags":["current-ghcr-image","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-005","query":"What is the current benchmark data root?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current benchmark data root is $HOME/.local/share/tagmem.","tags":["current-benchmark-data-root","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current benchmark data root used to be /data/tagmem.","tags":["current-benchmark-data-root","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is /srv/tagmem.","tags":["current-benchmark-data-root","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-006","query":"What is the current public MCP prefix?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current public MCP prefix is tagmem_.","tags":["current-public-mcp-prefix","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current public MCP prefix used to be tiered_memory_.","tags":["current-public-mcp-prefix","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is memory_.","tags":["current-public-mcp-prefix","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-007","query":"What is the current graph package name?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current graph package name is taggraph.","tags":["current-graph-package-name","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current graph package name used to be topicgraph.","tags":["current-graph-package-name","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is memorygraph.","tags":["current-graph-package-name","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-008","query":"What is the current default acceleration mode?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default acceleration mode is auto.","tags":["current-default-acceleration-mode","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current default acceleration mode used to be cuda.","tags":["current-default-acceleration-mode","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is cpu.","tags":["current-default-acceleration-mode","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-009","query":"What is the current install preference?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current install preference is Docker.","tags":["current-install-preference","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current install preference used to be local go build.","tags":["current-install-preference","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is manual source edits.","tags":["current-install-preference","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-010","query":"What is the current CLI primary surface?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current CLI primary surface is CLI and MCP.","tags":["current-cli-primary-surface","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current CLI primary surface used to be TUI and MCP.","tags":["current-cli-primary-surface","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is TUI only.","tags":["current-cli-primary-surface","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-011","query":"What is the current default runtime image?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default runtime image is tagmem:latest.","tags":["current-default-runtime-image","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current default runtime image used to be tagmem-opencode:latest.","tags":["current-default-runtime-image","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is tagmem-dev:latest.","tags":["current-default-runtime-image","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-012","query":"What is the current published benchmark file path?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current published benchmark file path is benchmarks/raw/bge-small-en-v1.5.","tags":["current-published-benchmark-file-path","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current published benchmark file path used to be bench-results-live.","tags":["current-published-benchmark-file-path","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is results/live.","tags":["current-published-benchmark-file-path","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-013","query":"What is the current package host?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current package host is ghcr.io.","tags":["current-package-host","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current package host used to be docker.io.","tags":["current-package-host","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is quay.io.","tags":["current-package-host","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-014","query":"What is the current repo owner?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current repo owner is codysnider.","tags":["current-repo-owner","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current repo owner used to be lhl.","tags":["current-repo-owner","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is openai.","tags":["current-repo-owner","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-015","query":"What is the current benchmark category count?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current benchmark category count is 5.","tags":["current-benchmark-category-count","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current benchmark category count used to be 4.","tags":["current-benchmark-category-count","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is 6.","tags":["current-benchmark-category-count","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-016","query":"What is the current benchmark default device?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current benchmark default device is GPU.","tags":["current-benchmark-default-device","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current benchmark default device used to be CPU.","tags":["current-benchmark-default-device","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is TPU.","tags":["current-benchmark-default-device","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-017","query":"What is the current memory grouping model?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current memory grouping model is tags plus depth.","tags":["current-memory-grouping-model","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current memory grouping model used to be rooms plus wings.","tags":["current-memory-grouping-model","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is folders only.","tags":["current-memory-grouping-model","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-018","query":"What is the current benchmark standalone project?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current benchmark standalone project is adversarial-memory-bench.","tags":["current-benchmark-standalone-project","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current benchmark standalone project used to be benchmarks.","tags":["current-benchmark-standalone-project","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is tagmem-bench.","tags":["current-benchmark-standalone-project","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-019","query":"What is the current install wrapper root?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current install wrapper root is ~/.local/share/tagmem/install.","tags":["current-install-wrapper-root","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current install wrapper root used to be ~/.local/share/tagmem/opencode-install.","tags":["current-install-wrapper-root","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is ~/.cache/tagmem/install.","tags":["current-install-wrapper-root","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"statex-020","query":"What is the current doctor image behavior?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current doctor image behavior is generic runtime image.","tags":["current-doctor-image-behavior","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current doctor image behavior used to be OpenCode-only image.","tags":["current-doctor-image-behavior","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"A related alternative is source-only binary.","tags":["current-doctor-image-behavior","alternative"],"depth":2,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Extra state-update case."}
{"id":"speaker-001","query":"What did you suggest for Terraform environments?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested separate Terraform states for shared, production, and staging instead of a single workspace-heavy state.","tags":["terraform-environments","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested one Terraform state with workspaces because it looked simpler on paper.","tags":["terraform-environments","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested isolating shared infrastructure so production drift could not be hidden behind staging changes.","tags":["terraform-environments","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later reviewed how imports and state moves become harder when all environments share one state file.","tags":["terraform-environments","discussion"],"depth":2},{"id":"e5","text":"The follow-up notes compared separate states against workspaces for blast-radius control, not just convenience.","tags":["terraform-environments","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded early speaker case with two relevant assistant turns and a more realistic user simplification distractor."}
{"id":"speaker-002","query":"What did you suggest for tagging pipeline?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested use deterministic extraction first and embedding ranking second.","tags":["tagging-pipeline","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested let the model invent tags directly.","tags":["tagging-pipeline","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed tagging pipeline implementation details later.","tags":["tagging-pipeline","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about tagging pipeline.","tags":["tagging-pipeline","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-003","query":"What did you suggest for OpenCode integration?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested run the image directly with the mcp subcommand.","tags":["opencode-integration","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested wrap everything in a shell pipeline.","tags":["opencode-integration","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed OpenCode integration implementation details later.","tags":["opencode-integration","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about OpenCode integration.","tags":["opencode-integration","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-004","query":"What did you suggest for release packaging?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested publish ghcr.io/codysnider/tagmem as the generic image.","tags":["release-packaging","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested ship only a local Docker build.","tags":["release-packaging","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed release packaging implementation details later.","tags":["release-packaging","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about release packaging.","tags":["release-packaging","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-005","query":"What did you suggest for depth model?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested treat depth as a secondary ranking bias.","tags":["depth-model","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested replace tags with rigid hierarchical folders.","tags":["depth-model","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed depth model implementation details later.","tags":["depth-model","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about depth model.","tags":["depth-model","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-006","query":"What did you suggest for README benchmarks?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keep a compact benchmark table in the main README.","tags":["readme-benchmarks","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested put the full benchmark suite in the top section.","tags":["readme-benchmarks","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed README benchmarks implementation details later.","tags":["readme-benchmarks","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about README benchmarks.","tags":["readme-benchmarks","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-007","query":"What did you suggest for Docker data root?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested use HOME-local defaults with an env override.","tags":["docker-data-root","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested hardcode everything to /data/tagmem.","tags":["docker-data-root","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed Docker data root implementation details later.","tags":["docker-data-root","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about Docker data root.","tags":["docker-data-root","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-008","query":"What did you suggest for MCP compatibility?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested use the official Go SDK.","tags":["mcp-compatibility","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested hand-roll every transport detail.","tags":["mcp-compatibility","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed MCP compatibility implementation details later.","tags":["mcp-compatibility","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about MCP compatibility.","tags":["mcp-compatibility","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-009","query":"What did you suggest for adversarial benchmark repo?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested split it into a standalone project.","tags":["adversarial-benchmark-repo","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested bury it inside the main app repo.","tags":["adversarial-benchmark-repo","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed adversarial benchmark repo implementation details later.","tags":["adversarial-benchmark-repo","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about adversarial benchmark repo.","tags":["adversarial-benchmark-repo","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speaker-010","query":"What did you suggest for OpenCode command naming?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested use remember instead of mine.","tags":["opencode-command-naming","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested keep mine because users will figure it out.","tags":["opencode-command-naming","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed OpenCode command naming implementation details later.","tags":["opencode-command-naming","discussion"],"depth":2},{"id":"e4","text":"The team wrote follow-up notes about OpenCode command naming.","tags":["opencode-command-naming","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Assistant/user suggestion swap with same subject area."}
{"id":"speakerx-001","query":"What did you suggest for adversarial benchmark design?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using exact relevant ids with plausible distractors so retrieval could be scored directly instead of judging generated summaries.","tags":["adversarial-benchmark-design","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just scoring generated answers because retrieval details felt secondary.","tags":["adversarial-benchmark-design","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping each case small enough to audit by hand even when it contains several realistic distractors.","tags":["adversarial-benchmark-design","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how exact ids make it possible to analyze which wrong memory won instead of only whether final wording looked acceptable.","tags":["adversarial-benchmark-design","discussion"],"depth":2},{"id":"e5","text":"The design notes continue to emphasize inspectable cases and retrieval-focused scoring.","tags":["adversarial-benchmark-design","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded early speaker case with dual relevant assistant guidance and a stronger retrieval-vs-generation contrast."}
{"id":"speakerx-002","query":"What did you suggest for Docker image naming?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested making the image generic rather than OpenCode-specific so the same artifact could run mcp, doctor, and benchmark commands.","tags":["docker-image-naming","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested keeping the image named tagmem-opencode forever because the current runtime was the only one I cared about.","tags":["docker-image-naming","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested avoiding runtime-specific names so future agent integrations would not need a separate image story.","tags":["docker-image-naming","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how a generic image reduces documentation drift when multiple runtimes share one container.","tags":["docker-image-naming","discussion"],"depth":2},{"id":"e5","text":"The publish notes now treat the image as a generic runtime artifact, not an OpenCode-only deliverable.","tags":["docker-image-naming","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with two relevant assistant recommendations and a stronger why-driven contrast."}
{"id":"speakerx-003","query":"What did you suggest for OpenCode commands?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested renaming mine to remember because the command should describe importing memory rather than sounding clever.","tags":["opencode-commands","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested keeping mine because some existing users already recognized it and I thought novelty mattered.","tags":["opencode-commands","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested preferring boring, obvious command names over metaphor-heavy ones so the CLI stays self-explanatory.","tags":["opencode-commands","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how command names should reduce explanation burden rather than add brand personality.","tags":["opencode-commands","discussion"],"depth":2},{"id":"e5","text":"The naming notes now consistently favor literal, well-understood verbs for memory actions.","tags":["opencode-commands","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant guidance and a clearer usability rationale."}
{"id":"speakerx-004","query":"What did you suggest for tagging pipeline?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using deterministic extraction before model ranking so obvious entities and symbols are captured consistently.","tags":["tagging-pipeline","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested letting the model invent every tag from scratch because it sounded more flexible.","tags":["tagging-pipeline","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested treating embedding ranking as a refinement layer rather than the first source of tags.","tags":["tagging-pipeline","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how deterministic extraction reduces drift for proper nouns, commands, and code symbols.","tags":["tagging-pipeline","discussion"],"depth":2},{"id":"e5","text":"The tagging notes now frame model ranking as a second-pass scorer, not the tag generator itself.","tags":["tagging-pipeline","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant turns and a stronger sequencing distinction."}
{"id":"speakerx-005","query":"What did you suggest for README benchmarks?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keeping the main README benchmark comparison compact so readers can orient quickly before diving into methodology elsewhere.","tags":["readme-benchmarks","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested putting every benchmark detail at the top of the README because I wanted nothing left implicit.","tags":["readme-benchmarks","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested linking out to fuller benchmark docs instead of forcing raw methodology into the first screen of the project.","tags":["readme-benchmarks","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how compact summaries and detailed methodology can complement each other rather than compete for the same space.","tags":["readme-benchmarks","discussion"],"depth":2},{"id":"e5","text":"The docs now split overview and methodology instead of collapsing them into one long README section.","tags":["readme-benchmarks","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with two relevant assistant recommendations and stronger overview-vs-methodology tension."}
{"id":"speakerx-006","query":"What did you suggest for public package visibility?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested verifying anonymous GHCR pulls from another machine because maintainer credentials can hide broken package visibility.","tags":["public-package-visibility","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested assuming package visibility from the maintainer machine because successful local pulls felt good enough.","tags":["public-package-visibility","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested checking the package from a fresh-user context so public install claims are proven rather than inferred.","tags":["public-package-visibility","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how visibility problems often stay invisible until a truly anonymous machine tries to pull.","tags":["public-package-visibility","discussion"],"depth":2},{"id":"e5","text":"The install notes now treat anonymous pulls as part of the public packaging story, not as an optional extra check.","tags":["public-package-visibility","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with two relevant assistant recommendations and stronger fresh-user framing."}
{"id":"speakerx-007","query":"What did you suggest for OpenCode MCP implementation?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using the official Go SDK so transport and protocol details would follow the current MCP conventions instead of drifting.","tags":["opencode-mcp-implementation","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested hand-rolling the protocol indefinitely because I wanted finer control over every detail.","tags":["opencode-mcp-implementation","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested avoiding custom protocol glue where the SDK already handles capability negotiation and message framing correctly.","tags":["opencode-mcp-implementation","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how hand-rolled MCP layers become maintenance liabilities when the protocol surface evolves.","tags":["opencode-mcp-implementation","discussion"],"depth":2},{"id":"e5","text":"The implementation notes now frame the SDK choice as reducing protocol drift, not just saving time.","tags":["opencode-mcp-implementation","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant guidance and a stronger protocol-maintenance rationale."}
{"id":"speakerx-008","query":"What did you suggest for TUI removal?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested removing the TUI completely for this phase instead of carrying a broken interface forward.","tags":["tui-removal","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested leaving the broken TUI in place because it might improve later and I did not want to close the door on it.","tags":["tui-removal","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested pruning dead surface area so docs, help output, and user expectations stop pointing at a path that is not actually supported.","tags":["tui-removal","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how half-working interfaces create more confusion than cleanly removed ones.","tags":["tui-removal","discussion"],"depth":2},{"id":"e5","text":"The project surface now emphasizes supported commands instead of preserving dormant UI affordances.","tags":["tui-removal","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant guidance and a stronger dead-surface-area rationale."}
{"id":"speakerx-009","query":"What did you suggest for depth model?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested treating depth as a ranking bias rather than the main organizer of memory.","tags":["depth-model","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested replacing tags with depth-only buckets because hierarchy felt cleaner to me.","tags":["depth-model","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping tags primary so related facts can connect across different depths instead of being trapped in one layer.","tags":["depth-model","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how depth works best as a retrieval hint once tags already define the semantic grouping.","tags":["depth-model","discussion"],"depth":2},{"id":"e5","text":"The later benchmark cases now assume depth is secondary and tags carry most of the conceptual load.","tags":["depth-model","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant turns and a clearer tags-vs-depth distinction."}
{"id":"speakerx-010","query":"What did you suggest for GPU support?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested allow GPU if available and fall back to CPU.","tags":["gpu-support","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested ban GPU because purity matters more than user experience.","tags":["gpu-support","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"We discussed GPU support again in follow-up planning.","tags":["gpu-support","discussion"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra speaker confusion case."}
{"id":"speakerx-011","query":"What did you suggest for docs cleanup?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested removing personal machine paths from public docs so published instructions stop leaking maintainer-specific assumptions.","tags":["docs-cleanup","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested leaving local paths because maintainers could mentally translate them and I thought that was good enough.","tags":["docs-cleanup","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested replacing machine-local examples with portable placeholders or home-relative defaults so docs stay trustworthy for fresh users.","tags":["docs-cleanup","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how local-path leakage quietly makes install docs look broken for anyone outside the maintainer environment.","tags":["docs-cleanup","discussion"],"depth":2},{"id":"e5","text":"The documentation cleanup notes now treat path portability as a public-facing correctness issue, not a cosmetic detail.","tags":["docs-cleanup","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with two relevant assistant turns and stronger portability rationale."}
{"id":"speakerx-012","query":"What did you suggest for fresh install testing?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested testing from another machine over SSH so package pulls, defaults, and path assumptions are exercised outside the maintainer environment.","tags":["fresh-install-testing","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested assuming local success implied public install success because the package worked on my box already.","tags":["fresh-install-testing","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested validating from a genuinely fresh-user context instead of a machine that already has hidden credentials, caches, and local checkouts.","tags":["fresh-install-testing","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how install flows often fail on environment assumptions that the maintainer machine can no longer reveal.","tags":["fresh-install-testing","discussion"],"depth":2},{"id":"e5","text":"The install validation notes now treat remote fresh-user testing as part of release readiness rather than as an optional sanity check.","tags":["fresh-install-testing","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant guidance and stronger fresh-user realism."}
{"id":"speakerx-013","query":"What did you suggest for image publish process?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested publishing the generic ghcr.io/codysnider/tagmem image so installs, OpenCode config, and other integrations all point at one public artifact.","tags":["image-publish-process","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested shipping only a local image and telling users to rebuild it themselves because that felt operationally lighter for maintainers.","tags":["image-publish-process","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested folding image publishing into the release checklist so public install viability is treated as part of shipping, not as an optional packaging chore.","tags":["image-publish-process","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how a local-only image story quietly breaks documentation, CI examples, and downstream tool integrations even when maintainers can still run everything themselves.","tags":["image-publish-process","discussion"],"depth":2},{"id":"e5","text":"The packaging notes now frame local builds as fallback or contributor paths, while the generic published image is treated as the canonical public artifact.","tags":["image-publish-process","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Further upgraded speaker case with more release-process realism and stronger downstream-integration consequences."}
{"id":"speakerx-014","query":"What did you suggest for memory terminology?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using plain terms like entries, tags, depth, facts, and diary so the system vocabulary stays legible to new users and maintainers.","tags":["memory-terminology","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested inventing new metaphor-heavy nouns because I thought they would feel more distinctive and give the system a stronger personality.","tags":["memory-terminology","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested that code, docs, CLI help, and benchmark labels should share the same plain terminology so people are not constantly translating between surfaces.","tags":["memory-terminology","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how metaphor-heavy naming often creates fake depth while making onboarding, debugging, and support conversations more brittle.","tags":["memory-terminology","discussion"],"depth":2},{"id":"e5","text":"The naming notes now favor words that already fit naturally in user questions, code comments, and evaluation docs instead of invented conceptual lore.","tags":["memory-terminology","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Further upgraded speaker case with stronger cross-surface realism and more believable rationale against metaphor-heavy naming."}
{"id":"speakerx-015","query":"What did you suggest for public benchmark reporting?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested including methodology, machine specs, and raw JSON so public benchmark claims can be inspected and reproduced.","tags":["public-benchmark-reporting","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested quoting one headline number and hiding everything else because I wanted the comparison to look cleaner.","tags":["public-benchmark-reporting","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested publishing enough detail that others can see whether differences came from hardware, configuration, or actual retrieval quality.","tags":["public-benchmark-reporting","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how benchmark trust collapses when people cannot inspect the exact commands, settings, and raw outputs behind a claim.","tags":["public-benchmark-reporting","discussion"],"depth":2},{"id":"e5","text":"The reporting notes now treat reproducibility details as part of the benchmark result, not as optional appendix material.","tags":["public-benchmark-reporting","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant turns and stronger reproducibility rationale."}
{"id":"speakerx-016","query":"What did you suggest for OpenAI-compatible support?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keeping the support generic rather than naming one specific server so the feature describes a protocol family, not one vendor.","tags":["openai-compatible-support","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested hard-coding it as an Ollama feature because that was the first server I had in mind.","tags":["openai-compatible-support","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested avoiding product-specific wording so future compatible servers can fit without renaming the feature later.","tags":["openai-compatible-support","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how vendor-specific naming quietly narrows expectations even when the actual implementation is more general.","tags":["openai-compatible-support","discussion"],"depth":2},{"id":"e5","text":"The docs now frame compatibility in terms of the API shape rather than one specific runtime brand.","tags":["openai-compatible-support","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with more realistic protocol-family framing and stronger future-compatibility rationale."}
{"id":"speakerx-017","query":"What did you suggest for CLI install docs?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested preferring Docker-first instructions because they give more users the same working environment without asking them to assemble toolchains first.","tags":["cli-install-docs","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested assuming everyone wanted go install first because that was the path most natural to me as a maintainer.","tags":["cli-install-docs","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested treating source installs as a secondary path for contributors rather than the default story for new users.","tags":["cli-install-docs","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how docs should optimize for the broadest successful first run, not for the maintainer's preferred development loop.","tags":["cli-install-docs","discussion"],"depth":2},{"id":"e5","text":"The install docs now separate quick-start and contributor paths instead of blending them into one ambiguous setup story.","tags":["cli-install-docs","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with dual relevant assistant turns and a more realistic onboarding-vs-contributor split."}
{"id":"speakerx-018","query":"What did you suggest for graph naming?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested renaming topicgraph to taggraph so the package name matches the actual organizing concept used throughout the system.","tags":["graph-naming","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested keeping topicgraph because it was internal and I thought renaming it would not matter much.","tags":["graph-naming","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested aligning the graph name with the user-facing docs so internal terminology does not quietly contradict the public mental model.","tags":["graph-naming","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how internal-only names still leak into logs, docs, code review, and maintenance conversations.","tags":["graph-naming","discussion"],"depth":2},{"id":"e5","text":"The later docs and package names now consistently treat tags, not topics, as the primary graph vocabulary.","tags":["graph-naming","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with realistic naming-consistency rationale across code and docs."}
{"id":"speakerx-019","query":"What did you suggest for CPU fallback verification?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested proving CPU fallback with a real add/search flow on a remote machine rather than trusting a diagnostic command alone.","tags":["cpu-fallback-verification","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just trusting the doctor output because it already reported the runtime and device state.","tags":["cpu-fallback-verification","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested exercising the full retrieval path so fallback is validated at the user-visible behavior layer, not only at the diagnostic layer.","tags":["cpu-fallback-verification","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how a runtime can report CPU execution yet still fail on real add/search if packaging or model assets are incomplete.","tags":["cpu-fallback-verification","discussion"],"depth":2},{"id":"e5","text":"The verification notes now treat doctor output as evidence, but not sufficient proof, of an actually working CPU fallback story.","tags":["cpu-fallback-verification","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with stronger diagnostics-vs-real-behavior contrast and two relevant assistant turns."}
{"id":"speakerx-020","query":"What did you suggest for benchmark expansion?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested adding adversarial distractor cases as a standalone project so the benchmark could grow, publish, and evolve without being buried inside product code concerns.","tags":["benchmark-expansion","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested burying the benchmark data inside the main repo forever because that felt simpler than treating it as its own asset.","tags":["benchmark-expansion","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested a standalone benchmark could develop its own documentation, dataset versioning, and evaluation workflow without inheriting unrelated product churn.","tags":["benchmark-expansion","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how benchmarks become harder to curate when they are treated as side files inside a repo optimized for other goals.","tags":["benchmark-expansion","discussion"],"depth":2},{"id":"e5","text":"The benchmark notes now treat the dataset as a first-class artifact rather than a hidden byproduct of the main repository.","tags":["benchmark-expansion","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded speaker case with stronger dataset-as-artifact reasoning and dual assistant relevance."}
{"id":"paraphrase-001","query":"What timeout do API calls use?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"API calls time out after 30 seconds.","tags":["config","limits","api-timeout"],"depth":1},{"id":"e2","text":"The API timeout is 60 seconds.","tags":["config","limits","api-timeout"],"depth":1},{"id":"e3","text":"External API requests are capped at thirty seconds before timing out.","tags":["config","limits","api-timeout"],"depth":1},{"id":"e4","text":"Background jobs retry for up to 30 seconds before marking a step transiently failed.","tags":["jobs","limits"],"depth":2},{"id":"e5","text":"Webhook signature checks also use a 30 second network timeout, but that belongs to inbound delivery handling rather than general API calls.","tags":["webhooks","limits"],"depth":2},{"id":"e6","text":"The timeout is evaluated after DNS resolution but before request retry policy kicks in.","tags":["api","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded early paraphrase case with two correct phrasings plus same-number distractors from nearby network workflows."}
{"id":"paraphrase-002","query":"How long do production cookies last?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Production cookies expire after 12 hours.","tags":["config","limits","production-cookies"],"depth":1},{"id":"e2","text":"Production cookies expire after 24 hours.","tags":["config","limits","production-cookies"],"depth":1},{"id":"e3","text":"In production, cookies are valid for twelve hours before expiration.","tags":["config","limits","production-cookies"],"depth":1},{"id":"e4","text":"Staging cookies expire after 12 hours.","tags":["staging","limits"],"depth":2},{"id":"e5","text":"Session replay links also last 12 hours in production, but that is a different expiry policy.","tags":["sessions","limits"],"depth":2},{"id":"e6","text":"Cookie expiry is measured from issuance rather than from last activity.","tags":["cookies","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with two correct phrasings and same-duration distractors from neighboring policies."}
{"id":"paraphrase-003","query":"What is the upload limit?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The upload limit is 25 megabytes.","tags":["config","limits","upload-limit"],"depth":1},{"id":"e2","text":"The upload limit is 50 megabytes.","tags":["config","limits","upload-limit"],"depth":1},{"id":"e3","text":"Uploads are capped at twenty-five megabytes.","tags":["config","limits","upload-limit"],"depth":1},{"id":"e4","text":"The attachment limit is 25 megabytes.","tags":["attachments","limits"],"depth":2},{"id":"e5","text":"Export bundle uploads are also capped at 25 megabytes, but that is a separate endpoint policy.","tags":["exports","limits"],"depth":2},{"id":"e6","text":"The upload cap is enforced before any server-side media transformation begins.","tags":["uploads","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with two correct phrasings and same-number distractors from adjacent upload workflows."}
{"id":"paraphrase-004","query":"How often does auth refresh run?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Auth refresh runs every 10 minutes.","tags":["config","limits","auth-refresh"],"depth":1},{"id":"e2","text":"Auth refresh runs every 15 minutes.","tags":["config","limits","auth-refresh"],"depth":1},{"id":"e3","text":"Authentication refresh runs on a ten minute cadence.","tags":["config","limits","auth-refresh"],"depth":1},{"id":"e4","text":"Metrics scraping runs every 10 minutes.","tags":["metrics","limits"],"depth":2},{"id":"e5","text":"Session cleanup also checks stale records every 10 minutes, but that is a housekeeping job rather than auth refresh.","tags":["sessions","limits"],"depth":2},{"id":"e6","text":"The auth refresh cadence is evaluated independently from token issuance and cookie expiry windows.","tags":["auth","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and same-interval distractors from adjacent background jobs."}
{"id":"paraphrase-005","query":"When does the reporting batch run?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The reporting batch runs at 02:00 UTC.","tags":["config","limits","reporting-batch"],"depth":1},{"id":"e2","text":"The reporting batch runs at 03:00 UTC.","tags":["config","limits","reporting-batch"],"depth":1},{"id":"e3","text":"The reporting batch is scheduled for 02:00 UTC.","tags":["config","limits","reporting-batch"],"depth":1},{"id":"e4","text":"The cleanup batch runs at 02:00 UTC.","tags":["cleanup","limits"],"depth":2},{"id":"e5","text":"The export bundle packer also wakes up around 02:00 UTC, but it is gated behind whether new benchmark outputs exist.","tags":["exports","limits"],"depth":2},{"id":"e6","text":"The reporting batch time is described in UTC even though some dashboards render it in local time for operators.","tags":["reporting","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and same-time distractors from neighboring nightly jobs."}
{"id":"paraphrase-006","query":"What port does the public API use?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The public API port is 8443.","tags":["config","limits","public-api-port"],"depth":1},{"id":"e2","text":"The public API port is 9443.","tags":["config","limits","public-api-port"],"depth":1},{"id":"e3","text":"The public API listens on port 8443.","tags":["config","limits","public-api-port"],"depth":1},{"id":"e4","text":"The internal admin port is 8443.","tags":["admin","limits"],"depth":2},{"id":"e5","text":"The TLS terminator also fronts 8443 in some deployments, which makes the public API port easy to confuse with neighboring network components.","tags":["tls","limits"],"depth":2},{"id":"e6","text":"Port references in the runbook distinguish external API traffic from internal admin traffic even when the numbers look similar.","tags":["api","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and stronger same-port neighboring-service confusion."}
{"id":"paraphrase-007","query":"What is the Docker data root default?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The Docker data root defaults to HOME-local storage.","tags":["config","limits","docker-data-root-default"],"depth":1},{"id":"e2","text":"The Docker data root defaults to /data/tagmem.","tags":["config","limits","docker-data-root-default"],"depth":1},{"id":"e3","text":"By default, the Docker data root lives under home-local storage.","tags":["config","limits","docker-data-root-default"],"depth":1},{"id":"e4","text":"The benchmark root defaults to HOME-local storage.","tags":["benchmark","limits"],"depth":2},{"id":"e5","text":"The install docs mention the default Docker data root next to the override variable, which makes the default path and override mechanism easy to blur together.","tags":["docs","limits"],"depth":2},{"id":"e6","text":"The Docker data root default and the benchmark data root default often coincide locally even though they are distinct settings.","tags":["docker","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and stronger default-vs-related-root confusion."}
{"id":"paraphrase-008","query":"What prefix do the MCP tools use?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The MCP tools use the tagmem_ prefix.","tags":["config","limits"],"depth":1},{"id":"e2","text":"The MCP tools use the tiered_memory_ prefix.","tags":["config","limits"],"depth":1},{"id":"e3","text":"The CLI binary is named tagmem.","tags":["jobs","limits"],"depth":2},{"id":"e4","text":"We reviewed the policy related to: The MCP tools use the tagmem_ prefix.","tags":["policy"],"depth":3}],"relevant_ids":["e1"],"notes":"Near-duplicate paraphrase with one wrong value distractor."}
{"id":"paraphrase-009","query":"What is the default GPU model?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The default GPU model is bge-small-en-v1.5.","tags":["config","limits","default-gpu-model"],"depth":1},{"id":"e2","text":"The default GPU model is bge-base-en-v1.5.","tags":["config","limits","default-gpu-model"],"depth":1},{"id":"e3","text":"By default, GPU execution uses bge-small-en-v1.5.","tags":["config","limits","default-gpu-model"],"depth":1},{"id":"e4","text":"The CPU fallback model is all-MiniLM-L6-v2.","tags":["cpu","limits"],"depth":2},{"id":"e5","text":"A larger GPU model was benchmarked in some experiments, which makes the default choice easy to misremember if someone recalls quality tests instead of shipped defaults.","tags":["gpu","benchmarks"],"depth":2},{"id":"e6","text":"The default GPU model is chosen separately from the CPU fallback and separately from any optional evaluation-only model overrides.","tags":["gpu","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and stronger default-vs-experiment-vs-fallback confusion."}
{"id":"paraphrase-010","query":"What is the runtime image name?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The runtime image is ghcr.io/codysnider/tagmem.","tags":["config","limits","runtime-image"],"depth":1},{"id":"e2","text":"The runtime image is ghcr.io/codysnider/tagmem-opencode.","tags":["config","limits","runtime-image"],"depth":1},{"id":"e3","text":"The runtime image should be ghcr.io/codysnider/tagmem.","tags":["config","limits","runtime-image"],"depth":1},{"id":"e4","text":"The repo path is github.com/codysnider/tagmem.","tags":["repo","limits"],"depth":2},{"id":"e5","text":"Older conversations about an OpenCode-specific image name still float around, which makes the generic published image easy to confuse with a transitional name.","tags":["images","history"],"depth":2},{"id":"e6","text":"The image name and repository path deliberately resemble each other, but one is the published container artifact and the other is the source code location.","tags":["runtime","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and stronger image-name vs repo-path vs legacy-name confusion."}
{"id":"paraphrase-011","query":"How is the MCP server implemented?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The MCP server is implemented with the official Go SDK.","tags":["config","limits"],"depth":1},{"id":"e2","text":"The MCP server is still handwritten.","tags":["config","limits"],"depth":1},{"id":"e3","text":"The transport runs over stdio.","tags":["jobs","limits"],"depth":2},{"id":"e4","text":"We reviewed the policy related to: The MCP server is implemented with the official Go SDK.","tags":["policy"],"depth":3}],"relevant_ids":["e1"],"notes":"Near-duplicate paraphrase with one wrong value distractor."}
{"id":"paraphrase-012","query":"What is the graph package name?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The local graph package is named taggraph.","tags":["config","limits"],"depth":1},{"id":"e2","text":"The local graph package is named topicgraph.","tags":["config","limits"],"depth":1},{"id":"e3","text":"The benchmark package is named tagbench.","tags":["jobs","limits"],"depth":2},{"id":"e4","text":"We reviewed the policy related to: The local graph package is named taggraph.","tags":["policy"],"depth":3}],"relevant_ids":["e1"],"notes":"Near-duplicate paraphrase with one wrong value distractor."}
{"id":"paraphrasex-001","query":"What is the benchmark data root default?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The benchmark data root defaults to the home-local tagmem directory.","tags":["docs","retrieval","benchmark-data-root"],"depth":1},{"id":"e2","text":"The benchmark data root defaults to /data/tagmem.","tags":["docs","retrieval","benchmark-data-root"],"depth":1},{"id":"e3","text":"By default, benchmark data is stored under the user's home-local tagmem directory.","tags":["docs","retrieval","benchmark-data-root"],"depth":1},{"id":"e4","text":"The benchmark output root defaults to the home-local tagmem directory.","tags":["docs","context","benchmark-output-root"],"depth":2},{"id":"e5","text":"The Docker data root can still be overridden independently of the benchmark data root.","tags":["docs","context","docker-data-root"],"depth":2},{"id":"e6","text":"Install docs note that the data root and output root often coincide locally even though they are conceptually different settings.","tags":["docs","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded early paraphrase case with two correct phrasings and adjacent root-setting distractors instead of a single binary swap."}
{"id":"paraphrasex-002","query":"What happens when CUDA is unavailable?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"When CUDA is unavailable, the embedding model falls back to the CPU runtime.","tags":["docs","retrieval","cuda-fallback"],"depth":1},{"id":"e2","text":"When CUDA is unavailable, the embedding model aborts instead of continuing.","tags":["docs","retrieval","cuda-fallback"],"depth":1},{"id":"e3","text":"CUDA failure causes the embedding stack to fall back onto CPU execution rather than stop outright.","tags":["docs","retrieval","cuda-fallback"],"depth":1},{"id":"e4","text":"The doctor command reports the current runtime path and execution device, but that is a diagnostic surface rather than the fallback behavior itself.","tags":["docs","context","doctor"],"depth":2},{"id":"e5","text":"A local build may still log a warning before the CPU fallback engages.","tags":["docs","context","runtime-warning"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with two correct phrasings and a stronger diagnostics-vs-behavior distractor."}
{"id":"paraphrasex-003","query":"How does the MCP server communicate?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The MCP server communicates over stdio.","tags":["docs","retrieval","mcp-transport"],"depth":1},{"id":"e2","text":"The MCP server communicates only over HTTP.","tags":["docs","retrieval","mcp-transport"],"depth":1},{"id":"e3","text":"Stdio is the transport used by the MCP server rather than an HTTP listener.","tags":["docs","retrieval","mcp-transport"],"depth":1},{"id":"e4","text":"The MCP client may launch the server through a local Docker command, but process launch is not the same thing as the transport.","tags":["docs","context","mcp-launch"],"depth":2},{"id":"e5","text":"The transport choice matters because many MCP runtimes expect stdin/stdout semantics directly.","tags":["docs","notes","mcp-transport"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a better launch-vs-transport distractor."}
{"id":"paraphrasex-004","query":"What does the install wrapper do first?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The install wrapper clones the public GitHub repository first.","tags":["docs","retrieval","install-wrapper"],"depth":1},{"id":"e2","text":"The install wrapper only copies a local checkout.","tags":["docs","retrieval","install-wrapper"],"depth":1},{"id":"e3","text":"The first install-wrapper step is cloning the public GitHub repository.","tags":["docs","retrieval","install-wrapper"],"depth":1},{"id":"e4","text":"The runtime wrapper uses the installed repo path after installation, but that is not the first action of the install wrapper.","tags":["docs","context","runtime-wrapper"],"depth":2},{"id":"e5","text":"A local checkout can still be used later for developer workflows, but it is not the default first install step.","tags":["docs","context","developer-workflow"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger first-step vs later-step distractor."}
{"id":"paraphrasex-005","query":"How is the benchmark section presented in the README?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The README benchmark section now uses a compact comparison table.","tags":["docs","retrieval","readme-benchmarks"],"depth":1},{"id":"e2","text":"The README benchmark section now embeds the full benchmark methodology.","tags":["docs","retrieval","readme-benchmarks"],"depth":1},{"id":"e3","text":"The benchmark section in the README is presented as a compact comparison table.","tags":["docs","retrieval","readme-benchmarks"],"depth":1},{"id":"e4","text":"Detailed benchmark docs live in the benchmarks folder, but that is where the methodology expands beyond the compact README summary.","tags":["docs","context","benchmark-methodology"],"depth":2},{"id":"e5","text":"The overview table is meant to orient readers before they open the fuller methodology materials.","tags":["docs","notes","readme-benchmarks"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and README-summary vs detailed-methodology confusion."}
{"id":"paraphrasex-006","query":"How should OpenCode launch the image?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The OpenCode config should launch the generic image with the mcp subcommand.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"The OpenCode config should launch an OpenCode-specific image name.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"The tagmem image is generic and can run doctor or bench too.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-007","query":"What should the graph browser use?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The graph browser should use tags instead of topics.","tags":["docs","retrieval","graph-browser"],"depth":1},{"id":"e2","text":"The graph browser should keep topic terminology because users expect it.","tags":["docs","retrieval","graph-browser"],"depth":1},{"id":"e3","text":"The browser should use tags rather than topics as its primary graph vocabulary.","tags":["docs","retrieval","graph-browser"],"depth":1},{"id":"e4","text":"The graph package is now named taggraph, but the package rename is related context rather than the browser vocabulary rule itself.","tags":["docs","context","taggraph"],"depth":2},{"id":"e5","text":"Depth may still appear as a secondary filter in the browser without replacing tag vocabulary.","tags":["docs","context","depth"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with two correct phrasings and browser-vocabulary vs package-name confusion."}
{"id":"paraphrasex-008","query":"Where should the public image be published?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The public image should be published to ghcr.io/codysnider/tagmem.","tags":["docs","retrieval","public-image"],"depth":1},{"id":"e2","text":"The public image should stay local-only.","tags":["docs","retrieval","public-image"],"depth":1},{"id":"e3","text":"The public container image belongs at ghcr.io/codysnider/tagmem.","tags":["docs","retrieval","public-image"],"depth":1},{"id":"e4","text":"The install flow can fall back to a local build if pull fails, but that is a fallback path rather than the intended public publishing location.","tags":["docs","context","fallback-build"],"depth":2},{"id":"e5","text":"The generic image name is meant to support multiple commands, not to imply a local-only distribution model.","tags":["docs","notes","image-genericity"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with two correct phrasings and fallback-vs-publish-location confusion."}
{"id":"paraphrasex-009","query":"What happens when tagmem runs without a command?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The main command with no arguments prints help.","tags":["docs","retrieval","no-command-behavior"],"depth":1},{"id":"e2","text":"The main command with no arguments opens a TUI.","tags":["docs","retrieval","no-command-behavior"],"depth":1},{"id":"e3","text":"Running tagmem with no command prints the help output.","tags":["docs","retrieval","no-command-behavior"],"depth":1},{"id":"e4","text":"The TUI was removed from the project surface, which is related context but not itself the exact no-command behavior statement.","tags":["docs","context","tui-removal"],"depth":2},{"id":"e5","text":"Subcommands still provide their own help text when invoked directly with help flags.","tags":["docs","notes","cli-help"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger no-command vs removed-TUI distinction."}
{"id":"paraphrasex-010","query":"Where should the adversarial benchmark live?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The standalone adversarial benchmark should live in its own repository.","tags":["docs","retrieval","benchmark-location"],"depth":1},{"id":"e2","text":"The adversarial benchmark should stay buried under the main repo only.","tags":["docs","retrieval","benchmark-location"],"depth":1},{"id":"e3","text":"The adversarial benchmark belongs in a dedicated standalone repository.","tags":["docs","retrieval","benchmark-location"],"depth":1},{"id":"e4","text":"The standalone benchmark can later be pushed to Hugging Face, but that is a distribution step rather than the repository-location decision itself.","tags":["docs","context","benchmark-distribution"],"depth":2},{"id":"e5","text":"Keeping the benchmark standalone makes it easier to version and publish independently from the main application repo.","tags":["docs","notes","benchmark-location"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and stronger location-vs-distribution confusion."}
{"id":"paraphrasex-011","query":"How does the improved tagger work?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The improved tagger uses deterministic extraction plus embedding ranking.","tags":["docs","retrieval","improved-tagger"],"depth":1},{"id":"e2","text":"The improved tagger relies entirely on a generative model.","tags":["docs","retrieval","improved-tagger"],"depth":1},{"id":"e3","text":"The improved tagger works by combining deterministic extraction with embedding-based ranking.","tags":["docs","retrieval","improved-tagger"],"depth":1},{"id":"e4","text":"The improved tagger is aware of proper nouns and code symbols, but that is a capability detail rather than the core pipeline description itself.","tags":["docs","context","tagger-capabilities"],"depth":2},{"id":"e5","text":"Generative models may still appear in surrounding tooling, but they are not the sole mechanism behind the improved tagger.","tags":["docs","context","generative-models"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger pipeline-vs-capability distractor."}
{"id":"paraphrasex-012","query":"Why is bge-small the default GPU model?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The default GPU model is bge-small because it is the best overall tradeoff.","tags":["docs","retrieval","gpu-model-choice"],"depth":1},{"id":"e2","text":"The default GPU model is bge-base because it wins every benchmark decisively.","tags":["docs","retrieval","gpu-model-choice"],"depth":1},{"id":"e3","text":"Bge-small is the default GPU model because it offers the best overall tradeoff.","tags":["docs","retrieval","gpu-model-choice"],"depth":1},{"id":"e4","text":"MiniLM remains the throughput-first fallback, but that explains the CPU fallback path rather than why bge-small is the default GPU choice.","tags":["docs","context","cpu-fallback"],"depth":2},{"id":"e5","text":"The GPU default was chosen as a balance of quality, speed, and practicality rather than because it dominated every benchmark dimension.","tags":["docs","notes","gpu-model-choice"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger default-choice vs fallback-model distractor."}
{"id":"paraphrasex-013","query":"What MCP tool prefix does the project use?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The MCP tool names now use the tagmem_ prefix.","tags":["docs","retrieval","mcp-tool-prefix"],"depth":1},{"id":"e2","text":"The MCP tool names still use the tiered_memory_ prefix for compatibility.","tags":["docs","retrieval","mcp-tool-prefix"],"depth":1},{"id":"e3","text":"The project's MCP tools are now prefixed with tagmem_.","tags":["docs","retrieval","mcp-tool-prefix"],"depth":1},{"id":"e4","text":"The README lists the current tool names explicitly, but that is documentation surface rather than the naming rule itself.","tags":["docs","context","tool-listing"],"depth":2},{"id":"e5","text":"Older compatibility discussions may still mention tiered_memory_, which makes the prefix migration easy to misremember.","tags":["docs","notes","prefix-history"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger current-prefix vs legacy-prefix history distractor."}
{"id":"paraphrasex-014","query":"How do you override the Docker data root?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The Docker data root can be overridden with TAGMEM_DATA_ROOT.","tags":["docs","retrieval","docker-data-root"],"depth":1},{"id":"e2","text":"The Docker data root is hardcoded to /data/tagmem.","tags":["docs","retrieval","docker-data-root"],"depth":1},{"id":"e3","text":"To override the Docker data root, set TAGMEM_DATA_ROOT.","tags":["docs","retrieval","docker-data-root"],"depth":1},{"id":"e4","text":"The default data root lives under the user's home directory, but that is the default location rather than the override mechanism.","tags":["docs","context","default-path"],"depth":2},{"id":"e5","text":"Install notes mention the environment variable and the default path in nearby sections, which makes the two easy to conflate.","tags":["docs","notes","docker-data-root"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Upgraded paraphrase case with dual correct phrasings and a stronger override-vs-default-path distinction."}
{"id":"paraphrasex-015","query":"What should the doctor command report?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The doctor command should report the execution device and runtime library.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"The doctor command only reports whether the process started.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"The doctor command is useful for checking CPU fallback.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-016","query":"What belongs in the benchmark package?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The benchmark package should include raw JSON outputs and machine specs.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"The benchmark package should only include a headline score.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"The methodology file lists exact commands and dataset hashes.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-017","query":"What is the correct import command name?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The CLI command for bringing data in is ingest, not mine.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"The CLI command for bringing data in should be called mine because it sounds more creative.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"The naming should stay boring and well-understood.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-018","query":"How do depth and tags relate?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Depth is secondary and tags are primary in the memory model.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"Depth should replace tags entirely as the primary organizer.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"Search may still use depth as a bias.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-019","query":"Why is the image generic?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The image should be generic enough to run mcp, doctor, or benchmarks.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"The image should be named specifically for OpenCode forever.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"The same image can be reused by multiple agent runtimes.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"paraphrasex-020","query":"How should install validation be done?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A fresh-user install test should be run from another machine, not assumed from local success.","tags":["docs","retrieval"],"depth":1},{"id":"e2","text":"A local maintainer machine is enough to validate public installation.","tags":["docs","retrieval"],"depth":1},{"id":"e3","text":"Anonymous GHCR pulls are part of the install story.","tags":["docs","context"],"depth":2}],"relevant_ids":["e1"],"notes":"Extra near-duplicate paraphrase case."}
{"id":"env-hard-001","query":"What document OCR endpoint does preview use?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The preview environment uses https://ocr-preview.internal for the document OCR endpoint.","tags":["preview","document-ocr-endpoint","infra"],"depth":1},{"id":"e2","text":"The production environment uses https://ocr.internal for the document OCR endpoint.","tags":["production","document-ocr-endpoint","infra"],"depth":1},{"id":"e3","text":"The development environment uses https://ocr-dev.internal for the document OCR endpoint.","tags":["development","document-ocr-endpoint","infra"],"depth":2},{"id":"e4","text":"The preview environment still keeps https://ocr-preview-readonly.internal for replaying historical OCR jobs.","tags":["preview","document-ocr-endpoint","replay"],"depth":2},{"id":"e5","text":"Before the preview split, staging temporarily used https://ocr-preview.internal during migration week.","tags":["staging","document-ocr-endpoint","historical"],"depth":3,"timestamp":"2025-11-14"},{"id":"e6","text":"The preview runbook tells operators to verify callbacks from the OCR endpoint before rotating credentials.","tags":["preview","document-ocr-endpoint","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with sibling environment, historical alias, and read-only shadow endpoint distractors."}
{"id":"env-hard-002","query":"What Kafka bootstrap server does failover use for billing events?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The failover environment uses kafka-dr.internal:9092 as the Kafka bootstrap server for billing events.","tags":["failover","billing-events","kafka-bootstrap","infra"],"depth":1},{"id":"e2","text":"The production environment uses kafka-prod.internal:9092 as the Kafka bootstrap server for billing events.","tags":["production","billing-events","kafka-bootstrap","infra"],"depth":1},{"id":"e3","text":"The staging environment uses kafka-staging.internal:9092 as the Kafka bootstrap server for billing events.","tags":["staging","billing-events","kafka-bootstrap","infra"],"depth":2},{"id":"e4","text":"The failover environment publishes audit events through kafka-dr.internal:9093 on a separate listener.","tags":["failover","audit-events","kafka-bootstrap","infra"],"depth":2},{"id":"e5","text":"An archived failover worksheet still lists kafka-recovery.internal:9092 from the pre-cutover topology.","tags":["failover","billing-events","historical"],"depth":3,"timestamp":"2025-08-02"},{"id":"e6","text":"The billing events consumer group is mirrored from production into failover every hour.","tags":["failover","billing-events","replication"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with service-family overlap, port-level distractor, and obsolete failover host."}
{"id":"entity-hard-001","query":"What editor does Maris prefer for schema reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Maris prefers Helix for schema reviews.","tags":["maris","schema-reviews","preference"],"depth":1},{"id":"e2","text":"Marius prefers Helix for schema reviews.","tags":["marius","schema-reviews","preference"],"depth":1},{"id":"e3","text":"Maris usually switches to VS Code when debugging generated fixtures.","tags":["maris","debugging","preference"],"depth":2},{"id":"e4","text":"Maren prefers Helix for RFC proofreading, not schema reviews.","tags":["maren","rfc-proofreading","preference"],"depth":2},{"id":"e5","text":"A planning note said Maris and Marius both wanted matching editor shortcuts during the migration week.","tags":["maris","marius","planning"],"depth":3},{"id":"e6","text":"Maris wrote the checklist for reviewing schema drift before releases.","tags":["maris","schema-reviews","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with near-name collisions, same preference value, and context-switch distractors."}
{"id":"entity-hard-002","query":"What breakfast does Jonelle bring to incident review mornings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Jonelle brings sesame bagels to incident review mornings.","tags":["jonelle","incident-review","preference"],"depth":1},{"id":"e2","text":"Joelle brings sesame bagels to incident review mornings.","tags":["joelle","incident-review","preference"],"depth":1},{"id":"e3","text":"Janelle brings seeded crackers to incident review mornings.","tags":["janelle","incident-review","preference"],"depth":2},{"id":"e4","text":"Jonelle brings fruit cups to customer interview days instead of bagels.","tags":["jonelle","customer-interviews","preference"],"depth":2},{"id":"e5","text":"The operations retro mentioned that Jonelle and Joelle both avoid sweet pastries during on-call weeks.","tags":["jonelle","joelle","ops-retro"],"depth":3},{"id":"e6","text":"Jonelle usually arrives first and labels the incident review breakfast tray.","tags":["jonelle","incident-review","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with phonetic name collisions and role-specific routine distractors."}
{"id":"time-hard-001","query":"When did we move benchmark raw artifacts to immutable paths?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We move benchmark raw artifacts to immutable paths in March.","tags":["timeline","benchmark-raw-artifacts","immutable-paths"],"depth":1,"timestamp":"2026-03-18"},{"id":"e2","text":"We planned to move benchmark raw artifacts to immutable paths in January.","tags":["planning","benchmark-raw-artifacts","immutable-paths"],"depth":2,"timestamp":"2026-01-09"},{"id":"e3","text":"We tested immutable artifact paths on a temporary branch in February.","tags":["trial","benchmark-raw-artifacts","immutable-paths"],"depth":2,"timestamp":"2026-02-21"},{"id":"e4","text":"We updated the release checklist for immutable artifact paths in April.","tags":["docs","benchmark-raw-artifacts","immutable-paths"],"depth":2,"timestamp":"2026-04-02"},{"id":"e5","text":"A later cleanup removed the old mutable-path publisher in May.","tags":["cleanup","benchmark-raw-artifacts","mutable-paths"],"depth":3,"timestamp":"2026-05-11"}],"relevant_ids":["e1"],"notes":"Hand-written time case with planning, trial, implementation, documentation, and cleanup milestones around one change."}
{"id":"time-hard-002","query":"When did we stop publishing the preview embedding index?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop publishing the preview embedding index in August.","tags":["timeline","preview-embedding-index","publishing"],"depth":1,"timestamp":"2026-08-07"},{"id":"e2","text":"We proposed stopping the preview embedding index in June.","tags":["planning","preview-embedding-index","publishing"],"depth":2,"timestamp":"2026-06-03"},{"id":"e3","text":"We ran one dry-run week without publishing the preview embedding index in July.","tags":["trial","preview-embedding-index","publishing"],"depth":2,"timestamp":"2026-07-12"},{"id":"e4","text":"We deleted the stale monitoring dashboard for that preview index in September.","tags":["cleanup","preview-embedding-index","monitoring"],"depth":2,"timestamp":"2026-09-05"},{"id":"e5","text":"The docs note about preview index retirement landed in October.","tags":["docs","preview-embedding-index","retirement"],"depth":3,"timestamp":"2026-10-01"}],"relevant_ids":["e1"],"notes":"Hand-written time case with proposal, dry run, actual change, cleanup, and documentation milestones."}
{"id":"state-hard-001","query":"What is the current default reranker in evaluation mode?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default reranker in evaluation mode is cross-encoder/ms-marco-MiniLM-L-6-v2.","tags":["current-default-reranker-eval","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current default reranker in evaluation mode used to be bge-reranker-base.","tags":["current-default-reranker-eval","historical"],"depth":2,"timestamp":"2025-12-14"},{"id":"e3","text":"The preview-only reranker in evaluation mode is mixedbread-ai/mxbai-rerank-base-v1.","tags":["current-default-reranker-eval","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An RFC proposed jina-reranker-v2-base-multilingual as a future experiment but it is not the default.","tags":["current-default-reranker-eval","proposal"],"depth":2,"timestamp":"2026-04-08"},{"id":"e5","text":"The runbook notes that evaluation mode and interactive mode now share the same reranker metrics dashboard.","tags":["current-default-reranker-eval","runbook"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with historical, preview, and proposed alternatives all plausible enough to mis-rank."}
{"id":"state-hard-002","query":"What is the current retention window for benchmark raw JSON artifacts?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current retention window for benchmark raw JSON artifacts is 45 days.","tags":["benchmark-raw-json-retention","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current retention window for benchmark raw JSON artifacts used to be 14 days.","tags":["benchmark-raw-json-retention","historical"],"depth":2,"timestamp":"2025-11-01"},{"id":"e3","text":"The preview bucket still expires benchmark raw JSON artifacts after 7 days.","tags":["benchmark-raw-json-retention","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A storage RFC argued for increasing benchmark raw JSON artifact retention to 90 days, but that proposal was not adopted.","tags":["benchmark-raw-json-retention","proposal"],"depth":2,"timestamp":"2026-03-26"},{"id":"e5","text":"The artifact cleanup job now logs separate counters for preview and canonical retention policies.","tags":["benchmark-raw-json-retention","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with current, historical, environment-specific, and proposed values competing on the same concept."}
{"id":"speaker-hard-001","query":"What did you suggest for handling contradictory memories during retrieval?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested prefer entries explicitly marked current or newer when contradictory memories describe the same fact.","tags":["contradictory-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just return whichever contradictory memory has the strongest lexical overlap.","tags":["contradictory-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested surfacing ambiguity when two contradictory entries are equally current instead of pretending one is authoritative.","tags":["contradictory-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later reviewed examples where stale runbooks outranked newer operational facts.","tags":["contradictory-memories","discussion"],"depth":2},{"id":"e5","text":"The benchmark notes mention contradiction handling as a likely future slice.","tags":["contradictory-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case with two complementary assistant recommendations and a tempting user shortcut."}
{"id":"speaker-hard-002","query":"What did you suggest for benchmark expansion without making it synthetic?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested adding a smaller number of dense hand-written cases instead of generating hundreds of shallow permutations.","tags":["benchmark-expansion-non-synthetic","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested scripting five hundred parameter swaps so volume alone would make the benchmark hard.","tags":["benchmark-expansion-non-synthetic","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested combining two confusion pressures inside one case while still keeping a single primary adversary label.","tags":["benchmark-expansion-non-synthetic","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how hand-written cases stay inspectable even when they carry more distractors.","tags":["benchmark-expansion-non-synthetic","discussion"],"depth":2},{"id":"e5","text":"The dataset design note still says cases should remain small enough to inspect by hand.","tags":["benchmark-expansion-non-synthetic","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case aligned with the dataset goal: harder cases from density and composition, not just more rows."}
{"id":"paraphrase-hard-001","query":"What is the quarantine period before a deleted memory can be purged?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A deleted memory stays in quarantine for 72 hours before it can be purged.","tags":["deletion","quarantine","retention"],"depth":1},{"id":"e2","text":"A deleted memory stays in quarantine for 24 hours before it can be purged.","tags":["deletion","quarantine","retention"],"depth":1},{"id":"e3","text":"Deleted memories remain tombstoned for three days before hard purge is allowed.","tags":["deletion","quarantine","tombstone"],"depth":1},{"id":"e4","text":"Export bundles are retained for 72 hours before cleanup begins.","tags":["exports","retention"],"depth":2},{"id":"e5","text":"The purge worker checks whether a memory is still under quarantine before deleting its vectors.","tags":["deletion","worker"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two correct phrasings, one wrong value, and same-number retention distractors."}
{"id":"paraphrase-hard-002","query":"How many exact tag matches are enough to skip embedding reranking?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Embedding reranking is skipped once a candidate has 4 exact tag matches.","tags":["ranking","tags","threshold"],"depth":1},{"id":"e2","text":"Embedding reranking is skipped once a candidate has 3 exact tag matches.","tags":["ranking","tags","threshold"],"depth":1},{"id":"e3","text":"Four exact tag hits are sufficient to bypass the embedding reranker.","tags":["ranking","tags","threshold"],"depth":1},{"id":"e4","text":"Lexical expansion is skipped after four exact path-segment matches in the docs indexer.","tags":["docs-indexer","threshold"],"depth":2},{"id":"e5","text":"The reranker threshold is evaluated after deduping candidates with identical ids.","tags":["ranking","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two semantically equivalent correct statements and nearby threshold distractors."}
{"id":"env-hard-003","query":"What vector store host does canary use for semantic search?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The canary environment uses qdrant-canary.internal for the semantic search vector store.","tags":["canary","semantic-search","vector-store","infra"],"depth":1},{"id":"e2","text":"The production environment uses qdrant-prod.internal for the semantic search vector store.","tags":["production","semantic-search","vector-store","infra"],"depth":1},{"id":"e3","text":"The preview environment uses qdrant-preview.internal for the semantic search vector store.","tags":["preview","semantic-search","vector-store","infra"],"depth":2},{"id":"e4","text":"The canary environment still reads archived embeddings from qdrant-canary-ro.internal during replay drills.","tags":["canary","semantic-search","vector-store","replay"],"depth":2},{"id":"e5","text":"A prelaunch checklist still mentions qdrant-shadow.internal from the week before canary traffic opened.","tags":["canary","semantic-search","historical"],"depth":3,"timestamp":"2025-09-17"},{"id":"e6","text":"The canary runbook says to compare semantic search latency against preview before promoting a model.","tags":["canary","semantic-search","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with canary-preview overlap, read-only shadow host, and obsolete prelaunch hostname."}
{"id":"env-hard-004","query":"What callback domain does sandbox use for partner SSO?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The sandbox environment uses sso-callback-sandbox.ecue.ai for the partner SSO callback domain.","tags":["sandbox","partner-sso","callback-domain","infra"],"depth":1},{"id":"e2","text":"The production environment uses sso-callback.ecue.ai for the partner SSO callback domain.","tags":["production","partner-sso","callback-domain","infra"],"depth":1},{"id":"e3","text":"The staging environment uses sso-callback-staging.ecue.ai for the partner SSO callback domain.","tags":["staging","partner-sso","callback-domain","infra"],"depth":2},{"id":"e4","text":"The sandbox environment sends test invite emails from sandbox-login.ecue.ai, which is not the callback domain.","tags":["sandbox","partner-sso","email-domain"],"depth":2},{"id":"e5","text":"An integration worksheet from the old tenant still lists callback-sandbox.ecue.ai before the hostname standardization.","tags":["sandbox","partner-sso","historical"],"depth":3,"timestamp":"2025-10-28"},{"id":"e6","text":"The partner SSO checklist warns that sandbox and staging use different certificate chains despite similar hostnames.","tags":["sandbox","partner-sso","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with similar domains, adjacent auth concepts, and a stale pre-standardization hostname."}
{"id":"env-hard-005","query":"What object prefix does disaster recovery use for nightly replay archives?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The disaster recovery environment uses the dr/replay-nightly/ object prefix for nightly replay archives.","tags":["disaster-recovery","replay-archives","object-prefix","infra"],"depth":1},{"id":"e2","text":"The production environment uses the prod/replay-nightly/ object prefix for nightly replay archives.","tags":["production","replay-archives","object-prefix","infra"],"depth":1},{"id":"e3","text":"The preview environment uses the preview/replay-nightly/ object prefix for nightly replay archives.","tags":["preview","replay-archives","object-prefix","infra"],"depth":2},{"id":"e4","text":"The disaster recovery environment writes ad hoc incident captures under dr/replay-manual/, not the nightly archive prefix.","tags":["disaster-recovery","replay-archives","incident-captures"],"depth":2},{"id":"e5","text":"A migration comment still references recovery/replay-nightly/ from before the environment was renamed to disaster recovery.","tags":["disaster-recovery","replay-archives","historical"],"depth":3,"timestamp":"2025-07-06"},{"id":"e6","text":"The archive cleanup job treats disaster recovery and production retention differently even when the prefix names look parallel.","tags":["disaster-recovery","replay-archives","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with renamed environment terminology, similar prefixes, and neighboring manual-capture storage paths."}
{"id":"entity-hard-003","query":"What notebook does Riona use for partner call notes?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Riona uses a black A5 dot-grid notebook for partner call notes.","tags":["riona","partner-calls","preference"],"depth":1},{"id":"e2","text":"Riordan uses a black A5 dot-grid notebook for partner call notes.","tags":["riordan","partner-calls","preference"],"depth":1},{"id":"e3","text":"Riona switches to a plain legal pad for outage command sessions.","tags":["riona","outages","preference"],"depth":2},{"id":"e4","text":"Fiona uses a black A5 dot-grid notebook for weekly hiring debriefs.","tags":["fiona","hiring-debriefs","preference"],"depth":2},{"id":"e5","text":"A coordination note said Riona and Riordan both wanted the same notebook pocket for partner business cards.","tags":["riona","riordan","planning"],"depth":3},{"id":"e6","text":"Riona keeps a running index of partner call notes on the inside cover.","tags":["riona","partner-calls","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with same object preference across near names but different contexts and work modes."}
{"id":"entity-hard-004","query":"What tea does Kamilah drink during release freeze days?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Kamilah drinks smoky lapsang souchong during release freeze days.","tags":["kamilah","release-freeze","preference"],"depth":1},{"id":"e2","text":"Kamila drinks smoky lapsang souchong during release freeze days.","tags":["kamila","release-freeze","preference"],"depth":1},{"id":"e3","text":"Kamilah drinks peppermint tea during late support shifts instead of smoky tea.","tags":["kamilah","support-shifts","preference"],"depth":2},{"id":"e4","text":"Camilla drinks lapsang souchong during architecture reviews, not release freezes.","tags":["camilla","architecture-reviews","preference"],"depth":2},{"id":"e5","text":"A snack order thread noted that Kamilah and Kamila both avoid sweetened tea on freeze days.","tags":["kamilah","kamila","ops-thread"],"depth":3},{"id":"e6","text":"Kamilah says the smoky tea is her signal that the release freeze checklist has started.","tags":["kamilah","release-freeze","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with orthographic name collision and a role-specific ritual that changes by work context."}
{"id":"entity-hard-005","query":"What bike route does Daren take on office archive days?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Daren takes the river path on office archive days.","tags":["daren","office-archive-days","preference"],"depth":1},{"id":"e2","text":"Darren takes the river path on office archive days.","tags":["darren","office-archive-days","preference"],"depth":1},{"id":"e3","text":"Daren takes Oak Street when carrying prototype hardware instead of archive boxes.","tags":["daren","hardware-runs","preference"],"depth":2},{"id":"e4","text":"Taren takes the river path on volunteer cleanup mornings.","tags":["taren","cleanup-mornings","preference"],"depth":2},{"id":"e5","text":"The facilities note says Daren leaves archive keys at the river-side loading door after each run.","tags":["daren","office-archive-days","ops"],"depth":3},{"id":"e6","text":"A calendar reminder accidentally tagged Darren on Daren's archive route checklist once before it was corrected.","tags":["daren","darren","historical"],"depth":3,"timestamp":"2026-01-22"}],"relevant_ids":["e1"],"notes":"Hand-written entity case with near-identical names, same route distractor, and corrected scheduling metadata."}
{"id":"time-hard-003","query":"When did we rename the replay exporter to the archive publisher?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the replay exporter to the archive publisher in February.","tags":["timeline","replay-exporter","archive-publisher"],"depth":1,"timestamp":"2026-02-14"},{"id":"e2","text":"We first proposed renaming the replay exporter to the archive publisher in December.","tags":["planning","replay-exporter","archive-publisher"],"depth":2,"timestamp":"2025-12-03"},{"id":"e3","text":"We updated package comments to say archive publisher in January on a feature branch.","tags":["trial","replay-exporter","archive-publisher"],"depth":2,"timestamp":"2026-01-09"},{"id":"e4","text":"We removed the last replay exporter alias from docs in March.","tags":["docs","replay-exporter","archive-publisher"],"depth":2,"timestamp":"2026-03-04"},{"id":"e5","text":"A cleanup in April deleted metrics still labeled replay exporter.","tags":["cleanup","replay-exporter","metrics"],"depth":3,"timestamp":"2026-04-12"}],"relevant_ids":["e1"],"notes":"Hand-written time case with proposal, branch trial, actual rename, docs catch-up, and metrics cleanup."}
{"id":"time-hard-004","query":"When did we start requiring explicit schema versions in import bundles?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We start requiring explicit schema versions in import bundles in July.","tags":["timeline","import-bundles","schema-versions"],"depth":1,"timestamp":"2026-07-15"},{"id":"e2","text":"We discussed requiring explicit schema versions in import bundles in May.","tags":["planning","import-bundles","schema-versions"],"depth":2,"timestamp":"2026-05-08"},{"id":"e3","text":"A validator warning for missing schema versions shipped in June before enforcement became mandatory.","tags":["trial","import-bundles","schema-versions"],"depth":2,"timestamp":"2026-06-11"},{"id":"e4","text":"The migration guide for old bundles was published in August.","tags":["docs","import-bundles","schema-versions"],"depth":2,"timestamp":"2026-08-02"},{"id":"e5","text":"A later cleanup removed the compatibility fallback in September.","tags":["cleanup","import-bundles","schema-versions"],"depth":3,"timestamp":"2026-09-09"}],"relevant_ids":["e1"],"notes":"Hand-written time case with warning-before-enforcement and documentation lag around the actual policy change."}
{"id":"time-hard-005","query":"When did we stop using the shadow index during nightly evaluation?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop using the shadow index during nightly evaluation in November.","tags":["timeline","shadow-index","nightly-evaluation"],"depth":1,"timestamp":"2026-11-06"},{"id":"e2","text":"We raised the idea of removing the shadow index in September.","tags":["planning","shadow-index","nightly-evaluation"],"depth":2,"timestamp":"2026-09-12"},{"id":"e3","text":"One experimental nightly evaluation skipped the shadow index in October but was rolled back the next day.","tags":["trial","shadow-index","nightly-evaluation"],"depth":2,"timestamp":"2026-10-18"},{"id":"e4","text":"We archived the shadow-index dashboard in December.","tags":["cleanup","shadow-index","nightly-evaluation"],"depth":2,"timestamp":"2026-12-03"},{"id":"e5","text":"The docs retrospect on simplifying nightly evaluation landed in January.","tags":["docs","shadow-index","nightly-evaluation"],"depth":3,"timestamp":"2027-01-05"}],"relevant_ids":["e1"],"notes":"Hand-written time case with a reverted experiment that competes with the actual permanent change."}
{"id":"state-hard-003","query":"What is the current default deduplication window for diary ingestion?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default deduplication window for diary ingestion is 36 hours.","tags":["diary-ingestion-dedup-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current default deduplication window for diary ingestion used to be 12 hours.","tags":["diary-ingestion-dedup-window","historical"],"depth":2,"timestamp":"2025-11-18"},{"id":"e3","text":"The preview workspace still uses a 6 hour deduplication window for diary ingestion experiments.","tags":["diary-ingestion-dedup-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A proposed long-horizon setting would have expanded the diary ingestion deduplication window to 72 hours, but that was rejected as too sticky.","tags":["diary-ingestion-dedup-window","proposal"],"depth":2,"timestamp":"2026-03-14"},{"id":"e5","text":"The ingestion metrics dashboard now breaks out duplicates seen inside and outside the active deduplication window.","tags":["diary-ingestion-dedup-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with realistic interval confusion: old default, preview override, and never-adopted proposal."}
{"id":"state-hard-004","query":"What is the current canonical field name for source provenance?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for source provenance is source_ref.","tags":["source-provenance-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current canonical field name for source provenance used to be origin_ref.","tags":["source-provenance-field","historical"],"depth":2,"timestamp":"2025-10-30"},{"id":"e3","text":"An internal preview adapter still emits source_uri before normalization.","tags":["source-provenance-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A schema proposal recommended provenance_ref as a more explicit replacement, but that proposal was declined.","tags":["source-provenance-field","proposal"],"depth":2,"timestamp":"2026-02-19"},{"id":"e5","text":"The import normalizer maps legacy provenance names onto the canonical field before validation.","tags":["source-provenance-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with naming drift and a preview adapter emitting a plausible but not canonical field."}
{"id":"state-hard-005","query":"What is the current maximum candidate pool before reranking?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current maximum candidate pool before reranking is 40 entries.","tags":["max-candidate-pool-before-reranking","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The current maximum candidate pool before reranking used to be 25 entries.","tags":["max-candidate-pool-before-reranking","historical"],"depth":2,"timestamp":"2025-12-09"},{"id":"e3","text":"The lightweight preview path still caps the candidate pool before reranking at 20 entries.","tags":["max-candidate-pool-before-reranking","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A benchmarking proposal asked for a 60 entry candidate pool before reranking, but that was not adopted as the default.","tags":["max-candidate-pool-before-reranking","proposal"],"depth":2,"timestamp":"2026-03-01"},{"id":"e5","text":"The reranking profiler records both retrieved and reranked candidate counts for each evaluation query.","tags":["max-candidate-pool-before-reranking","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with nearby numeric competitors that all sound operationally plausible."}
{"id":"speaker-hard-003","query":"What did you suggest for evaluating retrieval errors instead of just reporting a headline score?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keeping error buckets by adversary type so we could see whether systems fail on stale facts, entity confusion, or time confusion differently.","tags":["retrieval-error-analysis","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested reporting only one overall recall number so the benchmark would look cleaner.","tags":["retrieval-error-analysis","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested storing the retrieved ids so later analysis could inspect the exact wrong memories that won.","tags":["retrieval-error-analysis","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later reviewed how two systems with similar recall could still fail for very different reasons.","tags":["retrieval-error-analysis","discussion"],"depth":2},{"id":"e5","text":"The design notes still mention bucketed error counts in the scoring section.","tags":["retrieval-error-analysis","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case with two relevant assistant recommendations about measurement depth and post-hoc analysis."}
{"id":"speaker-hard-004","query":"What did you suggest for keeping renamed fields manageable in the dataset?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested preserving one canonical field name and normalizing older aliases at ingest time instead of teaching every consumer all historical names.","tags":["renamed-fields","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested leaving every historical field spelling in place because analysts could memorize the differences.","tags":["renamed-fields","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping a short alias map close to the schema so rename history stays inspectable.","tags":["renamed-fields","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how rename-heavy datasets quietly become benchmark traps for tools rather than retrieval systems.","tags":["renamed-fields","discussion"],"depth":2},{"id":"e5","text":"The ingest pipeline already contains some normalization logic for legacy names.","tags":["renamed-fields","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case focused on practical schema evolution rather than abstract style advice."}
{"id":"speaker-hard-005","query":"What did you suggest for using historical notes without letting them outrank current operations?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keeping historical notes searchable but biasing retrieval toward entries marked current when the query asks for present-day operational facts.","tags":["historical-notes-ranking","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested hiding historical notes entirely so nobody could ever retrieve an old mistake.","tags":["historical-notes-ranking","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested preserving timestamps and explicit historical tags so stale memories remain useful for audits and debugging.","tags":["historical-notes-ranking","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later looked at examples where old runbooks were still valuable, just not as the first answer to current-state questions.","tags":["historical-notes-ranking","discussion"],"depth":2},{"id":"e5","text":"The benchmark already includes several current-vs-stale retrieval patterns.","tags":["historical-notes-ranking","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case with realistic tradeoffs between auditability and current-state retrieval quality."}
{"id":"paraphrase-hard-003","query":"What confidence score is required before a guessed speaker tag is accepted?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A guessed speaker tag is only accepted once confidence reaches 0.92.","tags":["speaker-tags","confidence","threshold"],"depth":1},{"id":"e2","text":"A guessed speaker tag is only accepted once confidence reaches 0.82.","tags":["speaker-tags","confidence","threshold"],"depth":1},{"id":"e3","text":"The system requires at least 0.92 confidence before it keeps an inferred speaker tag.","tags":["speaker-tags","confidence","threshold"],"depth":1},{"id":"e4","text":"A 0.92 score is enough to auto-link a remembered project alias, but that is a different classifier.","tags":["aliases","confidence"],"depth":2},{"id":"e5","text":"Low-confidence speaker guesses are retained as debugging signals without being committed as tags.","tags":["speaker-tags","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two equivalent correct statements, one nearby wrong threshold, and a same-number classifier distractor."}
{"id":"paraphrase-hard-004","query":"How often is the contradiction scan run on active memories?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The contradiction scan runs every 6 hours on active memories.","tags":["contradiction-scan","schedule"],"depth":1},{"id":"e2","text":"The contradiction scan runs every 12 hours on active memories.","tags":["contradiction-scan","schedule"],"depth":1},{"id":"e3","text":"Active memories are checked for contradictions on a six hour cadence.","tags":["contradiction-scan","schedule"],"depth":1},{"id":"e4","text":"Inactive archive compaction also runs every 6 hours, but that is a separate maintenance job.","tags":["archive-compaction","schedule"],"depth":2},{"id":"e5","text":"The contradiction scan can also be triggered manually after large backfills.","tags":["contradiction-scan","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two correct phrasings and one tempting same-number maintenance-job distractor."}
{"id":"paraphrase-hard-005","query":"What is the minimum span before two diary entries are considered a duplicate cluster?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Two diary entries are treated as a duplicate cluster once they fall within a 90 minute span.","tags":["diary","duplicates","time-window"],"depth":1},{"id":"e2","text":"Two diary entries are treated as a duplicate cluster once they fall within a 60 minute span.","tags":["diary","duplicates","time-window"],"depth":1},{"id":"e3","text":"A duplicate cluster begins when diary entries land no more than ninety minutes apart.","tags":["diary","duplicates","time-window"],"depth":1},{"id":"e4","text":"Meeting-note consolidation also uses a 90 minute merge window, but that is outside diary ingestion.","tags":["meeting-notes","time-window"],"depth":2},{"id":"e5","text":"The duplicate cluster window is checked after author normalization and before semantic similarity scoring.","tags":["diary","duplicates","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with equivalent correct statements, a nearby wrong interval, and a parallel merge-window distractor."}
{"id":"env-hard-006","query":"What Redis sentinel host does warm standby use for session failover?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The warm standby environment uses redis-sentinel-standby.internal for session failover.","tags":["warm-standby","session-failover","redis-sentinel","infra"],"depth":1},{"id":"e2","text":"The production environment uses redis-sentinel-prod.internal for session failover.","tags":["production","session-failover","redis-sentinel","infra"],"depth":1},{"id":"e3","text":"The preview environment uses redis-sentinel-preview.internal for session failover.","tags":["preview","session-failover","redis-sentinel","infra"],"depth":2},{"id":"e4","text":"The warm standby environment still exposes redis-standby-readonly.internal for replaying historical session traces.","tags":["warm-standby","session-failover","readonly"],"depth":2},{"id":"e5","text":"A migration note from before the rename still refers to redis-sentinel-dr.internal for the same warm standby role.","tags":["warm-standby","session-failover","historical"],"depth":3,"timestamp":"2025-09-03"},{"id":"e6","text":"The failover checklist compares warm standby promotion lag against production every morning.","tags":["warm-standby","session-failover","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with renamed environment terminology, read-only sibling host, and old DR naming."}
{"id":"env-hard-007","query":"What upload bucket does partner preview use for signed document intake?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The partner preview environment uses signed-docs-partner-preview for signed document intake uploads.","tags":["partner-preview","signed-document-intake","upload-bucket","infra"],"depth":1},{"id":"e2","text":"The production environment uses signed-docs-prod for signed document intake uploads.","tags":["production","signed-document-intake","upload-bucket","infra"],"depth":1},{"id":"e3","text":"The staging environment uses signed-docs-staging for signed document intake uploads.","tags":["staging","signed-document-intake","upload-bucket","infra"],"depth":2},{"id":"e4","text":"The partner preview environment exports review PDFs to partner-preview-pdf-exports, which is not the intake bucket.","tags":["partner-preview","signed-document-intake","exports"],"depth":2},{"id":"e5","text":"An onboarding worksheet still lists signed-docs-preview-partner from before the bucket naming order was standardized.","tags":["partner-preview","signed-document-intake","historical"],"depth":3,"timestamp":"2025-10-15"},{"id":"e6","text":"The signed intake runbook warns that partner preview and staging share IAM patterns but not bucket names.","tags":["partner-preview","signed-document-intake","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with name-order confusion, adjacent export storage, and partner-specific preview terminology."}
{"id":"env-hard-008","query":"What metrics pushgateway does lab use for embedding bake-offs?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The lab environment uses pushgateway-lab.internal for embedding bake-off metrics.","tags":["lab","embedding-bakeoffs","metrics-pushgateway","infra"],"depth":1},{"id":"e2","text":"The production environment uses pushgateway-prod.internal for embedding bake-off metrics.","tags":["production","embedding-bakeoffs","metrics-pushgateway","infra"],"depth":1},{"id":"e3","text":"The benchmark preview environment uses pushgateway-bench-preview.internal for embedding bake-off metrics.","tags":["benchmark-preview","embedding-bakeoffs","metrics-pushgateway","infra"],"depth":2},{"id":"e4","text":"The lab environment ships long-term metrics to vm-lab.internal after the pushgateway step.","tags":["lab","embedding-bakeoffs","metrics-backend"],"depth":2},{"id":"e5","text":"A retired lab note still mentions pgw-lab.internal from before hostnames were expanded for clarity.","tags":["lab","embedding-bakeoffs","historical"],"depth":3,"timestamp":"2025-08-19"},{"id":"e6","text":"Operators use the lab pushgateway only during bake-offs, not for routine benchmark runs.","tags":["lab","embedding-bakeoffs","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written environment case with pushgateway vs backend confusion, lab vs benchmark-preview overlap, and shortened legacy hostname."}
{"id":"entity-hard-006","query":"What mug does Sorelle use during benchmark triage?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Sorelle uses a white enamel camping mug during benchmark triage.","tags":["sorelle","benchmark-triage","preference"],"depth":1},{"id":"e2","text":"Sorrel uses a white enamel camping mug during benchmark triage.","tags":["sorrel","benchmark-triage","preference"],"depth":1},{"id":"e3","text":"Sorelle switches to a tall ceramic mug during customer demos.","tags":["sorelle","customer-demos","preference"],"depth":2},{"id":"e4","text":"Coralie uses an enamel camping mug for warehouse inventory counts, not benchmark triage.","tags":["coralie","inventory-counts","preference"],"depth":2},{"id":"e5","text":"A facilities note said Sorelle and Sorrel requested identical mug hooks in the shared kitchen.","tags":["sorelle","sorrel","facilities"],"depth":3},{"id":"e6","text":"Sorelle says the enamel mug means she expects a long triage session.","tags":["sorelle","benchmark-triage","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with almost identical names, same object preference, and context-specific switching."}
{"id":"entity-hard-007","query":"What snack does Imani bring to late schema review sessions?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Imani brings salted pistachios to late schema review sessions.","tags":["imani","schema-review","preference"],"depth":1},{"id":"e2","text":"Iman brings salted pistachios to late schema review sessions.","tags":["iman","schema-review","preference"],"depth":1},{"id":"e3","text":"Imani brings dried mango to customer roadmap workshops instead of pistachios.","tags":["imani","roadmap-workshops","preference"],"depth":2},{"id":"e4","text":"Amani brings salted pistachios to security review nights, not schema review sessions.","tags":["amani","security-review","preference"],"depth":2},{"id":"e5","text":"A planning thread noted that Imani and Iman both prefer unsweetened snacks during late work.","tags":["imani","iman","planning"],"depth":3},{"id":"e6","text":"Imani leaves an extra cup of pistachio shells near the schema drift whiteboard.","tags":["imani","schema-review","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with near-name and near-context confusion, plus identical snack across different people."}
{"id":"entity-hard-008","query":"What train does Niko take home after overnight incident duty?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Niko takes the Blue Line home after overnight incident duty.","tags":["niko","overnight-incident-duty","preference"],"depth":1},{"id":"e2","text":"Nico takes the Blue Line home after overnight incident duty.","tags":["nico","overnight-incident-duty","preference"],"depth":1},{"id":"e3","text":"Niko takes the Green Line after daytime office visits instead of incident duty.","tags":["niko","office-visits","preference"],"depth":2},{"id":"e4","text":"Miko takes the Blue Line after support training sessions, not overnight incident duty.","tags":["miko","support-training","preference"],"depth":2},{"id":"e5","text":"A duty roster once auto-corrected Niko to Nico on the transit reimbursement form before finance fixed it.","tags":["niko","nico","historical"],"depth":3,"timestamp":"2026-02-06"},{"id":"e6","text":"Niko says the Blue Line is the only route quiet enough after night incidents.","tags":["niko","overnight-incident-duty","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Hand-written entity case with transliterated-name confusion and commute preferences that change by work mode."}
{"id":"time-hard-006","query":"When did we make contradiction flags visible in the review UI?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We make contradiction flags visible in the review UI in April.","tags":["timeline","contradiction-flags","review-ui"],"depth":1,"timestamp":"2026-04-22"},{"id":"e2","text":"We designed contradiction flags for the review UI in February.","tags":["planning","contradiction-flags","review-ui"],"depth":2,"timestamp":"2026-02-10"},{"id":"e3","text":"A hidden feature-flag trial showed contradiction flags in March for internal reviewers only.","tags":["trial","contradiction-flags","review-ui"],"depth":2,"timestamp":"2026-03-14"},{"id":"e4","text":"The reviewer guide explaining contradiction flags was published in May.","tags":["docs","contradiction-flags","review-ui"],"depth":2,"timestamp":"2026-05-03"},{"id":"e5","text":"A later cleanup removed the old warning badge implementation in June.","tags":["cleanup","contradiction-flags","review-ui"],"depth":3,"timestamp":"2026-06-09"}],"relevant_ids":["e1"],"notes":"Hand-written time case with design, gated trial, actual rollout, docs, and cleanup stages around one UI feature."}
{"id":"time-hard-007","query":"When did we stop indexing assistant signatures into search?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop indexing assistant signatures into search in September.","tags":["timeline","assistant-signatures","search-indexing"],"depth":1,"timestamp":"2026-09-16"},{"id":"e2","text":"We proposed stopping assistant signature indexing in July.","tags":["planning","assistant-signatures","search-indexing"],"depth":2,"timestamp":"2026-07-07"},{"id":"e3","text":"A temporary branch skipped assistant signatures during indexing in August, but it was not yet the default behavior.","tags":["trial","assistant-signatures","search-indexing"],"depth":2,"timestamp":"2026-08-18"},{"id":"e4","text":"We removed the legacy signature-specific search tests in October.","tags":["cleanup","assistant-signatures","search-indexing"],"depth":2,"timestamp":"2026-10-04"},{"id":"e5","text":"The search quality note explaining the change landed in November.","tags":["docs","assistant-signatures","search-indexing"],"depth":3,"timestamp":"2026-11-02"}],"relevant_ids":["e1"],"notes":"Hand-written time case with a realistic indexing-policy change surrounded by planning, trial, cleanup, and documentation."}
{"id":"time-hard-008","query":"When did we require explicit speaker attribution on imported transcripts?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require explicit speaker attribution on imported transcripts in January.","tags":["timeline","imported-transcripts","speaker-attribution"],"depth":1,"timestamp":"2026-01-20"},{"id":"e2","text":"We discussed explicit speaker attribution for imported transcripts in November.","tags":["planning","imported-transcripts","speaker-attribution"],"depth":2,"timestamp":"2025-11-27"},{"id":"e3","text":"A validator warning about missing speaker attribution was introduced in December before enforcement started.","tags":["trial","imported-transcripts","speaker-attribution"],"depth":2,"timestamp":"2025-12-13"},{"id":"e4","text":"The transcript import migration guide was updated in February.","tags":["docs","imported-transcripts","speaker-attribution"],"depth":2,"timestamp":"2026-02-08"},{"id":"e5","text":"A compatibility fallback for anonymous transcript lines was removed in March.","tags":["cleanup","imported-transcripts","speaker-attribution"],"depth":3,"timestamp":"2026-03-12"}],"relevant_ids":["e1"],"notes":"Hand-written time case with warning-before-enforcement and a cleanup milestone that comes after the actual requirement date."}
{"id":"state-hard-006","query":"What is the current maximum age for 'current' facts before they require review?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current maximum age for facts still marked current before they require review is 30 days.","tags":["current-fact-review-age","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The maximum age for facts still marked current used to be 14 days before review was required.","tags":["current-fact-review-age","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"The preview review workspace still tests a 7 day age threshold for facts marked current.","tags":["current-fact-review-age","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An audit proposal argued for a 45 day review threshold for current facts, but that was not adopted.","tags":["current-fact-review-age","proposal"],"depth":2,"timestamp":"2026-03-05"},{"id":"e5","text":"The review scheduler now emits separate counts for overdue current facts and merely aging current facts.","tags":["current-fact-review-age","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with plausible nearby age thresholds from old defaults, preview experiments, and audit proposals."}
{"id":"state-hard-007","query":"What is the current canonical tag for manually verified contradictions?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical tag for manually verified contradictions is contradiction_confirmed.","tags":["manual-contradiction-tag","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical tag for manually verified contradictions used to be contradiction_verified.","tags":["manual-contradiction-tag","historical"],"depth":2,"timestamp":"2025-10-12"},{"id":"e3","text":"An internal preview tool still emits contradiction_manual until its tag map is normalized.","tags":["manual-contradiction-tag","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A naming RFC recommended contradiction_reviewed as a more human-friendly replacement, but that proposal was rejected.","tags":["manual-contradiction-tag","proposal"],"depth":2,"timestamp":"2026-02-23"},{"id":"e5","text":"The normalization step maps legacy contradiction tags onto the canonical one before indexing.","tags":["manual-contradiction-tag","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case centered on tag rename drift and a preview emitter that still uses a tempting near-match."}
{"id":"state-hard-008","query":"What is the current default sort order for audit timelines?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default sort order for audit timelines is newest_first.","tags":["audit-timeline-sort-order","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default sort order for audit timelines used to be oldest_first.","tags":["audit-timeline-sort-order","historical"],"depth":2,"timestamp":"2025-11-09"},{"id":"e3","text":"A legacy audit export still writes oldest_first unless a caller overrides it explicitly.","tags":["audit-timeline-sort-order","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A UX proposal recommended grouped_by_day as the default audit timeline sort order, but that idea was not adopted.","tags":["audit-timeline-sort-order","proposal"],"depth":2,"timestamp":"2026-01-31"},{"id":"e5","text":"The audit viewer remembers a user's override, but the canonical default remains newest_first.","tags":["audit-timeline-sort-order","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Hand-written state-update case with default-vs-export-vs-user-override confusion and a rejected UI proposal."}
{"id":"speaker-hard-006","query":"What did you suggest for handling reverted experiments in benchmark memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested storing reverted experiments as historical milestones so the benchmark can test whether systems confuse a dry run with the final rollout.","tags":["reverted-experiments","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested deleting every reverted experiment note so the dataset would stay simpler.","tags":["reverted-experiments","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested wording those entries so it is obvious they were trials or reversions rather than permanent state changes.","tags":["reverted-experiments","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later reviewed several rollout histories where the reverted trial was lexically closer to the query than the real launch.","tags":["reverted-experiments","discussion"],"depth":2},{"id":"e5","text":"Many of the hard time cases now intentionally include dry runs and cleanups around the actual change.","tags":["reverted-experiments","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case about preserving trial history without turning reverted experiments into false current state."}
{"id":"speaker-hard-007","query":"What did you suggest for making entity confusion feel realistic instead of toy-like?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested giving near-name people overlapping habits or tools so the wrong memory sounds believable instead of obviously fake.","tags":["realistic-entity-confusion","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just changing one letter in a name while keeping everything else flat and repetitive.","tags":["realistic-entity-confusion","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested changing the work context around the same preference so the benchmark tests whether systems retrieve the right person and the right situation together.","tags":["realistic-entity-confusion","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later talked about how realistic confusion comes from overlap, not from random noise.","tags":["realistic-entity-confusion","discussion"],"depth":2},{"id":"e5","text":"The newer entity cases now mix person-name collisions with role-specific routines.","tags":["realistic-entity-confusion","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case aligned with the manual dataset direction: believable overlap rather than shallow string tricks."}
{"id":"speaker-hard-008","query":"What did you suggest for keeping contradiction cases inspectable by humans?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested keeping each contradiction case small enough that a reviewer can reconstruct the timeline without reading a whole chat log.","tags":["inspectable-contradiction-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested making each contradiction case huge so scale alone would create difficulty.","tags":["inspectable-contradiction-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested adding only the most confusing nearby milestones around the true fact rather than every possible adjacent event.","tags":["inspectable-contradiction-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later compared compact hard cases against bloated ones and found the compact ones easier to audit and often harder to retrieve correctly.","tags":["inspectable-contradiction-cases","discussion"],"depth":2},{"id":"e5","text":"The design notes already emphasize that cases should remain small enough to inspect by hand.","tags":["inspectable-contradiction-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written speaker case about density over size: difficult but still auditable contradiction-focused examples."}
{"id":"paraphrase-hard-006","query":"What score is needed before an alias match is treated as exact?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An alias match is treated as exact once its normalization score reaches 0.97.","tags":["alias-matching","score","threshold"],"depth":1},{"id":"e2","text":"An alias match is treated as exact once its normalization score reaches 0.87.","tags":["alias-matching","score","threshold"],"depth":1},{"id":"e3","text":"The system upgrades an alias match to exact only at a normalization score of at least 0.97.","tags":["alias-matching","score","threshold"],"depth":1},{"id":"e4","text":"A 0.97 score is also used by the duplicate-title detector, but that threshold belongs to a different step.","tags":["duplicate-titles","score"],"depth":2},{"id":"e5","text":"Sub-exact alias matches are still stored as soft evidence for debugging.","tags":["alias-matching","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two correct phrasings, one near-threshold distractor, and a same-number threshold from another subsystem."}
{"id":"paraphrase-hard-007","query":"How many contradictory entries trigger a manual review queue?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A manual review queue is triggered once 3 contradictory entries cluster around the same fact.","tags":["contradictions","review-queue","threshold"],"depth":1},{"id":"e2","text":"A manual review queue is triggered once 2 contradictory entries cluster around the same fact.","tags":["contradictions","review-queue","threshold"],"depth":1},{"id":"e3","text":"Three contradictory entries attached to one fact are enough to open a manual review queue.","tags":["contradictions","review-queue","threshold"],"depth":1},{"id":"e4","text":"Three duplicate diary entries are enough to open a deduplication queue, but that is a different workflow.","tags":["duplicates","review-queue"],"depth":2},{"id":"e5","text":"The contradiction review queue is checked after timeline consolidation, not before it.","tags":["contradictions","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with two correct statements, a nearby lower threshold, and a same-number queue trigger in another workflow."}
{"id":"paraphrase-hard-008","query":"What is the default cooldown before a retracted fact can be reintroduced?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A retracted fact has a default 48 hour cooldown before it can be reintroduced.","tags":["retracted-facts","cooldown","policy"],"depth":1},{"id":"e2","text":"A retracted fact has a default 24 hour cooldown before it can be reintroduced.","tags":["retracted-facts","cooldown","policy"],"depth":1},{"id":"e3","text":"The default policy waits forty-eight hours before a retracted fact may be introduced again.","tags":["retracted-facts","cooldown","policy"],"depth":1},{"id":"e4","text":"The export retry worker also uses a 48 hour cooldown after repeated failures, but that is unrelated to fact reintroduction.","tags":["export-retries","cooldown"],"depth":2},{"id":"e5","text":"A manually approved override can bypass the cooldown, but only for audited corrections.","tags":["retracted-facts","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Hand-written near-duplicate case with equivalent correct phrasings, a shorter wrong cooldown, and a same-duration operational distractor."}
{"id":"env-hard-009","query":"What billing ledger host does finance-sandbox use for invoice rehearsal?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The finance-sandbox environment uses ledger-sandbox.internal for invoice rehearsal billing ledger traffic.","tags":["finance-sandbox","invoice-rehearsal","billing-ledger","infra"],"depth":1},{"id":"e2","text":"The production environment uses ledger-prod.internal for invoice rehearsal billing ledger traffic.","tags":["production","invoice-rehearsal","billing-ledger","infra"],"depth":1},{"id":"e3","text":"The accounting-preview environment uses ledger-preview.internal for invoice rehearsal billing ledger traffic.","tags":["accounting-preview","invoice-rehearsal","billing-ledger","infra"],"depth":2},{"id":"e4","text":"Finance-sandbox also reads old invoices from ledger-sandbox-ro.internal during audits, which is not the primary ledger host.","tags":["finance-sandbox","invoice-rehearsal","readonly"],"depth":2},{"id":"e5","text":"A finance onboarding note still mentions sandbox-ledger.internal from before service names were normalized.","tags":["finance-sandbox","invoice-rehearsal","historical"],"depth":3,"timestamp":"2025-09-09"},{"id":"e6","text":"The rehearsal checklist warns that finance-sandbox and accounting-preview use parallel credentials but different hosts.","tags":["finance-sandbox","invoice-rehearsal","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse environment case from billing operations with sandbox vs preview and historical hostname confusion."}
{"id":"env-hard-010","query":"What document store does legal-review use for redline packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The legal-review environment uses docs-legal.internal for redline packet storage.","tags":["legal-review","redline-packets","document-store","infra"],"depth":1},{"id":"e2","text":"The production environment uses docs-prod.internal for redline packet storage.","tags":["production","redline-packets","document-store","infra"],"depth":1},{"id":"e3","text":"The compliance-preview environment uses docs-compliance-preview.internal for redline packet storage.","tags":["compliance-preview","redline-packets","document-store","infra"],"depth":2},{"id":"e4","text":"Legal-review sends packet exports through files-legal-export.internal, which is not the document store.","tags":["legal-review","redline-packets","exports"],"depth":2},{"id":"e5","text":"An archived merger checklist still refers to redlines-legal.internal from the old naming scheme.","tags":["legal-review","redline-packets","historical"],"depth":3,"timestamp":"2025-08-27"},{"id":"e6","text":"The legal hold runbook notes that legal-review and compliance-preview use different retention policies even when both store redline packets.","tags":["legal-review","redline-packets","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse environment case grounded in legal workflow storage with neighboring export and compliance systems."}
{"id":"env-hard-011","query":"What mail relay does support-training use for simulated escalations?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The support-training environment uses relay-support-training.internal for simulated escalation email.","tags":["support-training","simulated-escalations","mail-relay","infra"],"depth":1},{"id":"e2","text":"The production environment uses relay-prod.internal for simulated escalation email.","tags":["production","simulated-escalations","mail-relay","infra"],"depth":1},{"id":"e3","text":"The partner-demo environment uses relay-partner-demo.internal for simulated escalation email.","tags":["partner-demo","simulated-escalations","mail-relay","infra"],"depth":2},{"id":"e4","text":"Support-training also receives canned mailbox replies through inbound-support-training.internal, which is not the outbound relay.","tags":["support-training","simulated-escalations","inbound-mail"],"depth":2},{"id":"e5","text":"A trainer worksheet still lists relay-training.internal from before the support-training rename.","tags":["support-training","simulated-escalations","historical"],"depth":3,"timestamp":"2025-10-03"},{"id":"e6","text":"The training checklist warns that support-training intentionally uses different DKIM settings than partner-demo.","tags":["support-training","simulated-escalations","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse environment case around support operations with outbound vs inbound mail and renamed training environments."}
{"id":"env-hard-012","query":"What warehouse queue does returns-lab use for barcode replay?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The returns-lab environment uses queue-returns-lab.internal for barcode replay jobs.","tags":["returns-lab","barcode-replay","warehouse-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-returns-prod.internal for barcode replay jobs.","tags":["production","barcode-replay","warehouse-queue","infra"],"depth":1},{"id":"e3","text":"The logistics-preview environment uses queue-logistics-preview.internal for barcode replay jobs.","tags":["logistics-preview","barcode-replay","warehouse-queue","infra"],"depth":2},{"id":"e4","text":"Returns-lab archives replay traces in queue-returns-lab-deadletter.internal, which is not the primary warehouse queue.","tags":["returns-lab","barcode-replay","deadletter"],"depth":2},{"id":"e5","text":"A warehouse migration note still names returns-replay.internal from before queue naming became explicit.","tags":["returns-lab","barcode-replay","historical"],"depth":3,"timestamp":"2025-07-14"},{"id":"e6","text":"The replay runbook notes that returns-lab and logistics-preview share fixtures but not queue names.","tags":["returns-lab","barcode-replay","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse environment case from warehouse operations with dead-letter and historical queue naming distractors."}
{"id":"entity-hard-009","query":"What pen does Leorin use for vendor security questionnaires?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Leorin uses a blue extra-fine gel pen for vendor security questionnaires.","tags":["leorin","vendor-security-questionnaires","preference"],"depth":1},{"id":"e2","text":"Leora uses a blue extra-fine gel pen for vendor security questionnaires.","tags":["leora","vendor-security-questionnaires","preference"],"depth":1},{"id":"e3","text":"Leorin switches to a black felt-tip pen for contract markup sessions.","tags":["leorin","contract-markup","preference"],"depth":2},{"id":"e4","text":"Lorien uses a blue extra-fine gel pen for audit interviews, not vendor questionnaires.","tags":["lorien","audit-interviews","preference"],"depth":2},{"id":"e5","text":"A procurement note said Leorin and Leora both preferred pens that do not smear on glossy forms.","tags":["leorin","leora","procurement"],"depth":3},{"id":"e6","text":"Leorin keeps the blue pen clipped to the questionnaire binder during every vendor review.","tags":["leorin","vendor-security-questionnaires","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse entity case from vendor security workflow with overlapping stationery preferences and near-name collisions."}
{"id":"entity-hard-010","query":"What breakfast does Tamsin pack for payroll close mornings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Tamsin packs hard-boiled eggs and rye toast for payroll close mornings.","tags":["tamsin","payroll-close","preference"],"depth":1},{"id":"e2","text":"Tamzin packs hard-boiled eggs and rye toast for payroll close mornings.","tags":["tamzin","payroll-close","preference"],"depth":1},{"id":"e3","text":"Tamsin packs yogurt and fruit for travel reimbursement reviews instead of payroll close.","tags":["tamsin","travel-reimbursements","preference"],"depth":2},{"id":"e4","text":"Taryn packs hard-boiled eggs and rye toast for inventory counts, not payroll close mornings.","tags":["taryn","inventory-counts","preference"],"depth":2},{"id":"e5","text":"A kitchen order note said Tamsin and Tamzin both avoid sugary breakfasts on deadline days.","tags":["tamsin","tamzin","ops-thread"],"depth":3},{"id":"e6","text":"Tamsin says the rye toast is easier to eat one-handed while reconciling payroll exceptions.","tags":["tamsin","payroll-close","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse entity case tied to finance operations with near-name overlap and task-specific routine changes."}
{"id":"entity-hard-011","query":"What notebook does Darien carry to site audit walk-throughs?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Darien carries a weatherproof spiral notebook to site audit walk-throughs.","tags":["darien","site-audits","preference"],"depth":1},{"id":"e2","text":"Darian carries a weatherproof spiral notebook to site audit walk-throughs.","tags":["darian","site-audits","preference"],"depth":1},{"id":"e3","text":"Darien carries a leather folio to budget meetings instead of the weatherproof notebook.","tags":["darien","budget-meetings","preference"],"depth":2},{"id":"e4","text":"Marien carries a weatherproof spiral notebook to field inspections, not audit walk-throughs.","tags":["marien","field-inspections","preference"],"depth":2},{"id":"e5","text":"A facilities checklist once misspelled Darien as Darian on the audit supply order before it was corrected.","tags":["darien","darian","historical"],"depth":3,"timestamp":"2026-01-11"},{"id":"e6","text":"Darien keeps audit sticker samples in the back pocket of the weatherproof notebook.","tags":["darien","site-audits","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse entity case from audit operations with realistic supply preferences and clerical name confusion."}
{"id":"entity-hard-012","query":"What drink does Mireya bring to overnight migration cutovers?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mireya brings iced yerba mate to overnight migration cutovers.","tags":["mireya","migration-cutovers","preference"],"depth":1},{"id":"e2","text":"Mireille brings iced yerba mate to overnight migration cutovers.","tags":["mireille","migration-cutovers","preference"],"depth":1},{"id":"e3","text":"Mireya brings hot jasmine tea to daytime onboarding sessions instead of yerba mate.","tags":["mireya","onboarding","preference"],"depth":2},{"id":"e4","text":"Mariela brings iced yerba mate to travel booking sprints, not migration cutovers.","tags":["mariela","travel-booking","preference"],"depth":2},{"id":"e5","text":"A night-shift snack order noted that Mireya and Mireille both preferred unsweetened caffeinated drinks.","tags":["mireya","mireille","night-shift"],"depth":3},{"id":"e6","text":"Mireya says the iced mate is her signal that the migration has entered the overnight phase.","tags":["mireya","migration-cutovers","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Diverse entity case from migration operations with similar names, same drink, and different work contexts."}
{"id":"time-hard-009","query":"When did we require receipts on travel reimbursements over $75?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require receipts on travel reimbursements over $75 in June.","tags":["timeline","travel-reimbursements","receipts-policy"],"depth":1,"timestamp":"2026-06-17"},{"id":"e2","text":"We proposed requiring receipts on travel reimbursements over $75 in April.","tags":["planning","travel-reimbursements","receipts-policy"],"depth":2,"timestamp":"2026-04-28"},{"id":"e3","text":"A finance warning banner for missing receipts shipped in May before the policy became mandatory.","tags":["trial","travel-reimbursements","receipts-policy"],"depth":2,"timestamp":"2026-05-13"},{"id":"e4","text":"The travel FAQ was updated for the new receipt threshold in July.","tags":["docs","travel-reimbursements","receipts-policy"],"depth":2,"timestamp":"2026-07-02"},{"id":"e5","text":"A later cleanup removed the old manual exception form in August.","tags":["cleanup","travel-reimbursements","receipts-policy"],"depth":3,"timestamp":"2026-08-06"}],"relevant_ids":["e1"],"notes":"Diverse time case from finance policy with proposal, warning phase, enforcement, docs, and cleanup."}
{"id":"time-hard-010","query":"When did we rename the vendor checklist to the supplier trust packet?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the vendor checklist to the supplier trust packet in October.","tags":["timeline","vendor-checklist","supplier-trust-packet"],"depth":1,"timestamp":"2026-10-09"},{"id":"e2","text":"We discussed renaming the vendor checklist to the supplier trust packet in August.","tags":["planning","vendor-checklist","supplier-trust-packet"],"depth":2,"timestamp":"2026-08-19"},{"id":"e3","text":"An internal draft began using supplier trust packet in September before the rename was official.","tags":["trial","vendor-checklist","supplier-trust-packet"],"depth":2,"timestamp":"2026-09-07"},{"id":"e4","text":"The procurement glossary removed the old vendor checklist term in November.","tags":["docs","vendor-checklist","supplier-trust-packet"],"depth":2,"timestamp":"2026-11-05"},{"id":"e5","text":"A dashboard cleanup removed the last vendor-checklist label in December.","tags":["cleanup","vendor-checklist","supplier-trust-packet"],"depth":3,"timestamp":"2026-12-04"}],"relevant_ids":["e1"],"notes":"Diverse time case about procurement terminology with draft usage preceding the actual rename."}
{"id":"time-hard-011","query":"When did we stop auto-approving low-risk contract redlines?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-approving low-risk contract redlines in March.","tags":["timeline","contract-redlines","auto-approval"],"depth":1,"timestamp":"2026-03-21"},{"id":"e2","text":"We proposed stopping auto-approval of low-risk contract redlines in January.","tags":["planning","contract-redlines","auto-approval"],"depth":2,"timestamp":"2026-01-30"},{"id":"e3","text":"One legal-review pilot disabled low-risk auto-approval in February, but only for merger paperwork.","tags":["trial","contract-redlines","auto-approval"],"depth":2,"timestamp":"2026-02-12"},{"id":"e4","text":"The contract review handbook was updated in April to reflect manual review for all redlines.","tags":["docs","contract-redlines","auto-approval"],"depth":2,"timestamp":"2026-04-03"},{"id":"e5","text":"A metrics cleanup removed the old low-risk auto-approval chart in May.","tags":["cleanup","contract-redlines","auto-approval"],"depth":3,"timestamp":"2026-05-15"}],"relevant_ids":["e1"],"notes":"Diverse time case from legal workflow with a pilot that could be mistaken for the actual policy change."}
{"id":"time-hard-012","query":"When did we enforce manager sign-off on onboarding equipment requests?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We enforce manager sign-off on onboarding equipment requests in September.","tags":["timeline","onboarding-equipment","manager-signoff"],"depth":1,"timestamp":"2026-09-18"},{"id":"e2","text":"We discussed manager sign-off for onboarding equipment requests in July.","tags":["planning","onboarding-equipment","manager-signoff"],"depth":2,"timestamp":"2026-07-25"},{"id":"e3","text":"A procurement warning email for missing manager approval began in August before sign-off became mandatory.","tags":["trial","onboarding-equipment","manager-signoff"],"depth":2,"timestamp":"2026-08-11"},{"id":"e4","text":"The onboarding checklist was updated in October after manager sign-off became required.","tags":["docs","onboarding-equipment","manager-signoff"],"depth":2,"timestamp":"2026-10-06"},{"id":"e5","text":"The old self-service fast path was removed in November.","tags":["cleanup","onboarding-equipment","manager-signoff"],"depth":3,"timestamp":"2026-11-09"}],"relevant_ids":["e1"],"notes":"Diverse time case from onboarding and procurement with warning-before-enforcement structure."}
{"id":"state-hard-009","query":"What is the current auto-escalation threshold for unassigned support tickets?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current auto-escalation threshold for unassigned support tickets is 45 minutes.","tags":["support-ticket-auto-escalation-threshold","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The auto-escalation threshold for unassigned support tickets used to be 30 minutes.","tags":["support-ticket-auto-escalation-threshold","historical"],"depth":2,"timestamp":"2025-11-22"},{"id":"e3","text":"The training queue still uses a 15 minute auto-escalation threshold for unassigned tickets.","tags":["support-ticket-auto-escalation-threshold","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A staffing proposal recommended a 60 minute auto-escalation threshold, but that was not adopted.","tags":["support-ticket-auto-escalation-threshold","proposal"],"depth":2,"timestamp":"2026-03-18"},{"id":"e5","text":"The support dashboard shows separate timers for unassigned-ticket escalation and breached SLA response time.","tags":["support-ticket-auto-escalation-threshold","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Diverse state-update case grounded in support operations with realistic competing minute thresholds."}
{"id":"state-hard-010","query":"What is the current canonical field name for traveler cost center?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for traveler cost center is traveler_cost_center.","tags":["traveler-cost-center-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for traveler cost center used to be trip_cost_center.","tags":["traveler-cost-center-field","historical"],"depth":2,"timestamp":"2025-10-21"},{"id":"e3","text":"A legacy expense import still emits cost_center_traveler before normalization.","tags":["traveler-cost-center-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A schema proposal recommended employee_cost_center_travel, but it was rejected as too verbose.","tags":["traveler-cost-center-field","proposal"],"depth":2,"timestamp":"2026-02-16"},{"id":"e5","text":"The expense import normalizer maps older cost-center field names onto the canonical one before validation.","tags":["traveler-cost-center-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Diverse state-update case from travel and finance data modeling with plausible alias drift."}
{"id":"state-hard-011","query":"What is the current review window for unsigned supplier questionnaires?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current review window for unsigned supplier questionnaires is 10 business days.","tags":["unsigned-supplier-questionnaire-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The review window for unsigned supplier questionnaires used to be 5 business days.","tags":["unsigned-supplier-questionnaire-review-window","historical"],"depth":2,"timestamp":"2025-12-05"},{"id":"e3","text":"The expedited vendor lane still uses a 3 business day review window for unsigned supplier questionnaires.","tags":["unsigned-supplier-questionnaire-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A procurement proposal called for a 15 business day review window, but that was never made the default.","tags":["unsigned-supplier-questionnaire-review-window","proposal"],"depth":2,"timestamp":"2026-03-07"},{"id":"e5","text":"The supplier dashboard separates unsigned-questionnaire aging from completed review turnaround.","tags":["unsigned-supplier-questionnaire-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Diverse state-update case from procurement operations with business-day intervals and expedited-lane confusion."}
{"id":"state-hard-012","query":"What is the current default reminder cadence for overdue audit findings?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default reminder cadence for overdue audit findings is every 4 days.","tags":["overdue-audit-finding-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default reminder cadence for overdue audit findings used to be every 2 days.","tags":["overdue-audit-finding-reminders","historical"],"depth":2,"timestamp":"2025-11-03"},{"id":"e3","text":"The high-severity audit lane still sends overdue finding reminders every day.","tags":["overdue-audit-finding-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An audit governance proposal wanted weekly reminders for overdue findings, but that was not adopted.","tags":["overdue-audit-finding-reminders","proposal"],"depth":2,"timestamp":"2026-01-29"},{"id":"e5","text":"The audit console shows reminder cadence separately from escalation cadence for overdue findings.","tags":["overdue-audit-finding-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Diverse state-update case from audit tracking with default-vs-high-severity cadence confusion."}
{"id":"speaker-hard-009","query":"What did you suggest for making support memories realistic without turning them into ticket spam?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using a few believable support queue facts with nearby SLA, escalation, and training-lane distractors instead of dumping dozens of generic ticket snippets.","tags":["realistic-support-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested generating a huge pile of fake ticket titles so the retrieval system would drown in volume.","tags":["realistic-support-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested mixing queue state with policy timing so the wrong answer can sound operationally plausible.","tags":["realistic-support-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later compared small realistic support cases with bulk synthetic ticket noise and found the smaller ones more revealing.","tags":["realistic-support-memories","discussion"],"depth":2},{"id":"e5","text":"Several of the newer state cases now use support thresholds instead of only infrastructure settings.","tags":["realistic-support-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse speaker case focused on support operations and density over synthetic ticket volume."}
{"id":"speaker-hard-010","query":"What did you suggest for travel and reimbursement memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested modeling travel memories around policy changes, warning phases, and exception workflows so reimbursement questions feel like real administrative recall.","tags":["travel-reimbursement-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only airport names and hotel codes because those sound concrete enough.","tags":["travel-reimbursement-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested including business-day windows and receipt thresholds because those are exactly the details people misremember.","tags":["travel-reimbursement-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how administrative memories become hard when policy and practice diverge for a while.","tags":["travel-reimbursement-memories","discussion"],"depth":2},{"id":"e5","text":"The travel receipt case now includes a warning period before enforcement.","tags":["travel-reimbursement-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse speaker case about realistic administrative memory design rather than developer-only scenarios."}
{"id":"speaker-hard-011","query":"What did you suggest for adding diversity without losing the benchmark's identity?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested widening the domains of the memories while keeping the same failure modes so the benchmark stays coherent but less repetitive.","tags":["dataset-diversity","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested changing the schema and labels every few pages so the dataset would feel more varied.","tags":["dataset-diversity","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested reusing the existing adversary types while varying the operational settings, people, and institutions around them.","tags":["dataset-diversity","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later reviewed how domain diversity and adversary consistency can reinforce each other instead of competing.","tags":["dataset-diversity","discussion"],"depth":2},{"id":"e5","text":"This newer batch expands into travel, procurement, audit, support, legal, and warehouse contexts.","tags":["dataset-diversity","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse speaker case aligned with the user's request: more variety without inventing a new benchmark shape."}
{"id":"speaker-hard-012","query":"What did you suggest for realistic state-update cases?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested combining a current default with an old default, a lane-specific override, and a rejected proposal so the wrong answer feels believable from multiple angles.","tags":["realistic-state-updates","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only one outdated value because that would be easier to write in bulk.","tags":["realistic-state-updates","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested choosing values that differ in small, realistic increments like days, minutes, or field-name variants.","tags":["realistic-state-updates","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that realistic state-update cases often look deceptively simple while hiding several plausible wrong answers.","tags":["realistic-state-updates","discussion"],"depth":2},{"id":"e5","text":"Recent state cases now mix defaults, overrides, and declined proposals across several business domains.","tags":["realistic-state-updates","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse speaker case about constructing believable current-vs-stale cases without changing the schema."}
{"id":"paraphrase-hard-009","query":"What is the minimum approval count before a supplier exception can be fast-tracked?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A supplier exception can be fast-tracked once it has 2 approvals.","tags":["supplier-exceptions","approvals","threshold"],"depth":1},{"id":"e2","text":"A supplier exception can be fast-tracked once it has 3 approvals.","tags":["supplier-exceptions","approvals","threshold"],"depth":1},{"id":"e3","text":"Two approvals are enough to move a supplier exception onto the fast-track path.","tags":["supplier-exceptions","approvals","threshold"],"depth":1},{"id":"e4","text":"Two approvals are also enough to close a low-risk travel waiver, but that is a different workflow.","tags":["travel-waivers","approvals"],"depth":2},{"id":"e5","text":"Fast-track eligibility is checked after signature validation, not before it.","tags":["supplier-exceptions","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse paraphrase case from procurement policy with a same-number distractor from travel approvals."}
{"id":"paraphrase-hard-010","query":"How long is the cooling-off period before a terminated vendor can be re-invited?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A terminated vendor has a 30 day cooling-off period before re-invitation.","tags":["terminated-vendors","cooling-off","policy"],"depth":1},{"id":"e2","text":"A terminated vendor has a 14 day cooling-off period before re-invitation.","tags":["terminated-vendors","cooling-off","policy"],"depth":1},{"id":"e3","text":"The policy waits thirty days before a terminated vendor may be invited again.","tags":["terminated-vendors","cooling-off","policy"],"depth":1},{"id":"e4","text":"The contract archive worker also waits 30 days before final purge, but that delay belongs to storage cleanup.","tags":["contract-archive","cooling-off"],"depth":2},{"id":"e5","text":"A compliance override can shorten the cooling-off period only after legal sign-off.","tags":["terminated-vendors","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse paraphrase case from vendor management with same-duration storage cleanup distractor."}
{"id":"paraphrase-hard-011","query":"What confidence is required before a receipt is auto-classified as lodging?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A receipt is auto-classified as lodging only at 0.95 confidence or higher.","tags":["receipt-classification","lodging","confidence"],"depth":1},{"id":"e2","text":"A receipt is auto-classified as lodging only at 0.85 confidence or higher.","tags":["receipt-classification","lodging","confidence"],"depth":1},{"id":"e3","text":"The lodging classifier needs at least 0.95 confidence before it auto-files a receipt.","tags":["receipt-classification","lodging","confidence"],"depth":1},{"id":"e4","text":"A 0.95 score is also enough to auto-classify vendor insurance certificates, but that belongs to a different model.","tags":["certificate-classification","confidence"],"depth":2},{"id":"e5","text":"Below-threshold lodging guesses are surfaced as suggestions rather than committed automatically.","tags":["receipt-classification","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse paraphrase case from travel admin with same-threshold classifier confusion across domains."}
{"id":"paraphrase-hard-012","query":"What is the default reminder delay before an unsigned onboarding form is nudged?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned onboarding form is nudged after a default delay of 72 hours.","tags":["onboarding-forms","reminders","delay"],"depth":1},{"id":"e2","text":"An unsigned onboarding form is nudged after a default delay of 48 hours.","tags":["onboarding-forms","reminders","delay"],"depth":1},{"id":"e3","text":"The default onboarding reminder waits seventy-two hours before nudging an unsigned form.","tags":["onboarding-forms","reminders","delay"],"depth":1},{"id":"e4","text":"Background document export retries also wait 72 hours before their final alert, but that is unrelated to onboarding reminders.","tags":["document-exports","reminders"],"depth":2},{"id":"e5","text":"Managers can trigger an earlier reminder manually when a start date moves forward.","tags":["onboarding-forms","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Diverse paraphrase case from onboarding administration with same-duration alerting in another system."}
{"id":"env-hard-013","query":"What patient messaging gateway does clinic-training use for mock appointment reminders?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The clinic-training environment uses msg-clinic-training.internal for mock appointment reminder traffic.","tags":["clinic-training","appointment-reminders","messaging-gateway","infra"],"depth":1},{"id":"e2","text":"The production environment uses msg-clinic-prod.internal for mock appointment reminder traffic.","tags":["production","appointment-reminders","messaging-gateway","infra"],"depth":1},{"id":"e3","text":"The patient-demo environment uses msg-patient-demo.internal for mock appointment reminder traffic.","tags":["patient-demo","appointment-reminders","messaging-gateway","infra"],"depth":2},{"id":"e4","text":"Clinic-training receives canned replies through inbox-clinic-training.internal, which is not the outbound reminder gateway.","tags":["clinic-training","appointment-reminders","inbound-mail"],"depth":2},{"id":"e5","text":"A nursing worksheet still lists reminders-training.internal from before gateway names were standardized.","tags":["clinic-training","appointment-reminders","historical"],"depth":3,"timestamp":"2025-09-12"},{"id":"e6","text":"The training runbook notes that clinic-training and patient-demo use different send limits even when their reminder templates match.","tags":["clinic-training","appointment-reminders","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from healthcare admin with outbound vs inbound messaging and old training hostname confusion."}
{"id":"env-hard-014","query":"What grading queue does district-sandbox use for standardized essay rescoring?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The district-sandbox environment uses queue-district-sandbox.internal for standardized essay rescoring.","tags":["district-sandbox","essay-rescoring","grading-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-district-prod.internal for standardized essay rescoring.","tags":["production","essay-rescoring","grading-queue","infra"],"depth":1},{"id":"e3","text":"The teacher-preview environment uses queue-teacher-preview.internal for standardized essay rescoring.","tags":["teacher-preview","essay-rescoring","grading-queue","infra"],"depth":2},{"id":"e4","text":"District-sandbox archives failed rescoring jobs in queue-district-sandbox-deadletter.internal, which is not the primary grading queue.","tags":["district-sandbox","essay-rescoring","deadletter"],"depth":2},{"id":"e5","text":"A school pilot note still names district-grading-sandbox.internal from before queue naming was made consistent.","tags":["district-sandbox","essay-rescoring","historical"],"depth":3,"timestamp":"2025-08-24"},{"id":"e6","text":"The rescoring checklist warns that district-sandbox and teacher-preview use parallel fixtures but not the same queue names.","tags":["district-sandbox","essay-rescoring","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from education ops with sandbox vs preview and dead-letter queue confusion."}
{"id":"env-hard-015","query":"What freezer sensor endpoint does store-lab use for overnight spoilage drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The store-lab environment uses sensor-freezer-lab.internal for overnight spoilage drills.","tags":["store-lab","spoilage-drills","freezer-sensor","infra"],"depth":1},{"id":"e2","text":"The production environment uses sensor-freezer-prod.internal for overnight spoilage drills.","tags":["production","spoilage-drills","freezer-sensor","infra"],"depth":1},{"id":"e3","text":"The inventory-preview environment uses sensor-freezer-preview.internal for overnight spoilage drills.","tags":["inventory-preview","spoilage-drills","freezer-sensor","infra"],"depth":2},{"id":"e4","text":"Store-lab sends alarm snapshots to freezer-lab-capture.internal, which is not the live sensor endpoint.","tags":["store-lab","spoilage-drills","captures"],"depth":2},{"id":"e5","text":"A retired facilities note still lists freezer-lab.internal from before sensor hosts were prefixed more clearly.","tags":["store-lab","spoilage-drills","historical"],"depth":3,"timestamp":"2025-07-29"},{"id":"e6","text":"The spoilage drill checklist notes that store-lab and inventory-preview share alert thresholds but not sensor endpoints.","tags":["store-lab","spoilage-drills","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from retail operations with live endpoint vs capture host and inventory-preview confusion."}
{"id":"env-hard-016","query":"What badge printer host does facilities-drill use for evacuation practice?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The facilities-drill environment uses badge-printer-drill.internal for evacuation practice badges.","tags":["facilities-drill","evacuation-practice","badge-printer","infra"],"depth":1},{"id":"e2","text":"The production environment uses badge-printer-prod.internal for evacuation practice badges.","tags":["production","evacuation-practice","badge-printer","infra"],"depth":1},{"id":"e3","text":"The visitor-preview environment uses badge-printer-preview.internal for evacuation practice badges.","tags":["visitor-preview","evacuation-practice","badge-printer","infra"],"depth":2},{"id":"e4","text":"Facilities-drill stores badge templates on badge-templates-drill.internal, which is not the printer host.","tags":["facilities-drill","evacuation-practice","templates"],"depth":2},{"id":"e5","text":"A safety memo still refers to printer-drill-badges.internal from before facilities naming was cleaned up.","tags":["facilities-drill","evacuation-practice","historical"],"depth":3,"timestamp":"2025-10-02"},{"id":"e6","text":"The drill runbook reminds operators that facilities-drill and visitor-preview share layouts but not badge printer hosts.","tags":["facilities-drill","evacuation-practice","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from facilities operations with printer-vs-template confusion and preview overlap."}
{"id":"entity-hard-013","query":"What tea does Anjali drink during patient intake audits?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Anjali drinks ginger tulsi tea during patient intake audits.","tags":["anjali","patient-intake-audits","preference"],"depth":1},{"id":"e2","text":"Anjuli drinks ginger tulsi tea during patient intake audits.","tags":["anjuli","patient-intake-audits","preference"],"depth":1},{"id":"e3","text":"Anjali drinks mint tea during volunteer vaccination clinics instead of intake audits.","tags":["anjali","vaccination-clinics","preference"],"depth":2},{"id":"e4","text":"Anika drinks ginger tulsi tea during records retention reviews, not patient intake audits.","tags":["anika","records-retention","preference"],"depth":2},{"id":"e5","text":"A clinic snack note said Anjali and Anjuli both avoided sugary drinks on long audit days.","tags":["anjali","anjuli","clinic-ops"],"depth":3},{"id":"e6","text":"Anjali says the ginger tea helps her stay warm in the intake records room.","tags":["anjali","patient-intake-audits","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from healthcare admin with near-name overlap and context-specific beverage habits."}
{"id":"entity-hard-014","query":"What bag does Corin carry to substitute teacher orientation?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Corin carries a canvas messenger bag to substitute teacher orientation.","tags":["corin","substitute-teacher-orientation","preference"],"depth":1},{"id":"e2","text":"Corinne carries a canvas messenger bag to substitute teacher orientation.","tags":["corinne","substitute-teacher-orientation","preference"],"depth":1},{"id":"e3","text":"Corin carries a rolling briefcase to district budget meetings instead of orientation sessions.","tags":["corin","budget-meetings","preference"],"depth":2},{"id":"e4","text":"Torin carries a canvas messenger bag to curriculum fairs, not substitute orientation.","tags":["torin","curriculum-fairs","preference"],"depth":2},{"id":"e5","text":"A staff room note said Corin and Corinne both wanted bags with outside folders for orientation packets.","tags":["corin","corinne","staff-room"],"depth":3},{"id":"e6","text":"Corin keeps substitute roster forms in the front sleeve of the messenger bag.","tags":["corin","substitute-teacher-orientation","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from education operations with similar names and overlapping bag preferences."}
{"id":"entity-hard-015","query":"What snack does Paloma bring to overnight inventory resets?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Paloma brings chili-lime popcorn to overnight inventory resets.","tags":["paloma","inventory-resets","preference"],"depth":1},{"id":"e2","text":"Palmer brings chili-lime popcorn to overnight inventory resets.","tags":["palmer","inventory-resets","preference"],"depth":1},{"id":"e3","text":"Paloma brings plain pretzels to daytime merchandising walks instead of inventory resets.","tags":["paloma","merchandising-walks","preference"],"depth":2},{"id":"e4","text":"Pomona brings chili-lime popcorn to warehouse photo shoots, not overnight resets.","tags":["pomona","photo-shoots","preference"],"depth":2},{"id":"e5","text":"A break room note said Paloma and Palmer both preferred salty snacks over sweets on late shifts.","tags":["paloma","palmer","break-room"],"depth":3},{"id":"e6","text":"Paloma says the popcorn keeps her awake without making a mess on scanner tables.","tags":["paloma","inventory-resets","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from retail operations with same snack overlap and task-specific substitutions."}
{"id":"entity-hard-016","query":"What notebook does Joriel use for lease walk-throughs?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Joriel uses a narrow ruled field notebook for lease walk-throughs.","tags":["joriel","lease-walkthroughs","preference"],"depth":1},{"id":"e2","text":"Jorelle uses a narrow ruled field notebook for lease walk-throughs.","tags":["jorelle","lease-walkthroughs","preference"],"depth":1},{"id":"e3","text":"Joriel uses a binder folio during contractor bid reviews instead of the field notebook.","tags":["joriel","contractor-bids","preference"],"depth":2},{"id":"e4","text":"Liora uses a narrow ruled field notebook during parking inspections, not lease walk-throughs.","tags":["liora","parking-inspections","preference"],"depth":2},{"id":"e5","text":"A facilities supply order once swapped Joriel and Jorelle on the field notebook line item before admin corrected it.","tags":["joriel","jorelle","historical"],"depth":3,"timestamp":"2026-02-09"},{"id":"e6","text":"Joriel keeps spare unit keys in the elastic band of the field notebook during walk-throughs.","tags":["joriel","lease-walkthroughs","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from real estate and facilities with clerical name confusion and context-specific supplies."}
{"id":"time-hard-013","query":"When did we require supervisor review on disputed clinic copay waivers?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require supervisor review on disputed clinic copay waivers in May.","tags":["timeline","clinic-copay-waivers","supervisor-review"],"depth":1,"timestamp":"2026-05-18"},{"id":"e2","text":"We proposed supervisor review on disputed clinic copay waivers in March.","tags":["planning","clinic-copay-waivers","supervisor-review"],"depth":2,"timestamp":"2026-03-27"},{"id":"e3","text":"A billing warning banner for disputed copay waivers appeared in April before supervisor review became mandatory.","tags":["trial","clinic-copay-waivers","supervisor-review"],"depth":2,"timestamp":"2026-04-14"},{"id":"e4","text":"The patient billing guide was updated in June after supervisor review became required.","tags":["docs","clinic-copay-waivers","supervisor-review"],"depth":2,"timestamp":"2026-06-04"},{"id":"e5","text":"The old nurse-only exception path was removed in July.","tags":["cleanup","clinic-copay-waivers","supervisor-review"],"depth":3,"timestamp":"2026-07-08"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from healthcare administration with warning-before-enforcement and retired exception path."}
{"id":"time-hard-014","query":"When did we stop auto-posting district attendance corrections?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-posting district attendance corrections in November.","tags":["timeline","attendance-corrections","auto-posting"],"depth":1,"timestamp":"2026-11-13"},{"id":"e2","text":"We discussed stopping district attendance auto-posting in September.","tags":["planning","attendance-corrections","auto-posting"],"depth":2,"timestamp":"2026-09-21"},{"id":"e3","text":"A one-school pilot disabled auto-posting in October, but only for substitute-marked absences.","tags":["trial","attendance-corrections","auto-posting"],"depth":2,"timestamp":"2026-10-06"},{"id":"e4","text":"The district handbook was updated in December to describe manual attendance correction review.","tags":["docs","attendance-corrections","auto-posting"],"depth":2,"timestamp":"2026-12-02"},{"id":"e5","text":"A cleanup in January removed the old auto-posting dashboard tiles.","tags":["cleanup","attendance-corrections","auto-posting"],"depth":3,"timestamp":"2027-01-10"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from education administration with a partial pilot preceding the full policy change."}
{"id":"time-hard-015","query":"When did we require dual sign-off on overnight cash drops?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require dual sign-off on overnight cash drops in February.","tags":["timeline","overnight-cash-drops","dual-signoff"],"depth":1,"timestamp":"2026-02-22"},{"id":"e2","text":"We proposed dual sign-off on overnight cash drops in December.","tags":["planning","overnight-cash-drops","dual-signoff"],"depth":2,"timestamp":"2025-12-14"},{"id":"e3","text":"A pilot store used dual sign-off in January before the rule became chain-wide.","tags":["trial","overnight-cash-drops","dual-signoff"],"depth":2,"timestamp":"2026-01-17"},{"id":"e4","text":"The retail operations handbook was updated in March after dual sign-off became mandatory.","tags":["docs","overnight-cash-drops","dual-signoff"],"depth":2,"timestamp":"2026-03-03"},{"id":"e5","text":"The old single-clerk exception toggle was removed in April.","tags":["cleanup","overnight-cash-drops","dual-signoff"],"depth":3,"timestamp":"2026-04-07"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from retail cash handling with store pilot vs chain-wide enforcement confusion."}
{"id":"time-hard-016","query":"When did we rename the incident binder to the building event log?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the incident binder to the building event log in August.","tags":["timeline","incident-binder","building-event-log"],"depth":1,"timestamp":"2026-08-20"},{"id":"e2","text":"We proposed renaming the incident binder to the building event log in June.","tags":["planning","incident-binder","building-event-log"],"depth":2,"timestamp":"2026-06-29"},{"id":"e3","text":"A facilities draft began using building event log in July before the rename was official.","tags":["trial","incident-binder","building-event-log"],"depth":2,"timestamp":"2026-07-11"},{"id":"e4","text":"The safety glossary removed the old incident binder term in September.","tags":["docs","incident-binder","building-event-log"],"depth":2,"timestamp":"2026-09-05"},{"id":"e5","text":"The old binder label was removed from inspection stickers in October.","tags":["cleanup","incident-binder","building-event-log"],"depth":3,"timestamp":"2026-10-02"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from facilities terminology with draft usage preceding the official rename."}
{"id":"state-hard-013","query":"What is the current timeout before an unanswered patient portal message escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unanswered patient portal message escalates is 18 hours.","tags":["patient-portal-message-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The escalation timeout for unanswered patient portal messages used to be 12 hours.","tags":["patient-portal-message-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-28"},{"id":"e3","text":"The training lane still uses a 6 hour escalation timeout for unanswered patient portal messages.","tags":["patient-portal-message-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A patient access proposal recommended a 24 hour escalation timeout, but that was not adopted.","tags":["patient-portal-message-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-12"},{"id":"e5","text":"The portal dashboard tracks unanswered-message escalation separately from callback overdue timers.","tags":["patient-portal-message-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from healthcare messaging with realistic hour-level alternatives and lane-specific override."}
{"id":"state-hard-014","query":"What is the current canonical field name for student guardian language preference?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for student guardian language preference is guardian_language_preference.","tags":["guardian-language-preference-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for student guardian language preference used to be family_language_pref.","tags":["guardian-language-preference-field","historical"],"depth":2,"timestamp":"2025-10-18"},{"id":"e3","text":"A legacy import still emits guardian_pref_language before normalization.","tags":["guardian-language-preference-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A data governance proposal recommended guardian_home_language, but it was rejected as too ambiguous.","tags":["guardian-language-preference-field","proposal"],"depth":2,"timestamp":"2026-02-02"},{"id":"e5","text":"The student import pipeline maps older guardian language fields onto the canonical name before validation.","tags":["guardian-language-preference-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from education data modeling with plausible alias drift and canonical naming pressure."}
{"id":"state-hard-015","query":"What is the current default recount window for overnight shelf audits?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default recount window for overnight shelf audits is 6 hours.","tags":["shelf-audit-recount-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default recount window for overnight shelf audits used to be 4 hours.","tags":["shelf-audit-recount-window","historical"],"depth":2,"timestamp":"2025-12-08"},{"id":"e3","text":"The high-shrink pilot lane still uses a 2 hour recount window for overnight shelf audits.","tags":["shelf-audit-recount-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A store ops proposal asked for an 8 hour recount window, but that was not adopted.","tags":["shelf-audit-recount-window","proposal"],"depth":2,"timestamp":"2026-03-08"},{"id":"e5","text":"The inventory console tracks recount windows separately from final shrink investigation deadlines.","tags":["shelf-audit-recount-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from retail inventory with small realistic hour differences and pilot-lane confusion."}
{"id":"state-hard-016","query":"What is the current reminder cadence for open elevator inspection findings?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for open elevator inspection findings is every 5 days.","tags":["elevator-inspection-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for open elevator inspection findings used to be every 3 days.","tags":["elevator-inspection-reminders","historical"],"depth":2,"timestamp":"2025-11-06"},{"id":"e3","text":"The urgent-safety lane still reminds on open elevator inspection findings every day.","tags":["elevator-inspection-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A facilities governance proposal recommended weekly reminders, but that was not adopted as the default.","tags":["elevator-inspection-reminders","proposal"],"depth":2,"timestamp":"2026-01-21"},{"id":"e5","text":"The building dashboard shows reminder cadence separately from regulatory due dates for each elevator finding.","tags":["elevator-inspection-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from facilities compliance with default-vs-urgent-lane reminder differences."}
{"id":"speaker-hard-013","query":"What did you suggest for making healthcare admin memories realistic?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using policy timing, supervisor review rules, and training-lane exceptions so healthcare admin memories feel like real operational recall instead of generic hospital jargon.","tags":["healthcare-admin-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just scattering medical words around otherwise generic cases so they would seem realistic enough.","tags":["healthcare-admin-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested focusing on appointment reminders, copay waivers, portal messages, and intake audits because those create concrete but easily confusable facts.","tags":["healthcare-admin-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how administrative healthcare memory often fails on timing and approval details rather than medical terminology.","tags":["healthcare-admin-memories","discussion"],"depth":2},{"id":"e5","text":"This broader batch adds clinic-training, patient portal, and copay waiver examples.","tags":["healthcare-admin-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case for healthcare admin realism rather than software-only operations."}
{"id":"speaker-hard-014","query":"What did you suggest for education and training memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested centering education memories on attendance corrections, substitute orientation, grading rescoring, and guardian data so the facts feel administratively real.","tags":["education-training-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only classroom object lists because school settings would already imply realism.","tags":["education-training-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using district, teacher-preview, and pilot-lane distinctions because education systems often have those layered operational contexts.","tags":["education-training-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that education memories become hard when district-wide policy and one-school pilots overlap.","tags":["education-training-memories","discussion"],"depth":2},{"id":"e5","text":"This batch now includes attendance, rescoring, orientation, and guardian-field examples.","tags":["education-training-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case focused on education administration and training workflows."}
{"id":"speaker-hard-015","query":"What did you suggest for retail and facilities memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using overnight resets, cash drops, spoilage drills, badge printing, inspections, and lease walk-throughs so retail and facilities memories feel grounded in routine operations.","tags":["retail-facilities-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using generic store and building nouns without operational detail because the setting alone would carry the realism.","tags":["retail-facilities-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested leaning on adjacent systems like dead-letter queues, template hosts, and urgent-safety lanes so the wrong answer has a believable neighbor.","tags":["retail-facilities-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that store and facilities memories get hard when the distractor is operationally related rather than random.","tags":["retail-facilities-memories","discussion"],"depth":2},{"id":"e5","text":"The newer retail and facilities cases intentionally include neighboring but non-primary systems.","tags":["retail-facilities-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case about realistic neighboring systems in retail and facilities workflows."}
{"id":"speaker-hard-016","query":"What did you suggest for compliance and investigations memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using reminder cadences, review windows, disputed exceptions, and termination cooldowns so compliance memories revolve around the kinds of details people actually misremember.","tags":["compliance-investigations-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested making compliance memories mostly about intimidating terminology because that would sound official enough.","tags":["compliance-investigations-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested mixing business-day windows, reminder cadences, and override paths because those create several plausible but wrong answers in a compact case.","tags":["compliance-investigations-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how compliance retrieval usually fails on small policy numbers, not on the broad topic label.","tags":["compliance-investigations-memories","discussion"],"depth":2},{"id":"e5","text":"The new procurement, audit, and vendor cases follow that pattern of small but consequential policy details.","tags":["compliance-investigations-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case for compliance and investigation style memories with number-heavy realistic details."}
{"id":"paraphrase-hard-013","query":"What score is required before a clinic message is auto-routed to billing?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A clinic message is auto-routed to billing only at 0.91 confidence or higher.","tags":["clinic-messages","billing-routing","confidence"],"depth":1},{"id":"e2","text":"A clinic message is auto-routed to billing only at 0.81 confidence or higher.","tags":["clinic-messages","billing-routing","confidence"],"depth":1},{"id":"e3","text":"The billing router needs at least 0.91 confidence before it auto-sends a clinic message.","tags":["clinic-messages","billing-routing","confidence"],"depth":1},{"id":"e4","text":"A 0.91 score is also enough to auto-route school transport requests, but that belongs to a different classifier.","tags":["transport-requests","confidence"],"depth":2},{"id":"e5","text":"Lower-confidence clinic billing guesses are retained as suggestions rather than committed automatically.","tags":["clinic-messages","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from healthcare routing with same-threshold distractor from education transport."}
{"id":"paraphrase-hard-014","query":"How many missing forms trigger a substitute onboarding escalation?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A substitute onboarding escalation is triggered once 4 forms are missing.","tags":["substitute-onboarding","missing-forms","threshold"],"depth":1},{"id":"e2","text":"A substitute onboarding escalation is triggered once 3 forms are missing.","tags":["substitute-onboarding","missing-forms","threshold"],"depth":1},{"id":"e3","text":"Four missing forms are enough to open a substitute onboarding escalation.","tags":["substitute-onboarding","missing-forms","threshold"],"depth":1},{"id":"e4","text":"Four missing lease photos are enough to open a property handoff review, but that is a different workflow.","tags":["property-handoffs","threshold"],"depth":2},{"id":"e5","text":"The substitute onboarding escalation is checked after identity verification, not before it.","tags":["substitute-onboarding","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from education onboarding with same-number review threshold in property operations."}
{"id":"paraphrase-hard-015","query":"What is the default hold time before a spoilage alert is escalated to a manager?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A spoilage alert is escalated to a manager after a default hold time of 20 minutes.","tags":["spoilage-alerts","manager-escalation","delay"],"depth":1},{"id":"e2","text":"A spoilage alert is escalated to a manager after a default hold time of 10 minutes.","tags":["spoilage-alerts","manager-escalation","delay"],"depth":1},{"id":"e3","text":"The default spoilage workflow waits twenty minutes before escalating an alert to a manager.","tags":["spoilage-alerts","manager-escalation","delay"],"depth":1},{"id":"e4","text":"The freezer sensor self-test also waits 20 minutes before its second warning, but that timer is unrelated to managerial escalation.","tags":["freezer-sensors","delay"],"depth":2},{"id":"e5","text":"An overnight lead can override the hold time manually during known maintenance windows.","tags":["spoilage-alerts","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from retail operations with same-duration sensor-warning distractor."}
{"id":"paraphrase-hard-016","query":"What is the default waiting period before a lease inspection is marked missed?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A lease inspection is marked missed after a default waiting period of 15 minutes.","tags":["lease-inspections","missed-appointments","delay"],"depth":1},{"id":"e2","text":"A lease inspection is marked missed after a default waiting period of 30 minutes.","tags":["lease-inspections","missed-appointments","delay"],"depth":1},{"id":"e3","text":"The default lease inspection workflow waits fifteen minutes before marking the appointment missed.","tags":["lease-inspections","missed-appointments","delay"],"depth":1},{"id":"e4","text":"Visitor badge pickup also uses a 15 minute hold before cancellation, but that belongs to facilities reception.","tags":["visitor-badges","delay"],"depth":2},{"id":"e5","text":"An agent can extend the lease inspection waiting period manually when a tenant confirms they are parking.","tags":["lease-inspections","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from real estate and facilities with same-duration cancellation behavior in a nearby workflow."}
{"id":"env-hard-017","query":"What claims image store does adjuster-training use for hail damage mock reviews?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The adjuster-training environment uses images-adjuster-training.internal for hail damage mock review images.","tags":["adjuster-training","hail-damage-reviews","claims-image-store","infra"],"depth":1},{"id":"e2","text":"The production environment uses images-claims-prod.internal for hail damage mock review images.","tags":["production","hail-damage-reviews","claims-image-store","infra"],"depth":1},{"id":"e3","text":"The fraud-preview environment uses images-fraud-preview.internal for hail damage mock review images.","tags":["fraud-preview","hail-damage-reviews","claims-image-store","infra"],"depth":2},{"id":"e4","text":"Adjuster-training archives rejected photos to images-adjuster-training-archive.internal, which is not the primary review store.","tags":["adjuster-training","hail-damage-reviews","archive"],"depth":2},{"id":"e5","text":"A claims handbook still lists photos-adjuster-training.internal from before image store hosts were renamed.","tags":["adjuster-training","hail-damage-reviews","historical"],"depth":3,"timestamp":"2025-09-05"},{"id":"e6","text":"The training checklist notes that adjuster-training and fraud-preview use similar sample claims but different image stores.","tags":["adjuster-training","hail-damage-reviews","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector environment case from insurance claims with archive host and fraud-preview distractors."}
{"id":"env-hard-018","query":"What room booking host does event-rehearsal use for banquet seating tests?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The event-rehearsal environment uses booking-event-rehearsal.internal for banquet seating tests.","tags":["event-rehearsal","banquet-seating-tests","room-booking","infra"],"depth":1},{"id":"e2","text":"The production environment uses booking-hospitality-prod.internal for banquet seating tests.","tags":["production","banquet-seating-tests","room-booking","infra"],"depth":1},{"id":"e3","text":"The guest-preview environment uses booking-guest-preview.internal for banquet seating tests.","tags":["guest-preview","banquet-seating-tests","room-booking","infra"],"depth":2},{"id":"e4","text":"Event-rehearsal prints seating cards through cards-event-rehearsal.internal, which is not the room booking host.","tags":["event-rehearsal","banquet-seating-tests","printing"],"depth":2},{"id":"e5","text":"A catering worksheet still mentions rehearsal-booking.internal from before the hospitality naming cleanup.","tags":["event-rehearsal","banquet-seating-tests","historical"],"depth":3,"timestamp":"2025-10-18"},{"id":"e6","text":"The rehearsal runbook notes that event-rehearsal and guest-preview share guest lists but not room booking hosts.","tags":["event-rehearsal","banquet-seating-tests","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector environment case from hospitality operations with adjacent printing system and preview overlap."}
{"id":"env-hard-019","query":"What calibration queue does plant-sandbox use for torque wrench verification?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The plant-sandbox environment uses queue-plant-sandbox.internal for torque wrench verification jobs.","tags":["plant-sandbox","torque-wrench-verification","calibration-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-plant-prod.internal for torque wrench verification jobs.","tags":["production","torque-wrench-verification","calibration-queue","infra"],"depth":1},{"id":"e3","text":"The qa-preview environment uses queue-qa-preview.internal for torque wrench verification jobs.","tags":["qa-preview","torque-wrench-verification","calibration-queue","infra"],"depth":2},{"id":"e4","text":"Plant-sandbox routes failed verification jobs into queue-plant-sandbox-deadletter.internal, which is not the primary calibration queue.","tags":["plant-sandbox","torque-wrench-verification","deadletter"],"depth":2},{"id":"e5","text":"An old line setup note still lists torque-queue-sandbox.internal from before calibration queues were renamed.","tags":["plant-sandbox","torque-wrench-verification","historical"],"depth":3,"timestamp":"2025-08-08"},{"id":"e6","text":"The calibration runbook notes that plant-sandbox and qa-preview share sample fixtures but not queue names.","tags":["plant-sandbox","torque-wrench-verification","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector environment case from manufacturing quality control with dead-letter and renamed queue confusion."}
{"id":"env-hard-020","query":"What permit archive host does zoning-preview use for resubmission packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The zoning-preview environment uses archive-zoning-preview.internal for permit resubmission packets.","tags":["zoning-preview","permit-resubmissions","archive-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses archive-permits-prod.internal for permit resubmission packets.","tags":["production","permit-resubmissions","archive-host","infra"],"depth":1},{"id":"e3","text":"The hearings-sandbox environment uses archive-hearings-sandbox.internal for permit resubmission packets.","tags":["hearings-sandbox","permit-resubmissions","archive-host","infra"],"depth":2},{"id":"e4","text":"Zoning-preview publishes reviewer PDFs through pdf-zoning-preview.internal, which is not the archive host.","tags":["zoning-preview","permit-resubmissions","pdfs"],"depth":2},{"id":"e5","text":"A permit office memo still references permits-preview-archive.internal from before the archive hosts were standardized.","tags":["zoning-preview","permit-resubmissions","historical"],"depth":3,"timestamp":"2025-09-30"},{"id":"e6","text":"The resubmission checklist notes that zoning-preview and hearings-sandbox share packet templates but not archive hosts.","tags":["zoning-preview","permit-resubmissions","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector environment case from public sector permitting with archive-vs-PDF host confusion."}
{"id":"entity-hard-017","query":"What snack does Sabria bring to workers' comp intake reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Sabria brings roasted chickpeas to workers' comp intake reviews.","tags":["sabria","workers-comp-intake","preference"],"depth":1},{"id":"e2","text":"Sabriya brings roasted chickpeas to workers' comp intake reviews.","tags":["sabriya","workers-comp-intake","preference"],"depth":1},{"id":"e3","text":"Sabria brings apple slices to policy renewal meetings instead of intake reviews.","tags":["sabria","policy-renewals","preference"],"depth":2},{"id":"e4","text":"Sabrina brings roasted chickpeas to benefits orientation, not workers' comp intake.","tags":["sabrina","benefits-orientation","preference"],"depth":2},{"id":"e5","text":"A claims break-room note said Sabria and Sabriya both preferred dry snacks that did not smudge paper files.","tags":["sabria","sabriya","break-room"],"depth":3},{"id":"e6","text":"Sabria keeps the roasted chickpeas in the side pocket of her intake binder.","tags":["sabria","workers-comp-intake","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector entity case from insurance admin with near-name overlap and same-snack distractors."}
{"id":"entity-hard-018","query":"What shoes does Elowen wear during ballroom setup inspections?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Elowen wears black soft-soled flats during ballroom setup inspections.","tags":["elowen","ballroom-setup-inspections","preference"],"depth":1},{"id":"e2","text":"Elowyn wears black soft-soled flats during ballroom setup inspections.","tags":["elowyn","ballroom-setup-inspections","preference"],"depth":1},{"id":"e3","text":"Elowen wears leather ankle boots during loading dock checks instead of ballroom inspections.","tags":["elowen","loading-dock-checks","preference"],"depth":2},{"id":"e4","text":"Marwen wears black soft-soled flats during stage rehearsals, not ballroom setup inspections.","tags":["marwen","stage-rehearsals","preference"],"depth":2},{"id":"e5","text":"A banquet checklist once swapped Elowen and Elowyn on the footwear reimbursement form before it was fixed.","tags":["elowen","elowyn","historical"],"depth":3,"timestamp":"2026-02-01"},{"id":"e6","text":"Elowen says the soft-soled flats keep the ballroom floor quiet during pre-event inspections.","tags":["elowen","ballroom-setup-inspections","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector entity case from hospitality operations with near-name overlap and context-specific footwear choices."}
{"id":"entity-hard-019","query":"What mug does Tarek carry during final line quality reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Tarek carries a stainless travel mug during final line quality reviews.","tags":["tarek","line-quality-reviews","preference"],"depth":1},{"id":"e2","text":"Tarik carries a stainless travel mug during final line quality reviews.","tags":["tarik","line-quality-reviews","preference"],"depth":1},{"id":"e3","text":"Tarek uses a ceramic cup during daytime vendor demos instead of the travel mug.","tags":["tarek","vendor-demos","preference"],"depth":2},{"id":"e4","text":"Marek carries a stainless travel mug during forklift certification mornings, not final line quality reviews.","tags":["marek","forklift-certification","preference"],"depth":2},{"id":"e5","text":"A plant kitchen note said Tarek and Tarik both preferred mugs with locking lids on the production floor.","tags":["tarek","tarik","kitchen-note"],"depth":3},{"id":"e6","text":"Tarek clips the travel mug to the rail by the final inspection table.","tags":["tarek","line-quality-reviews","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector entity case from manufacturing with transliterated-name confusion and safety-specific mug choices."}
{"id":"entity-hard-020","query":"What folder does Narelle bring to zoning appeal hearings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Narelle brings a red accordion folder to zoning appeal hearings.","tags":["narelle","zoning-appeals","preference"],"depth":1},{"id":"e2","text":"Narel brings a red accordion folder to zoning appeal hearings.","tags":["narel","zoning-appeals","preference"],"depth":1},{"id":"e3","text":"Narelle brings a slim blue folio to permit counter reviews instead of zoning hearings.","tags":["narelle","permit-counter","preference"],"depth":2},{"id":"e4","text":"Marielle brings a red accordion folder to tenant mediation sessions, not zoning appeal hearings.","tags":["marielle","tenant-mediation","preference"],"depth":2},{"id":"e5","text":"A city clerk note said Narelle and Narel both wanted folders with enough pockets for oversized maps.","tags":["narelle","narel","clerk-note"],"depth":3},{"id":"e6","text":"Narelle keeps hearing exhibits in the back section of the red accordion folder.","tags":["narelle","zoning-appeals","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-sector entity case from public sector hearings with near-name overlap and hearing-vs-counter context changes."}
{"id":"time-hard-017","query":"When did we require manager concurrence on employee relations note closures?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require manager concurrence on employee relations note closures in July.","tags":["timeline","employee-relations-notes","manager-concurrence"],"depth":1,"timestamp":"2026-07-19"},{"id":"e2","text":"We proposed manager concurrence on employee relations note closures in May.","tags":["planning","employee-relations-notes","manager-concurrence"],"depth":2,"timestamp":"2026-05-24"},{"id":"e3","text":"A limited HR pilot required manager concurrence in June for attendance-related note closures only.","tags":["trial","employee-relations-notes","manager-concurrence"],"depth":2,"timestamp":"2026-06-09"},{"id":"e4","text":"The HR guide was updated in August to describe manager concurrence for note closure.","tags":["docs","employee-relations-notes","manager-concurrence"],"depth":2,"timestamp":"2026-08-03"},{"id":"e5","text":"The single-reviewer closure toggle was removed in September.","tags":["cleanup","employee-relations-notes","manager-concurrence"],"depth":3,"timestamp":"2026-09-01"}],"relevant_ids":["e1"],"notes":"Broad-sector time case from HR investigations with a partial pilot preceding the organization-wide policy change."}
{"id":"time-hard-018","query":"When did we rename the guest issue log to the stay resolution ledger?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the guest issue log to the stay resolution ledger in April.","tags":["timeline","guest-issue-log","stay-resolution-ledger"],"depth":1,"timestamp":"2026-04-16"},{"id":"e2","text":"We discussed renaming the guest issue log to the stay resolution ledger in February.","tags":["planning","guest-issue-log","stay-resolution-ledger"],"depth":2,"timestamp":"2026-02-22"},{"id":"e3","text":"An operations draft began using stay resolution ledger in March before the rename was official.","tags":["trial","guest-issue-log","stay-resolution-ledger"],"depth":2,"timestamp":"2026-03-10"},{"id":"e4","text":"The front desk glossary removed the guest issue log term in May.","tags":["docs","guest-issue-log","stay-resolution-ledger"],"depth":2,"timestamp":"2026-05-04"},{"id":"e5","text":"The old guest issue log label disappeared from the dashboard tiles in June.","tags":["cleanup","guest-issue-log","stay-resolution-ledger"],"depth":3,"timestamp":"2026-06-07"}],"relevant_ids":["e1"],"notes":"Broad-sector time case from hospitality operations with draft usage preceding the actual rename."}
{"id":"time-hard-019","query":"When did we stop auto-closing torque variance findings?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-closing torque variance findings in October.","tags":["timeline","torque-variance-findings","auto-closing"],"depth":1,"timestamp":"2026-10-14"},{"id":"e2","text":"We proposed stopping auto-closing torque variance findings in August.","tags":["planning","torque-variance-findings","auto-closing"],"depth":2,"timestamp":"2026-08-27"},{"id":"e3","text":"One production cell stopped auto-closing torque variance findings in September, but only during a quality pilot.","tags":["trial","torque-variance-findings","auto-closing"],"depth":2,"timestamp":"2026-09-18"},{"id":"e4","text":"The plant quality handbook was updated in November to reflect manual closure review.","tags":["docs","torque-variance-findings","auto-closing"],"depth":2,"timestamp":"2026-11-05"},{"id":"e5","text":"A metrics cleanup removed the old auto-close chart in December.","tags":["cleanup","torque-variance-findings","auto-closing"],"depth":3,"timestamp":"2026-12-06"}],"relevant_ids":["e1"],"notes":"Broad-sector time case from manufacturing quality control with pilot-vs-final policy confusion."}
{"id":"time-hard-020","query":"When did we require secondary reviewer sign-off on expedited permit denials?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require secondary reviewer sign-off on expedited permit denials in January.","tags":["timeline","expedited-permit-denials","secondary-signoff"],"depth":1,"timestamp":"2026-01-15"},{"id":"e2","text":"We proposed secondary reviewer sign-off on expedited permit denials in November.","tags":["planning","expedited-permit-denials","secondary-signoff"],"depth":2,"timestamp":"2025-11-23"},{"id":"e3","text":"A hearings pilot added secondary sign-off in December for demolition permits only.","tags":["trial","expedited-permit-denials","secondary-signoff"],"depth":2,"timestamp":"2025-12-19"},{"id":"e4","text":"The permit office handbook was updated in February after sign-off became mandatory.","tags":["docs","expedited-permit-denials","secondary-signoff"],"depth":2,"timestamp":"2026-02-06"},{"id":"e5","text":"The old single-reviewer expedited path was removed in March.","tags":["cleanup","expedited-permit-denials","secondary-signoff"],"depth":3,"timestamp":"2026-03-09"}],"relevant_ids":["e1"],"notes":"Broad-sector time case from public sector permitting with a permit-type pilot that competes with the actual rollout date."}
{"id":"state-hard-017","query":"What is the current default reserve review window for suspicious auto claims?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default reserve review window for suspicious auto claims is 7 business days.","tags":["suspicious-auto-claims-reserve-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reserve review window for suspicious auto claims used to be 5 business days.","tags":["suspicious-auto-claims-reserve-review-window","historical"],"depth":2,"timestamp":"2025-11-12"},{"id":"e3","text":"The expedited fraud lane still uses a 2 business day reserve review window.","tags":["suspicious-auto-claims-reserve-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A claims governance proposal recommended a 10 business day reserve review window, but that was not adopted.","tags":["suspicious-auto-claims-reserve-review-window","proposal"],"depth":2,"timestamp":"2026-03-11"},{"id":"e5","text":"The claims dashboard tracks reserve review windows separately from payout approval deadlines.","tags":["suspicious-auto-claims-reserve-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-sector state case from insurance claims with business-day timing and fraud-lane override confusion."}
{"id":"state-hard-018","query":"What is the current canonical field name for banquet guarantee cutoff?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for banquet guarantee cutoff is guarantee_cutoff_at.","tags":["banquet-guarantee-cutoff-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for banquet guarantee cutoff used to be banquet_guarantee_deadline.","tags":["banquet-guarantee-cutoff-field","historical"],"depth":2,"timestamp":"2025-10-25"},{"id":"e3","text":"A legacy events import still emits guarantee_deadline_ts before normalization.","tags":["banquet-guarantee-cutoff-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An events schema proposal recommended banquet_cutoff_time, but it was rejected as too vague.","tags":["banquet-guarantee-cutoff-field","proposal"],"depth":2,"timestamp":"2026-01-28"},{"id":"e5","text":"The events import normalizer maps older banquet guarantee fields onto the canonical one before validation.","tags":["banquet-guarantee-cutoff-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-sector state case from hospitality data modeling with plausible legacy field variants."}
{"id":"state-hard-019","query":"What is the current cooldown before a failed calibration batch is requeued?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current cooldown before a failed calibration batch is requeued is 90 minutes.","tags":["failed-calibration-batch-cooldown","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The cooldown before a failed calibration batch is requeued used to be 45 minutes.","tags":["failed-calibration-batch-cooldown","historical"],"depth":2,"timestamp":"2025-12-02"},{"id":"e3","text":"The high-priority quality lane still uses a 15 minute cooldown before a failed calibration batch is requeued.","tags":["failed-calibration-batch-cooldown","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A plant ops proposal recommended a 2 hour cooldown, but that was not adopted as the default.","tags":["failed-calibration-batch-cooldown","proposal"],"depth":2,"timestamp":"2026-03-03"},{"id":"e5","text":"The calibration console tracks requeue cooldown separately from final failure escalation thresholds.","tags":["failed-calibration-batch-cooldown","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-sector state case from manufacturing operations with nearby interval choices and a high-priority lane override."}
{"id":"state-hard-020","query":"What is the current reminder cadence for unresolved employee relations follow-ups?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unresolved employee relations follow-ups is every 6 days.","tags":["employee-relations-followup-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unresolved employee relations follow-ups used to be every 3 days.","tags":["employee-relations-followup-reminders","historical"],"depth":2,"timestamp":"2025-11-15"},{"id":"e3","text":"The urgent conduct lane still sends unresolved employee relations follow-up reminders every day.","tags":["employee-relations-followup-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An HR governance proposal wanted weekly follow-up reminders, but that was not adopted as the default.","tags":["employee-relations-followup-reminders","proposal"],"depth":2,"timestamp":"2026-02-18"},{"id":"e5","text":"The HR dashboard separates unresolved follow-up reminders from case-aging escalations.","tags":["employee-relations-followup-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-sector state case from HR employee relations with default-vs-urgent-lane cadence confusion."}
{"id":"speaker-hard-017","query":"What did you suggest for insurance and claims memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested centering insurance memories on reserve windows, intake reviews, fraud lanes, and image handling so the difficulty comes from realistic claims operations rather than abstract finance jargon.","tags":["insurance-claims-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested filling insurance cases with policy buzzwords because domain language alone would make them feel specialized.","tags":["insurance-claims-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using business-day windows and lane-specific overrides because those are the details claims teams routinely misremember.","tags":["insurance-claims-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how claims memories become hard when archive, fraud, and training systems all look adjacent to the right answer.","tags":["insurance-claims-memories","discussion"],"depth":2},{"id":"e5","text":"This batch now adds claims image stores, reserve review windows, and intake-review routines.","tags":["insurance-claims-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector speaker case for insurance and claims realism rather than software-only ops patterns."}
{"id":"speaker-hard-018","query":"What did you suggest for hospitality and events memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using guest issue logs, seating tests, banquet guarantees, and rehearsal environments so hospitality memories revolve around concrete coordination details.","tags":["hospitality-events-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using room names and menu items only, because events already sound specific enough.","tags":["hospitality-events-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested pairing booking hosts with neighboring printing or guest-preview systems so distractors feel operationally believable.","tags":["hospitality-events-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later observed that hospitality memory errors often come from adjacent planning systems rather than from the main booking fact itself.","tags":["hospitality-events-memories","discussion"],"depth":2},{"id":"e5","text":"The new event-rehearsal and banquet field cases follow that pattern of neighboring-system confusion.","tags":["hospitality-events-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector speaker case focused on hospitality coordination realism and adjacent-system distractors."}
{"id":"speaker-hard-019","query":"What did you suggest for manufacturing and quality control memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using calibration queues, torque variance findings, line quality reviews, and requeue cooldowns so manufacturing memories hinge on operationally precise details.","tags":["manufacturing-quality-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only tool names and part numbers because factory language already sounds technical.","tags":["manufacturing-quality-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping pilot-lane and dead-letter distractors close to the real system so wrong answers stay plausible.","tags":["manufacturing-quality-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that manufacturing retrieval gets hardest when a trial cell or high-priority lane almost matches the production answer.","tags":["manufacturing-quality-memories","discussion"],"depth":2},{"id":"e5","text":"This wave adds plant-sandbox, torque variance, and calibration cooldown cases built that way.","tags":["manufacturing-quality-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector speaker case about realistic manufacturing/quality retrieval pressure from near-production neighbors."}
{"id":"speaker-hard-020","query":"What did you suggest for public sector and HR memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using permit denials, zoning appeals, employee relations follow-ups, and closure sign-offs because public sector and HR memories often fail on small procedural details.","tags":["public-sector-hr-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested relying on official-sounding labels and committee names because that would already make the cases feel bureaucratic.","tags":["public-sector-hr-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using review windows, reminder cadences, and exception paths because those produce believable wrong answers without bloating each case.","tags":["public-sector-hr-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how both public sector and HR memories are hard for the same reason: lots of similar procedures with slightly different numbers and sign-off rules.","tags":["public-sector-hr-memories","discussion"],"depth":2},{"id":"e5","text":"The new permit and employee-relations cases lean on those small procedural differences.","tags":["public-sector-hr-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector speaker case linking public sector and HR through realistic procedural memory failure modes."}
{"id":"paraphrase-hard-017","query":"What confidence is required before a hail damage photo is auto-tagged as windshield-only?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A hail damage photo is auto-tagged as windshield-only only at 0.93 confidence or higher.","tags":["hail-damage-photos","windshield-only","confidence"],"depth":1},{"id":"e2","text":"A hail damage photo is auto-tagged as windshield-only only at 0.83 confidence or higher.","tags":["hail-damage-photos","windshield-only","confidence"],"depth":1},{"id":"e3","text":"The windshield-only tagger needs at least 0.93 confidence before it auto-labels a hail damage photo.","tags":["hail-damage-photos","windshield-only","confidence"],"depth":1},{"id":"e4","text":"A 0.93 score is also enough to auto-tag banquet table layouts, but that belongs to a different classifier.","tags":["table-layouts","confidence"],"depth":2},{"id":"e5","text":"Below-threshold windshield guesses are stored as suggestions for adjuster review.","tags":["hail-damage-photos","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector paraphrase case from insurance image routing with same-threshold distractor from hospitality layouts."}
{"id":"paraphrase-hard-018","query":"How many unresolved guest complaints trigger a banquet operations review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A banquet operations review is triggered once 5 guest complaints remain unresolved.","tags":["guest-complaints","banquet-review","threshold"],"depth":1},{"id":"e2","text":"A banquet operations review is triggered once 4 guest complaints remain unresolved.","tags":["guest-complaints","banquet-review","threshold"],"depth":1},{"id":"e3","text":"Five unresolved guest complaints are enough to open a banquet operations review.","tags":["guest-complaints","banquet-review","threshold"],"depth":1},{"id":"e4","text":"Five unresolved safety punch-list items are enough to open a ballroom handoff review, but that is a different workflow.","tags":["safety-punchlists","threshold"],"depth":2},{"id":"e5","text":"The banquet review threshold is checked after duplicate complaint consolidation, not before it.","tags":["guest-complaints","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector paraphrase case from hospitality operations with same-number threshold in a nearby facilities workflow."}
{"id":"paraphrase-hard-019","query":"What is the default delay before a failed calibration alert is sent to the shift lead?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A failed calibration alert is sent to the shift lead after a default delay of 25 minutes.","tags":["failed-calibration-alerts","shift-lead","delay"],"depth":1},{"id":"e2","text":"A failed calibration alert is sent to the shift lead after a default delay of 15 minutes.","tags":["failed-calibration-alerts","shift-lead","delay"],"depth":1},{"id":"e3","text":"The default calibration workflow waits twenty-five minutes before alerting the shift lead.","tags":["failed-calibration-alerts","shift-lead","delay"],"depth":1},{"id":"e4","text":"The torque wrench self-test also waits 25 minutes before its final warning, but that timer is unrelated to shift-lead escalation.","tags":["torque-wrench-selftests","delay"],"depth":2},{"id":"e5","text":"The quality manager can shorten the delay manually during active containment procedures.","tags":["failed-calibration-alerts","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector paraphrase case from manufacturing quality with same-duration warning behavior in a neighboring system."}
{"id":"paraphrase-hard-020","query":"What is the default hold period before an unsigned disciplinary notice is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned disciplinary notice is escalated after a default hold period of 5 business days.","tags":["disciplinary-notices","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned disciplinary notice is escalated after a default hold period of 3 business days.","tags":["disciplinary-notices","escalation","delay"],"depth":1},{"id":"e3","text":"The default disciplinary workflow waits five business days before escalating an unsigned notice.","tags":["disciplinary-notices","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned permit affidavits also wait five business days before legal review, but that belongs to a different workflow.","tags":["permit-affidavits","delay"],"depth":2},{"id":"e5","text":"An HR director can waive the hold period only for documented urgent conduct cases.","tags":["disciplinary-notices","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-sector paraphrase case from HR employee relations with same-duration public-sector paperwork distractor."}
{"id":"env-hard-021","query":"What ledger replica host does treasury-sandbox use for settlement rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The treasury-sandbox environment uses ledger-treasury-sandbox.internal for settlement rehearsals.","tags":["treasury-sandbox","settlement-rehearsals","ledger-replica","infra"],"depth":1},{"id":"e2","text":"The production environment uses ledger-banking-prod.internal for settlement rehearsals.","tags":["production","settlement-rehearsals","ledger-replica","infra"],"depth":1},{"id":"e3","text":"The risk-preview environment uses ledger-risk-preview.internal for settlement rehearsals.","tags":["risk-preview","settlement-rehearsals","ledger-replica","infra"],"depth":2},{"id":"e4","text":"Treasury-sandbox replays old payment batches from ledger-treasury-sandbox-ro.internal, which is not the primary ledger replica.","tags":["treasury-sandbox","settlement-rehearsals","readonly"],"depth":2},{"id":"e5","text":"A treasury memo still lists sandbox-ledger-treasury.internal from before hostnames were normalized.","tags":["treasury-sandbox","settlement-rehearsals","historical"],"depth":3,"timestamp":"2025-09-16"},{"id":"e6","text":"The settlement rehearsal checklist notes that treasury-sandbox and risk-preview use similar sample wires but different ledger replicas.","tags":["treasury-sandbox","settlement-rehearsals","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from banking ops with risk-preview and read-only ledger distractors."}
{"id":"env-hard-022","query":"What routing queue does depot-training use for cold-chain exception drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The depot-training environment uses queue-depot-training.internal for cold-chain exception drills.","tags":["depot-training","cold-chain-exceptions","routing-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-logistics-prod.internal for cold-chain exception drills.","tags":["production","cold-chain-exceptions","routing-queue","infra"],"depth":1},{"id":"e3","text":"The fleet-preview environment uses queue-fleet-preview.internal for cold-chain exception drills.","tags":["fleet-preview","cold-chain-exceptions","routing-queue","infra"],"depth":2},{"id":"e4","text":"Depot-training stores failed route events in queue-depot-training-deadletter.internal, which is not the primary routing queue.","tags":["depot-training","cold-chain-exceptions","deadletter"],"depth":2},{"id":"e5","text":"An operations worksheet still refers to training-coldchain-queue.internal from before naming became environment-first.","tags":["depot-training","cold-chain-exceptions","historical"],"depth":3,"timestamp":"2025-08-03"},{"id":"e6","text":"The drill runbook notes that depot-training and fleet-preview share route maps but not routing queues.","tags":["depot-training","cold-chain-exceptions","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from logistics with dead-letter and route-map neighbor confusion."}
{"id":"env-hard-023","query":"What CMS asset host does editorial-preview use for embargoed homepage packages?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The editorial-preview environment uses assets-editorial-preview.internal for embargoed homepage packages.","tags":["editorial-preview","homepage-packages","cms-assets","infra"],"depth":1},{"id":"e2","text":"The production environment uses assets-media-prod.internal for embargoed homepage packages.","tags":["production","homepage-packages","cms-assets","infra"],"depth":1},{"id":"e3","text":"The ad-preview environment uses assets-ad-preview.internal for embargoed homepage packages.","tags":["ad-preview","homepage-packages","cms-assets","infra"],"depth":2},{"id":"e4","text":"Editorial-preview renders screenshots through renders-editorial-preview.internal, which is not the CMS asset host.","tags":["editorial-preview","homepage-packages","renders"],"depth":2},{"id":"e5","text":"A newsroom checklist still mentions preview-assets-editorial.internal from before asset hosts were renamed.","tags":["editorial-preview","homepage-packages","historical"],"depth":3,"timestamp":"2025-10-01"},{"id":"e6","text":"The embargo runbook notes that editorial-preview and ad-preview share package metadata but not asset hosts.","tags":["editorial-preview","homepage-packages","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from media publishing with render-host and ad-preview distractors."}
{"id":"env-hard-024","query":"What grant portal host does donor-sandbox use for renewal packet tests?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The donor-sandbox environment uses portal-donor-sandbox.internal for renewal packet tests.","tags":["donor-sandbox","renewal-packets","grant-portal","infra"],"depth":1},{"id":"e2","text":"The production environment uses portal-grants-prod.internal for renewal packet tests.","tags":["production","renewal-packets","grant-portal","infra"],"depth":1},{"id":"e3","text":"The reviewer-preview environment uses portal-reviewer-preview.internal for renewal packet tests.","tags":["reviewer-preview","renewal-packets","grant-portal","infra"],"depth":2},{"id":"e4","text":"Donor-sandbox generates PDF packets through pdf-donor-sandbox.internal, which is not the grant portal host.","tags":["donor-sandbox","renewal-packets","pdfs"],"depth":2},{"id":"e5","text":"An old foundation note still lists donor-portal-sandbox.internal from before hostnames were standardized.","tags":["donor-sandbox","renewal-packets","historical"],"depth":3,"timestamp":"2025-09-08"},{"id":"e6","text":"The grant test checklist notes that donor-sandbox and reviewer-preview share sample applications but not portal hosts.","tags":["donor-sandbox","renewal-packets","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from nonprofit grant administration with PDF-host and reviewer-preview confusion."}
{"id":"entity-hard-021","query":"What tea does Liora drink during wire exception reconciliation?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Liora drinks genmaicha during wire exception reconciliation.","tags":["liora","wire-exceptions","preference"],"depth":1},{"id":"e2","text":"Leora drinks genmaicha during wire exception reconciliation.","tags":["leora","wire-exceptions","preference"],"depth":1},{"id":"e3","text":"Liora drinks peppermint tea during branch onboarding days instead of wire reconciliation.","tags":["liora","branch-onboarding","preference"],"depth":2},{"id":"e4","text":"Lioraen drinks genmaicha during fraud case scrubs, not wire exception reconciliation.","tags":["lioraen","fraud-case-scrubs","preference"],"depth":2},{"id":"e5","text":"A treasury kitchen note said Liora and Leora both preferred unsweetened tea during long reconciliation sessions.","tags":["liora","leora","kitchen-note"],"depth":3},{"id":"e6","text":"Liora keeps the genmaicha sachets in the side pocket of the wire exception binder.","tags":["liora","wire-exceptions","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from banking operations with near-name overlap and task-specific beverage changes."}
{"id":"entity-hard-022","query":"What snack does Marisol bring to overnight route audits?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Marisol brings tamari almonds to overnight route audits.","tags":["marisol","route-audits","preference"],"depth":1},{"id":"e2","text":"Marisolyn brings tamari almonds to overnight route audits.","tags":["marisolyn","route-audits","preference"],"depth":1},{"id":"e3","text":"Marisol brings dried figs to daytime dispatch reviews instead of route audits.","tags":["marisol","dispatch-reviews","preference"],"depth":2},{"id":"e4","text":"Maribel brings tamari almonds to vehicle handoff checks, not overnight route audits.","tags":["maribel","vehicle-handoffs","preference"],"depth":2},{"id":"e5","text":"A depot snack order said Marisol and Marisolyn both preferred salty snacks that could be eaten one-handed.","tags":["marisol","marisolyn","depot-ops"],"depth":3},{"id":"e6","text":"Marisol says the tamari almonds are easier than chips when handling printed route manifests.","tags":["marisol","route-audits","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from logistics with near-name overlap and overnight-vs-daytime routine changes."}
{"id":"entity-hard-023","query":"What notebook does Iven use for fact-check huddles?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Iven uses a gray stenographer notebook for fact-check huddles.","tags":["iven","fact-check-huddles","preference"],"depth":1},{"id":"e2","text":"Evan uses a gray stenographer notebook for fact-check huddles.","tags":["evan","fact-check-huddles","preference"],"depth":1},{"id":"e3","text":"Iven uses a yellow reporter pad for interview prep instead of fact-check huddles.","tags":["iven","interview-prep","preference"],"depth":2},{"id":"e4","text":"Yvonne uses a gray stenographer notebook for copy desk shifts, not fact-check huddles.","tags":["yvonne","copy-desk","preference"],"depth":2},{"id":"e5","text":"A newsroom supply list once autocorrected Iven to Evan on the fact-check notebook line.","tags":["iven","evan","historical"],"depth":3,"timestamp":"2026-01-19"},{"id":"e6","text":"Iven keeps source confirmation marks in the margin of the gray stenographer notebook.","tags":["iven","fact-check-huddles","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from media operations with orthographic name collision and context-specific notebook choice."}
{"id":"entity-hard-024","query":"What drink does Saanvi bring to grant scoring panels?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Saanvi brings sparkling water with lime to grant scoring panels.","tags":["saanvi","grant-scoring-panels","preference"],"depth":1},{"id":"e2","text":"Sanvi brings sparkling water with lime to grant scoring panels.","tags":["sanvi","grant-scoring-panels","preference"],"depth":1},{"id":"e3","text":"Saanvi brings hot black tea to donor board meetings instead of scoring panels.","tags":["saanvi","donor-board","preference"],"depth":2},{"id":"e4","text":"Sarvi brings sparkling water with lime to site visit debriefs, not grant scoring panels.","tags":["sarvi","site-visit-debriefs","preference"],"depth":2},{"id":"e5","text":"A catering note said Saanvi and Sanvi both preferred non-sugary drinks during long panel sessions.","tags":["saanvi","sanvi","catering"],"depth":3},{"id":"e6","text":"Saanvi keeps the lime wedges in a separate cup so they do not spill on score sheets.","tags":["saanvi","grant-scoring-panels","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from nonprofit grant review with near-name overlap and panel-specific habits."}
{"id":"time-hard-021","query":"When did we require dual review on high-value escrow releases?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require dual review on high-value escrow releases in March.","tags":["timeline","escrow-releases","dual-review"],"depth":1,"timestamp":"2026-03-18"},{"id":"e2","text":"We proposed dual review on high-value escrow releases in January.","tags":["planning","escrow-releases","dual-review"],"depth":2,"timestamp":"2026-01-23"},{"id":"e3","text":"A treasury pilot required dual review in February for international escrow releases only.","tags":["trial","escrow-releases","dual-review"],"depth":2,"timestamp":"2026-02-11"},{"id":"e4","text":"The treasury operations guide was updated in April after dual review became mandatory.","tags":["docs","escrow-releases","dual-review"],"depth":2,"timestamp":"2026-04-04"},{"id":"e5","text":"The old single-approver emergency path was removed in May.","tags":["cleanup","escrow-releases","dual-review"],"depth":3,"timestamp":"2026-05-09"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from banking operations with an international-only pilot preceding full rollout."}
{"id":"time-hard-022","query":"When did we stop auto-clearing delayed cold-chain exceptions?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-clearing delayed cold-chain exceptions in September.","tags":["timeline","cold-chain-exceptions","auto-clearing"],"depth":1,"timestamp":"2026-09-15"},{"id":"e2","text":"We proposed stopping auto-clearing delayed cold-chain exceptions in July.","tags":["planning","cold-chain-exceptions","auto-clearing"],"depth":2,"timestamp":"2026-07-28"},{"id":"e3","text":"One depot disabled auto-clearing in August, but only for biologic shipments.","tags":["trial","cold-chain-exceptions","auto-clearing"],"depth":2,"timestamp":"2026-08-13"},{"id":"e4","text":"The logistics playbook was updated in October to reflect manual exception closure.","tags":["docs","cold-chain-exceptions","auto-clearing"],"depth":2,"timestamp":"2026-10-06"},{"id":"e5","text":"A cleanup in November removed the old auto-clear dashboard cards.","tags":["cleanup","cold-chain-exceptions","auto-clearing"],"depth":3,"timestamp":"2026-11-05"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from logistics with a shipment-type pilot that competes with the actual policy change."}
{"id":"time-hard-023","query":"When did we rename the correction desk to the standards desk?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the correction desk to the standards desk in June.","tags":["timeline","correction-desk","standards-desk"],"depth":1,"timestamp":"2026-06-20"},{"id":"e2","text":"We discussed renaming the correction desk to the standards desk in April.","tags":["planning","correction-desk","standards-desk"],"depth":2,"timestamp":"2026-04-26"},{"id":"e3","text":"An editorial draft started using standards desk in May before the rename was official.","tags":["trial","correction-desk","standards-desk"],"depth":2,"timestamp":"2026-05-12"},{"id":"e4","text":"The newsroom glossary removed the correction desk term in July.","tags":["docs","correction-desk","standards-desk"],"depth":2,"timestamp":"2026-07-03"},{"id":"e5","text":"The old correction desk dashboard label disappeared in August.","tags":["cleanup","correction-desk","standards-desk"],"depth":3,"timestamp":"2026-08-08"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from media publishing with draft usage preceding the official terminology change."}
{"id":"time-hard-024","query":"When did we require lab director sign-off on specimen relabel requests?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require lab director sign-off on specimen relabel requests in December.","tags":["timeline","specimen-relabel-requests","director-signoff"],"depth":1,"timestamp":"2026-12-12"},{"id":"e2","text":"We proposed lab director sign-off on specimen relabel requests in October.","tags":["planning","specimen-relabel-requests","director-signoff"],"depth":2,"timestamp":"2026-10-24"},{"id":"e3","text":"A pathology pilot required director sign-off in November for blood culture relabels only.","tags":["trial","specimen-relabel-requests","director-signoff"],"depth":2,"timestamp":"2026-11-07"},{"id":"e4","text":"The specimen handling manual was updated in January after director sign-off became required.","tags":["docs","specimen-relabel-requests","director-signoff"],"depth":2,"timestamp":"2027-01-04"},{"id":"e5","text":"The old supervisor-only relabel path was removed in February.","tags":["cleanup","specimen-relabel-requests","director-signoff"],"depth":3,"timestamp":"2027-02-05"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from laboratory administration with specimen-type pilot preceding the full requirement."}
{"id":"state-hard-021","query":"What is the current timeout before an unconfirmed wire callback escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unconfirmed wire callback escalates is 25 minutes.","tags":["wire-callback-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unconfirmed wire callback escalates used to be 15 minutes.","tags":["wire-callback-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-19"},{"id":"e3","text":"The high-risk wire lane still uses a 5 minute callback escalation timeout.","tags":["wire-callback-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A treasury governance proposal recommended a 30 minute callback escalation timeout, but that was not adopted.","tags":["wire-callback-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-06"},{"id":"e5","text":"The payment dashboard separates callback escalation timers from settlement batch overdue timers.","tags":["wire-callback-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from banking with realistic minute-level alternatives and a high-risk override lane."}
{"id":"state-hard-022","query":"What is the current default review window for frozen shipment incident notes?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for frozen shipment incident notes is 36 hours.","tags":["frozen-shipment-incident-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for frozen shipment incident notes used to be 24 hours.","tags":["frozen-shipment-incident-review-window","historical"],"depth":2,"timestamp":"2025-12-07"},{"id":"e3","text":"The biologics lane still uses a 12 hour review window for frozen shipment incident notes.","tags":["frozen-shipment-incident-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A logistics governance proposal recommended a 48 hour review window, but that was not adopted.","tags":["frozen-shipment-incident-review-window","proposal"],"depth":2,"timestamp":"2026-03-10"},{"id":"e5","text":"The cold-chain dashboard tracks note review windows separately from delivery exception aging.","tags":["frozen-shipment-incident-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from logistics with lane-specific override and nearby hour-level alternatives."}
{"id":"state-hard-023","query":"What is the current canonical field name for homepage embargo lift time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for homepage embargo lift time is embargo_lift_at.","tags":["homepage-embargo-lift-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for homepage embargo lift time used to be homepage_embargo_time.","tags":["homepage-embargo-lift-field","historical"],"depth":2,"timestamp":"2025-10-31"},{"id":"e3","text":"A legacy publishing import still emits embargo_release_ts before normalization.","tags":["homepage-embargo-lift-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An editorial schema proposal recommended embargo_go_live_at, but it was rejected as too product-specific.","tags":["homepage-embargo-lift-field","proposal"],"depth":2,"timestamp":"2026-02-14"},{"id":"e5","text":"The publishing import normalizer maps older embargo field names onto the canonical one before validation.","tags":["homepage-embargo-lift-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from media publishing with plausible legacy field variants and canonical naming pressure."}
{"id":"state-hard-024","query":"What is the current reminder cadence for unsigned grant award letters?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unsigned grant award letters is every 8 days.","tags":["unsigned-grant-award-letter-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unsigned grant award letters used to be every 4 days.","tags":["unsigned-grant-award-letter-reminders","historical"],"depth":2,"timestamp":"2025-11-10"},{"id":"e3","text":"The urgent renewal lane still sends unsigned grant award letter reminders every 2 days.","tags":["unsigned-grant-award-letter-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A foundation governance proposal wanted biweekly reminders for unsigned award letters, but that was not adopted.","tags":["unsigned-grant-award-letter-reminders","proposal"],"depth":2,"timestamp":"2026-01-25"},{"id":"e5","text":"The grants dashboard separates unsigned award reminders from missing-report escalations.","tags":["unsigned-grant-award-letter-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from nonprofit grants with reminder cadence and urgent-lane confusion."}
{"id":"speaker-hard-021","query":"What did you suggest for banking and lending memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using callback timers, escrow sign-offs, settlement rehearsals, and high-risk override lanes so banking memories hinge on small procedural facts rather than generic finance language.","tags":["banking-lending-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested filling banking cases with account terms and product names because that would already sound specialized.","tags":["banking-lending-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested mixing ledger replicas, read-only neighbors, and lane-specific timings so the wrong answer stays operationally believable.","tags":["banking-lending-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how banking retrieval often fails on review steps and small timing thresholds, not on the broad product category.","tags":["banking-lending-memories","discussion"],"depth":2},{"id":"e5","text":"This batch adds treasury-sandbox, wire callback, and escrow dual-review examples built that way.","tags":["banking-lending-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case for banking/lending realism centered on procedural detail rather than vocabulary alone."}
{"id":"speaker-hard-022","query":"What did you suggest for logistics and shipping memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using route audits, cold-chain exceptions, frozen shipment notes, and depot training lanes so logistics memories feel like real operational recall.","tags":["logistics-shipping-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested relying on package terms and truck names because shipping language already sounds concrete.","tags":["logistics-shipping-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using route-map, dead-letter, and biologics-lane distractors because they are adjacent enough to fool retrieval without feeling random.","tags":["logistics-shipping-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that logistics memories get hardest when the wrong system is a real operational neighbor rather than a synthetic decoy.","tags":["logistics-shipping-memories","discussion"],"depth":2},{"id":"e5","text":"The newest depot and cold-chain cases now intentionally use those nearby-neighbor patterns.","tags":["logistics-shipping-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case about realistic logistics and shipping memory design with operational neighbors."}
{"id":"speaker-hard-023","query":"What did you suggest for media and publishing memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using embargo lifts, standards desks, fact-check huddles, and editorial-preview systems so publishing memories revolve around deadline-sensitive coordination details.","tags":["media-publishing-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using headline nouns and section names because journalism terms would already make the cases feel real.","tags":["media-publishing-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping render hosts, ad-preview systems, and draft terminology close to the real fact so the wrong answer feels plausible.","tags":["media-publishing-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how publishing retrieval often fails on embargo timing and desk naming rather than on the article topic itself.","tags":["media-publishing-memories","discussion"],"depth":2},{"id":"e5","text":"This batch now includes editorial-preview, standards-desk, and embargo field examples.","tags":["media-publishing-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case focused on media workflow realism with draft-vs-official and neighbor-system confusion."}
{"id":"speaker-hard-024","query":"What did you suggest for nonprofit and laboratory administration memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using grant reminders, donor renewal packets, specimen relabel requests, and scoring panels so nonprofit and lab memories stay grounded in administrative details rather than abstract domain labels.","tags":["nonprofit-lab-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using mostly donor and lab jargon because the specialized words would do most of the realism work.","tags":["nonprofit-lab-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested business-day reminders, sign-off rules, and lane-specific overrides because those are the kinds of tiny details memory systems actually confuse.","tags":["nonprofit-lab-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that both nonprofit and laboratory memories become difficult when the process is bureaucratic but still compact enough to inspect.","tags":["nonprofit-lab-memories","discussion"],"depth":2},{"id":"e5","text":"The donor-sandbox, unsigned award letter, and specimen relabel cases follow that same pattern.","tags":["nonprofit-lab-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case connecting nonprofit and lab admin through realistic procedural detail and compact case design."}
{"id":"paraphrase-hard-021","query":"What confidence is required before a wire note is auto-tagged as callback-complete?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A wire note is auto-tagged as callback-complete only at 0.94 confidence or higher.","tags":["wire-notes","callback-complete","confidence"],"depth":1},{"id":"e2","text":"A wire note is auto-tagged as callback-complete only at 0.84 confidence or higher.","tags":["wire-notes","callback-complete","confidence"],"depth":1},{"id":"e3","text":"The callback-complete tagger needs at least 0.94 confidence before it auto-labels a wire note.","tags":["wire-notes","callback-complete","confidence"],"depth":1},{"id":"e4","text":"A 0.94 score is also enough to auto-tag lab specimen pickups, but that belongs to a different classifier.","tags":["specimen-pickups","confidence"],"depth":2},{"id":"e5","text":"Below-threshold callback-complete guesses are stored as suggestions for reviewer confirmation.","tags":["wire-notes","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from banking with same-threshold distractor from laboratory logistics."}
{"id":"paraphrase-hard-022","query":"How many late depot scans trigger a cold-chain route review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A cold-chain route review is triggered once 6 depot scans are late.","tags":["depot-scans","cold-chain-route-review","threshold"],"depth":1},{"id":"e2","text":"A cold-chain route review is triggered once 4 depot scans are late.","tags":["depot-scans","cold-chain-route-review","threshold"],"depth":1},{"id":"e3","text":"Six late depot scans are enough to open a cold-chain route review.","tags":["depot-scans","cold-chain-route-review","threshold"],"depth":1},{"id":"e4","text":"Six missing homepage assets are enough to open a package readiness review, but that is a different workflow.","tags":["homepage-assets","threshold"],"depth":2},{"id":"e5","text":"The route review threshold is checked after duplicate scan consolidation, not before it.","tags":["depot-scans","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from logistics with same-number threshold in media publishing as distractor."}
{"id":"paraphrase-hard-023","query":"What is the default delay before an unsigned grant rubric is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned grant rubric is escalated after a default delay of 9 business days.","tags":["grant-rubrics","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned grant rubric is escalated after a default delay of 6 business days.","tags":["grant-rubrics","escalation","delay"],"depth":1},{"id":"e3","text":"The default grant workflow waits nine business days before escalating an unsigned rubric.","tags":["grant-rubrics","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned loan exception forms also wait nine business days before compliance review, but that belongs to a different workflow.","tags":["loan-exception-forms","delay"],"depth":2},{"id":"e5","text":"A program director can shorten the delay manually for emergency response grants.","tags":["grant-rubrics","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from nonprofit grants with same-duration banking compliance distractor."}
{"id":"paraphrase-hard-024","query":"What is the default hold time before a specimen pickup failure is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A specimen pickup failure is escalated after a default hold time of 35 minutes.","tags":["specimen-pickups","escalation","delay"],"depth":1},{"id":"e2","text":"A specimen pickup failure is escalated after a default hold time of 20 minutes.","tags":["specimen-pickups","escalation","delay"],"depth":1},{"id":"e3","text":"The default specimen pickup workflow waits thirty-five minutes before escalating a failure.","tags":["specimen-pickups","escalation","delay"],"depth":1},{"id":"e4","text":"The donor packet printer also waits 35 minutes before its final retry alert, but that timer is unrelated to pickup escalation.","tags":["donor-packets","delay"],"depth":2},{"id":"e5","text":"A lab supervisor can shorten the hold time manually during courier shortages.","tags":["specimen-pickups","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from laboratory administration with same-duration alert timing in nonprofit operations."}
{"id":"env-hard-025","query":"What outage queue does noc-training use for fiber cut simulations?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The noc-training environment uses queue-noc-training.internal for fiber cut simulations.","tags":["noc-training","fiber-cut-simulations","outage-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-telecom-prod.internal for fiber cut simulations.","tags":["production","fiber-cut-simulations","outage-queue","infra"],"depth":1},{"id":"e3","text":"The field-preview environment uses queue-field-preview.internal for fiber cut simulations.","tags":["field-preview","fiber-cut-simulations","outage-queue","infra"],"depth":2},{"id":"e4","text":"Noc-training stores failed simulation events in queue-noc-training-deadletter.internal, which is not the primary outage queue.","tags":["noc-training","fiber-cut-simulations","deadletter"],"depth":2},{"id":"e5","text":"An operations note still lists training-outage-queue.internal from before queue names were standardized.","tags":["noc-training","fiber-cut-simulations","historical"],"depth":3,"timestamp":"2025-09-11"},{"id":"e6","text":"The drill runbook notes that noc-training and field-preview share maps but not outage queues.","tags":["noc-training","fiber-cut-simulations","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from telecom operations with dead-letter and field-preview distractors."}
{"id":"env-hard-026","query":"What dispatch host does gate-sandbox use for deicing turnaround drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The gate-sandbox environment uses dispatch-gate-sandbox.internal for deicing turnaround drills.","tags":["gate-sandbox","deicing-turnarounds","dispatch-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses dispatch-airport-prod.internal for deicing turnaround drills.","tags":["production","deicing-turnarounds","dispatch-host","infra"],"depth":1},{"id":"e3","text":"The ops-preview environment uses dispatch-ops-preview.internal for deicing turnaround drills.","tags":["ops-preview","deicing-turnarounds","dispatch-host","infra"],"depth":2},{"id":"e4","text":"Gate-sandbox prints bay sheets through print-gate-sandbox.internal, which is not the dispatch host.","tags":["gate-sandbox","deicing-turnarounds","printing"],"depth":2},{"id":"e5","text":"A ramp checklist still mentions sandbox-dispatch-gate.internal from before hostnames were normalized.","tags":["gate-sandbox","deicing-turnarounds","historical"],"depth":3,"timestamp":"2025-10-07"},{"id":"e6","text":"The turnaround drill notes that gate-sandbox and ops-preview share aircraft fixtures but not dispatch hosts.","tags":["gate-sandbox","deicing-turnarounds","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from aviation operations with print-host and ops-preview confusion."}
{"id":"env-hard-027","query":"What telemetry broker does grid-lab use for transformer overload drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The grid-lab environment uses broker-grid-lab.internal for transformer overload drills.","tags":["grid-lab","transformer-overloads","telemetry-broker","infra"],"depth":1},{"id":"e2","text":"The production environment uses broker-utility-prod.internal for transformer overload drills.","tags":["production","transformer-overloads","telemetry-broker","infra"],"depth":1},{"id":"e3","text":"The dispatch-preview environment uses broker-dispatch-preview.internal for transformer overload drills.","tags":["dispatch-preview","transformer-overloads","telemetry-broker","infra"],"depth":2},{"id":"e4","text":"Grid-lab stores replay captures on capture-grid-lab.internal, which is not the telemetry broker.","tags":["grid-lab","transformer-overloads","captures"],"depth":2},{"id":"e5","text":"A substation memo still references lab-broker-grid.internal from before brokers were renamed.","tags":["grid-lab","transformer-overloads","historical"],"depth":3,"timestamp":"2025-08-20"},{"id":"e6","text":"The overload drill checklist notes that grid-lab and dispatch-preview share threshold templates but not broker hosts.","tags":["grid-lab","transformer-overloads","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from utilities with capture-host and dispatch-preview distractors."}
{"id":"env-hard-028","query":"What intake archive host does clinic-rights-preview use for eviction defense packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The clinic-rights-preview environment uses archive-clinic-rights-preview.internal for eviction defense packets.","tags":["clinic-rights-preview","eviction-defense","intake-archive","infra"],"depth":1},{"id":"e2","text":"The production environment uses archive-legal-aid-prod.internal for eviction defense packets.","tags":["production","eviction-defense","intake-archive","infra"],"depth":1},{"id":"e3","text":"The hearings-sandbox environment uses archive-hearings-sandbox.internal for eviction defense packets.","tags":["hearings-sandbox","eviction-defense","intake-archive","infra"],"depth":2},{"id":"e4","text":"Clinic-rights-preview generates packet PDFs through pdf-clinic-rights-preview.internal, which is not the archive host.","tags":["clinic-rights-preview","eviction-defense","pdfs"],"depth":2},{"id":"e5","text":"A legal aid note still lists rights-preview-archive.internal from before hostnames were standardized.","tags":["clinic-rights-preview","eviction-defense","historical"],"depth":3,"timestamp":"2025-09-27"},{"id":"e6","text":"The intake checklist notes that clinic-rights-preview and hearings-sandbox share sample cases but not archive hosts.","tags":["clinic-rights-preview","eviction-defense","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain environment case from legal aid with PDF-host and hearings-sandbox confusion."}
{"id":"entity-hard-025","query":"What tea does Nayeli drink during outage bridge moderation?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Nayeli drinks roasted barley tea during outage bridge moderation.","tags":["nayeli","outage-bridges","preference"],"depth":1},{"id":"e2","text":"Naeli drinks roasted barley tea during outage bridge moderation.","tags":["naeli","outage-bridges","preference"],"depth":1},{"id":"e3","text":"Nayeli drinks green tea during field handoff reviews instead of outage bridges.","tags":["nayeli","field-handoffs","preference"],"depth":2},{"id":"e4","text":"Nayela drinks roasted barley tea during switchroom tours, not outage bridge moderation.","tags":["nayela","switchroom-tours","preference"],"depth":2},{"id":"e5","text":"A NOC kitchen note said Nayeli and Naeli both preferred unsweetened tea during overnight incidents.","tags":["nayeli","naeli","kitchen-note"],"depth":3},{"id":"e6","text":"Nayeli keeps the barley tea thermos next to the incident bridge attendance sheet.","tags":["nayeli","outage-bridges","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from telecom ops with near-name overlap and overnight-vs-field context changes."}
{"id":"entity-hard-026","query":"What shoes does Mirel wear during runway lighting inspections?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mirel wears insulated ankle boots during runway lighting inspections.","tags":["mirel","runway-lighting-inspections","preference"],"depth":1},{"id":"e2","text":"Mirelle wears insulated ankle boots during runway lighting inspections.","tags":["mirelle","runway-lighting-inspections","preference"],"depth":1},{"id":"e3","text":"Mirel wears soft-soled deck shoes during terminal wayfinding checks instead of runway inspections.","tags":["mirel","terminal-wayfinding","preference"],"depth":2},{"id":"e4","text":"Muriel wears insulated ankle boots during hangar inventory rounds, not runway lighting inspections.","tags":["muriel","hangar-inventory","preference"],"depth":2},{"id":"e5","text":"A ramp safety note said Mirel and Mirelle both preferred boots with reflective tabs for night work.","tags":["mirel","mirelle","safety-note"],"depth":3},{"id":"e6","text":"Mirel keeps spare boot covers in the side pouch of the runway inspection bag.","tags":["mirel","runway-lighting-inspections","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from aviation with near-name overlap and night-work footwear preferences."}
{"id":"entity-hard-027","query":"What notebook does Tovah use for outage restoration briefings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Tovah uses a green grid notebook for outage restoration briefings.","tags":["tovah","restoration-briefings","preference"],"depth":1},{"id":"e2","text":"Tova uses a green grid notebook for outage restoration briefings.","tags":["tova","restoration-briefings","preference"],"depth":1},{"id":"e3","text":"Tovah uses a yellow legal pad during meter dispute reviews instead of restoration briefings.","tags":["tovah","meter-disputes","preference"],"depth":2},{"id":"e4","text":"Torah uses a green grid notebook during storm call-center drills, not restoration briefings.","tags":["torah","storm-call-center","preference"],"depth":2},{"id":"e5","text":"A utility supply order once swapped Tovah and Tova on the grid notebook line item.","tags":["tovah","tova","historical"],"depth":3,"timestamp":"2026-01-26"},{"id":"e6","text":"Tovah keeps feeder maps tucked into the back cover of the green grid notebook.","tags":["tovah","restoration-briefings","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from utilities with orthographic name collision and briefing-vs-review notebook changes."}
{"id":"entity-hard-028","query":"What drink does Roshni bring to protocol deviation board reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Roshni brings cold brew coffee to protocol deviation board reviews.","tags":["roshni","protocol-deviations","preference"],"depth":1},{"id":"e2","text":"Roshani brings cold brew coffee to protocol deviation board reviews.","tags":["roshani","protocol-deviations","preference"],"depth":1},{"id":"e3","text":"Roshni brings jasmine tea to consent form workshops instead of board reviews.","tags":["roshni","consent-workshops","preference"],"depth":2},{"id":"e4","text":"Roshan brings cold brew coffee to specimen chain audits, not protocol deviation board reviews.","tags":["roshan","specimen-audits","preference"],"depth":2},{"id":"e5","text":"An IRB catering note said Roshni and Roshani both preferred unsweetened caffeinated drinks during long review sessions.","tags":["roshni","roshani","catering"],"depth":3},{"id":"e6","text":"Roshni says the cold brew is the only thing that keeps her awake during protocol board reviews.","tags":["roshni","protocol-deviations","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Broad-domain entity case from research ethics with near-name overlap and review-vs-workshop context shifts."}
{"id":"time-hard-025","query":"When did we require supervisor confirmation on customer-owned modem swaps?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require supervisor confirmation on customer-owned modem swaps in October.","tags":["timeline","modem-swaps","supervisor-confirmation"],"depth":1,"timestamp":"2026-10-22"},{"id":"e2","text":"We proposed supervisor confirmation on customer-owned modem swaps in August.","tags":["planning","modem-swaps","supervisor-confirmation"],"depth":2,"timestamp":"2026-08-29"},{"id":"e3","text":"A field pilot required supervisor confirmation in September for business-line modem swaps only.","tags":["trial","modem-swaps","supervisor-confirmation"],"depth":2,"timestamp":"2026-09-13"},{"id":"e4","text":"The field operations guide was updated in November after supervisor confirmation became mandatory.","tags":["docs","modem-swaps","supervisor-confirmation"],"depth":2,"timestamp":"2026-11-04"},{"id":"e5","text":"The old self-approve swap path was removed in December.","tags":["cleanup","modem-swaps","supervisor-confirmation"],"depth":3,"timestamp":"2026-12-02"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from telecom with business-line pilot preceding a full operational requirement."}
{"id":"time-hard-026","query":"When did we stop auto-clearing gate tug checklist exceptions?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-clearing gate tug checklist exceptions in February.","tags":["timeline","gate-tug-exceptions","auto-clearing"],"depth":1,"timestamp":"2026-02-18"},{"id":"e2","text":"We proposed stopping auto-clearing gate tug checklist exceptions in December.","tags":["planning","gate-tug-exceptions","auto-clearing"],"depth":2,"timestamp":"2025-12-21"},{"id":"e3","text":"One snow-ops pilot disabled auto-clearing in January, but only for deicing tug checklists.","tags":["trial","gate-tug-exceptions","auto-clearing"],"depth":2,"timestamp":"2026-01-14"},{"id":"e4","text":"The apron operations handbook was updated in March to reflect manual exception closure.","tags":["docs","gate-tug-exceptions","auto-clearing"],"depth":2,"timestamp":"2026-03-05"},{"id":"e5","text":"A cleanup in April removed the old auto-clear dashboard tile.","tags":["cleanup","gate-tug-exceptions","auto-clearing"],"depth":3,"timestamp":"2026-04-09"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from aviation ops with a deicing-specific pilot preceding full policy change."}
{"id":"time-hard-027","query":"When did we rename the outage council to the restoration board?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the outage council to the restoration board in July.","tags":["timeline","outage-council","restoration-board"],"depth":1,"timestamp":"2026-07-17"},{"id":"e2","text":"We discussed renaming the outage council to the restoration board in May.","tags":["planning","outage-council","restoration-board"],"depth":2,"timestamp":"2026-05-30"},{"id":"e3","text":"A storm response draft started using restoration board in June before the rename was official.","tags":["trial","outage-council","restoration-board"],"depth":2,"timestamp":"2026-06-12"},{"id":"e4","text":"The utility glossary removed the outage council term in August.","tags":["docs","outage-council","restoration-board"],"depth":2,"timestamp":"2026-08-06"},{"id":"e5","text":"The old dashboard label disappeared in September.","tags":["cleanup","outage-council","restoration-board"],"depth":3,"timestamp":"2026-09-03"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from utilities with draft terminology preceding the official rename."}
{"id":"time-hard-028","query":"When did we require ethics chair sign-off on expedited consent amendments?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require ethics chair sign-off on expedited consent amendments in April.","tags":["timeline","consent-amendments","ethics-chair-signoff"],"depth":1,"timestamp":"2026-04-24"},{"id":"e2","text":"We proposed ethics chair sign-off on expedited consent amendments in February.","tags":["planning","consent-amendments","ethics-chair-signoff"],"depth":2,"timestamp":"2026-02-27"},{"id":"e3","text":"An oncology pilot required ethics chair sign-off in March for expedited consent amendments only.","tags":["trial","consent-amendments","ethics-chair-signoff"],"depth":2,"timestamp":"2026-03-16"},{"id":"e4","text":"The research compliance manual was updated in May after ethics chair sign-off became required.","tags":["docs","consent-amendments","ethics-chair-signoff"],"depth":2,"timestamp":"2026-05-02"},{"id":"e5","text":"The old coordinator-only expedited path was removed in June.","tags":["cleanup","consent-amendments","ethics-chair-signoff"],"depth":3,"timestamp":"2026-06-10"}],"relevant_ids":["e1"],"notes":"Broad-domain time case from research ethics with department-specific pilot preceding full requirement."}
{"id":"state-hard-025","query":"What is the current timeout before an unacknowledged tower alarm escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unacknowledged tower alarm escalates is 14 minutes.","tags":["tower-alarm-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unacknowledged tower alarm escalates used to be 8 minutes.","tags":["tower-alarm-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-21"},{"id":"e3","text":"The storm-priority lane still uses a 3 minute tower alarm escalation timeout.","tags":["tower-alarm-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A network governance proposal recommended a 20 minute tower alarm escalation timeout, but that was not adopted.","tags":["tower-alarm-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-09"},{"id":"e5","text":"The NOC dashboard separates tower alarm escalation timers from dispatch callback timers.","tags":["tower-alarm-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from telecom with minute-level alternatives and a storm-priority override lane."}
{"id":"state-hard-026","query":"What is the current default review window for runway light outage notes?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for runway light outage notes is 16 hours.","tags":["runway-light-outage-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for runway light outage notes used to be 8 hours.","tags":["runway-light-outage-review-window","historical"],"depth":2,"timestamp":"2025-12-03"},{"id":"e3","text":"The snow-ops lane still uses a 4 hour review window for runway light outage notes.","tags":["runway-light-outage-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An airport governance proposal recommended a 24 hour review window, but that was not adopted.","tags":["runway-light-outage-review-window","proposal"],"depth":2,"timestamp":"2026-03-17"},{"id":"e5","text":"The airfield dashboard tracks outage note review windows separately from NOTAM publication deadlines.","tags":["runway-light-outage-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from aviation with review-window alternatives and snow-ops override confusion."}
{"id":"state-hard-027","query":"What is the current canonical field name for emergency load shed trigger time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for emergency load shed trigger time is shed_triggered_at.","tags":["load-shed-trigger-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for emergency load shed trigger time used to be load_shed_time.","tags":["load-shed-trigger-field","historical"],"depth":2,"timestamp":"2025-10-29"},{"id":"e3","text":"A legacy outage import still emits shed_time_utc before normalization.","tags":["load-shed-trigger-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A grid data proposal recommended emergency_shed_at, but it was rejected as too broad.","tags":["load-shed-trigger-field","proposal"],"depth":2,"timestamp":"2026-02-12"},{"id":"e5","text":"The utility import normalizer maps older load shed field names onto the canonical one before validation.","tags":["load-shed-trigger-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from utilities data modeling with plausible legacy field variants and canonical rename pressure."}
{"id":"state-hard-028","query":"What is the current reminder cadence for unsigned participant reconsent forms?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unsigned participant reconsent forms is every 11 days.","tags":["participant-reconsent-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unsigned participant reconsent forms used to be every 5 days.","tags":["participant-reconsent-reminders","historical"],"depth":2,"timestamp":"2025-11-13"},{"id":"e3","text":"The urgent safety protocol lane still sends reconsent reminders every 2 days.","tags":["participant-reconsent-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An IRB governance proposal wanted biweekly reconsent reminders, but that was not adopted.","tags":["participant-reconsent-reminders","proposal"],"depth":2,"timestamp":"2026-01-20"},{"id":"e5","text":"The research dashboard separates unsigned reconsent reminders from missed visit follow-up alerts.","tags":["participant-reconsent-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Broad-domain state case from research ethics with cadence confusion and urgent-protocol override lane."}
{"id":"speaker-hard-025","query":"What did you suggest for telecom and network operations memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using outage bridges, tower alarm timers, fiber cut drills, and supervisor confirmation paths so telecom memories hinge on procedural details rather than just network terminology.","tags":["telecom-network-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using mostly device names and acronyms because telecom jargon already feels specific.","tags":["telecom-network-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested mixing training queues, priority lanes, and nearby dispatch timers so wrong answers sound operationally believable.","tags":["telecom-network-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how telecom retrieval usually fails on steps, timers, and environment neighbors rather than on the device label itself.","tags":["telecom-network-memories","discussion"],"depth":2},{"id":"e5","text":"This wave now adds noc-training, tower alarm, and modem swap cases that follow that pattern.","tags":["telecom-network-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case for telecom/network ops realism centered on timers, approvals, and neighboring systems."}
{"id":"speaker-hard-026","query":"What did you suggest for aviation and airport operations memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using dispatch hosts, runway inspections, tug checklist exceptions, and snow-ops lanes so aviation memories revolve around concrete airfield operations.","tags":["aviation-airport-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested relying on airport nouns and aircraft types because that would already sound operational.","tags":["aviation-airport-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested pairing dispatch and print systems, plus draft-vs-official timing, so the wrong answer stays close to a real airport workflow.","tags":["aviation-airport-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that aviation retrieval often fails on procedural exceptions and lane-specific timing rather than on the airport setting itself.","tags":["aviation-airport-memories","discussion"],"depth":2},{"id":"e5","text":"The new gate-sandbox and runway light cases were written to exploit those realistic near-misses.","tags":["aviation-airport-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case about realistic aviation operations memory pressure from adjacent systems and special lanes."}
{"id":"speaker-hard-027","query":"What did you suggest for energy and utilities memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using restoration boards, load shed fields, outage notes, and storm-priority lanes so utility memories depend on small but consequential operational details.","tags":["energy-utilities-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using mostly outage jargon and transformer terms because the domain words would already make the cases hard.","tags":["energy-utilities-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping broker hosts, capture systems, and glossary renames adjacent to the right answer so distractors feel like real grid operations.","tags":["energy-utilities-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how utility retrieval errors often come from neighboring control-room systems and renamed committees rather than from the outage topic itself.","tags":["energy-utilities-memories","discussion"],"depth":2},{"id":"e5","text":"The grid-lab, restoration-board, and load-shed field cases now follow that pattern deliberately.","tags":["energy-utilities-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case focused on utilities realism through neighboring control systems and renamed governance terms."}
{"id":"speaker-hard-028","query":"What did you suggest for legal aid and research ethics memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using archive hosts, hearing packets, sign-off rules, reconsent reminders, and protocol board reviews so legal aid and ethics memories revolve around small procedural facts.","tags":["legal-aid-ethics-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested leaning on legal and research jargon because the official terms alone would make the memories believable.","tags":["legal-aid-ethics-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using business-day delays, approval paths, and lane-specific overrides because that is where realistic retrieval confusion actually comes from.","tags":["legal-aid-ethics-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that both legal aid and research ethics become hard when similar forms and approvals differ by just one procedural step.","tags":["legal-aid-ethics-memories","discussion"],"depth":2},{"id":"e5","text":"This wave adds clinic-rights-preview, ethics chair sign-off, and reconsent reminder cases built that way.","tags":["legal-aid-ethics-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain speaker case linking legal aid and research ethics through compact but realistic procedural confusion."}
{"id":"paraphrase-hard-025","query":"What confidence is required before a fiber outage note is auto-tagged as customer-visible?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A fiber outage note is auto-tagged as customer-visible only at 0.89 confidence or higher.","tags":["fiber-outages","customer-visible","confidence"],"depth":1},{"id":"e2","text":"A fiber outage note is auto-tagged as customer-visible only at 0.79 confidence or higher.","tags":["fiber-outages","customer-visible","confidence"],"depth":1},{"id":"e3","text":"The customer-visible tagger needs at least 0.89 confidence before it auto-labels a fiber outage note.","tags":["fiber-outages","customer-visible","confidence"],"depth":1},{"id":"e4","text":"A 0.89 score is also enough to auto-tag runway debris reports, but that belongs to a different classifier.","tags":["runway-debris","confidence"],"depth":2},{"id":"e5","text":"Below-threshold customer-visible guesses are retained as suggestions for NOC review.","tags":["fiber-outages","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from telecom with same-threshold distractor from aviation safety reporting."}
{"id":"paraphrase-hard-026","query":"How many feeder alarms trigger a restoration board review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A restoration board review is triggered once 7 feeder alarms cluster together.","tags":["feeder-alarms","restoration-board","threshold"],"depth":1},{"id":"e2","text":"A restoration board review is triggered once 5 feeder alarms cluster together.","tags":["feeder-alarms","restoration-board","threshold"],"depth":1},{"id":"e3","text":"Seven feeder alarms attached to one event are enough to open a restoration board review.","tags":["feeder-alarms","restoration-board","threshold"],"depth":1},{"id":"e4","text":"Seven unresolved intake notes are enough to open an eviction defense escalation, but that is a different workflow.","tags":["intake-notes","threshold"],"depth":2},{"id":"e5","text":"The restoration board threshold is checked after duplicate alarm consolidation, not before it.","tags":["feeder-alarms","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from utilities with same-number threshold in legal aid intake as distractor."}
{"id":"paraphrase-hard-027","query":"What is the default delay before an unsigned runway checklist is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned runway checklist is escalated after a default delay of 4 hours.","tags":["runway-checklists","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned runway checklist is escalated after a default delay of 2 hours.","tags":["runway-checklists","escalation","delay"],"depth":1},{"id":"e3","text":"The default runway checklist workflow waits four hours before escalating an unsigned checklist.","tags":["runway-checklists","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned outage board notes also wait four hours before manager review, but that belongs to a different workflow.","tags":["outage-board-notes","delay"],"depth":2},{"id":"e5","text":"A duty manager can shorten the delay manually during active storm operations.","tags":["runway-checklists","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from aviation with same-duration utilities paperwork distractor."}
{"id":"paraphrase-hard-028","query":"What is the default hold period before an unsigned protocol deviation memo is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned protocol deviation memo is escalated after a default hold period of 6 business days.","tags":["protocol-deviation-memos","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned protocol deviation memo is escalated after a default hold period of 3 business days.","tags":["protocol-deviation-memos","escalation","delay"],"depth":1},{"id":"e3","text":"The default protocol deviation workflow waits six business days before escalating an unsigned memo.","tags":["protocol-deviation-memos","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned utility waiver letters also wait six business days before compliance review, but that belongs to a different workflow.","tags":["utility-waiver-letters","delay"],"depth":2},{"id":"e5","text":"An ethics chair can waive the hold period only for documented immediate-risk cases.","tags":["protocol-deviation-memos","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Broad-domain paraphrase case from research ethics with same-duration utilities compliance distractor."}
{"id":"env-hard-029","query":"What permit queue does streets-sandbox use for curb closure rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The streets-sandbox environment uses queue-streets-sandbox.internal for curb closure rehearsal permits.","tags":["streets-sandbox","curb-closure-rehearsals","permit-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-publicworks-prod.internal for curb closure rehearsal permits.","tags":["production","curb-closure-rehearsals","permit-queue","infra"],"depth":1},{"id":"e3","text":"The hearings-preview environment uses queue-hearings-preview.internal for curb closure rehearsal permits.","tags":["hearings-preview","curb-closure-rehearsals","permit-queue","infra"],"depth":2},{"id":"e4","text":"Streets-sandbox stores failed submissions in queue-streets-sandbox-deadletter.internal, which is not the primary permit queue.","tags":["streets-sandbox","curb-closure-rehearsals","deadletter"],"depth":2},{"id":"e5","text":"A public works note still lists sandbox-curb-queue.internal from before queue naming was standardized.","tags":["streets-sandbox","curb-closure-rehearsals","historical"],"depth":3,"timestamp":"2025-08-14"},{"id":"e6","text":"The rehearsal checklist notes that streets-sandbox and hearings-preview share mock applications but not queue hosts.","tags":["streets-sandbox","curb-closure-rehearsals","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh environment case from municipal public works with dead-letter and hearings-preview distractors."}
{"id":"env-hard-030","query":"What assay results host does pharma-preview use for batch release drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The pharma-preview environment uses results-pharma-preview.internal for batch release assay results.","tags":["pharma-preview","batch-release-drills","assay-results","infra"],"depth":1},{"id":"e2","text":"The production environment uses results-pharma-prod.internal for batch release assay results.","tags":["production","batch-release-drills","assay-results","infra"],"depth":1},{"id":"e3","text":"The qa-sandbox environment uses results-qa-sandbox.internal for batch release assay results.","tags":["qa-sandbox","batch-release-drills","assay-results","infra"],"depth":2},{"id":"e4","text":"Pharma-preview renders certificate PDFs through certs-pharma-preview.internal, which is not the assay results host.","tags":["pharma-preview","batch-release-drills","certificates"],"depth":2},{"id":"e5","text":"A validation worksheet still mentions preview-assays.internal from before results hosts were renamed.","tags":["pharma-preview","batch-release-drills","historical"],"depth":3,"timestamp":"2025-09-22"},{"id":"e6","text":"The drill notes that pharma-preview and qa-sandbox share sample batches but not results hosts.","tags":["pharma-preview","batch-release-drills","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh environment case from pharmaceutical QA with certificate-host and QA sandbox neighbors."}
{"id":"env-hard-031","query":"What blueprint host does site-lab use for crane path rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The site-lab environment uses blueprints-site-lab.internal for crane path rehearsals.","tags":["site-lab","crane-path-rehearsals","blueprint-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses blueprints-construction-prod.internal for crane path rehearsals.","tags":["production","crane-path-rehearsals","blueprint-host","infra"],"depth":1},{"id":"e3","text":"The safety-preview environment uses blueprints-safety-preview.internal for crane path rehearsals.","tags":["safety-preview","crane-path-rehearsals","blueprint-host","infra"],"depth":2},{"id":"e4","text":"Site-lab stores rendered crane overlays on overlays-site-lab.internal, which is not the blueprint host.","tags":["site-lab","crane-path-rehearsals","overlays"],"depth":2},{"id":"e5","text":"A foreman note still lists crane-blueprints-lab.internal from before blueprint hosts were renamed.","tags":["site-lab","crane-path-rehearsals","historical"],"depth":3,"timestamp":"2025-10-10"},{"id":"e6","text":"The rehearsal runbook notes that site-lab and safety-preview share path templates but not blueprint hosts.","tags":["site-lab","crane-path-rehearsals","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh environment case from construction operations with overlay-host and safety-preview confusion."}
{"id":"env-hard-032","query":"What catalog host does archive-preview use for traveling exhibit mock installs?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The archive-preview environment uses catalog-archive-preview.internal for traveling exhibit mock installs.","tags":["archive-preview","traveling-exhibits","catalog-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses catalog-museum-prod.internal for traveling exhibit mock installs.","tags":["production","traveling-exhibits","catalog-host","infra"],"depth":1},{"id":"e3","text":"The registrar-sandbox environment uses catalog-registrar-sandbox.internal for traveling exhibit mock installs.","tags":["registrar-sandbox","traveling-exhibits","catalog-host","infra"],"depth":2},{"id":"e4","text":"Archive-preview generates crate labels through labels-archive-preview.internal, which is not the catalog host.","tags":["archive-preview","traveling-exhibits","labels"],"depth":2},{"id":"e5","text":"An exhibit prep note still refers to preview-catalog-archive.internal from before hostnames were reordered.","tags":["archive-preview","traveling-exhibits","historical"],"depth":3,"timestamp":"2025-08-31"},{"id":"e6","text":"The mock install checklist notes that archive-preview and registrar-sandbox share object lists but not catalog hosts.","tags":["archive-preview","traveling-exhibits","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh environment case from museum/archive operations with label-host and registrar sandbox distractors."}
{"id":"entity-hard-029","query":"What drink does Kalina bring to permit counter reconciliations?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Kalina brings unsweetened hibiscus tea to permit counter reconciliations.","tags":["kalina","permit-counter","preference"],"depth":1},{"id":"e2","text":"Karina brings unsweetened hibiscus tea to permit counter reconciliations.","tags":["karina","permit-counter","preference"],"depth":1},{"id":"e3","text":"Kalina brings sparkling water to council packet assembly instead of permit counter reconciliations.","tags":["kalina","council-packets","preference"],"depth":2},{"id":"e4","text":"Malina brings unsweetened hibiscus tea to grant intake reviews, not permit counter reconciliations.","tags":["malina","grant-intake","preference"],"depth":2},{"id":"e5","text":"A clerk kitchen note said Kalina and Karina both avoided sugary drinks on long reconciliation mornings.","tags":["kalina","karina","kitchen-note"],"depth":3},{"id":"e6","text":"Kalina keeps the tea bottle beside the cash drawer discrepancy sheet.","tags":["kalina","permit-counter","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh entity case from public counter operations with near-name overlap and context-shifted beverage habits."}
{"id":"entity-hard-030","query":"What notebook does Devika use during batch disposition reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Devika uses a cream hardbound lab notebook during batch disposition reviews.","tags":["devika","batch-disposition","preference"],"depth":1},{"id":"e2","text":"Davika uses a cream hardbound lab notebook during batch disposition reviews.","tags":["davika","batch-disposition","preference"],"depth":1},{"id":"e3","text":"Devika uses a wirebound scratch pad during cleaning validation walks instead of batch reviews.","tags":["devika","cleaning-validation","preference"],"depth":2},{"id":"e4","text":"Devina uses a cream hardbound lab notebook during deviation triage, not batch disposition reviews.","tags":["devina","deviation-triage","preference"],"depth":2},{"id":"e5","text":"A QA supply order said Devika and Davika both requested notebooks with solvent-resistant covers.","tags":["devika","davika","supply-order"],"depth":3},{"id":"e6","text":"Devika keeps disposition signatures on the inside front cover of the cream notebook.","tags":["devika","batch-disposition","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh entity case from pharmaceutical QA with near-name overlap and work-mode notebook changes."}
{"id":"entity-hard-031","query":"What snack does Orin bring to crane safety briefings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Orin brings sesame rice crackers to crane safety briefings.","tags":["orin","crane-safety","preference"],"depth":1},{"id":"e2","text":"Orrin brings sesame rice crackers to crane safety briefings.","tags":["orrin","crane-safety","preference"],"depth":1},{"id":"e3","text":"Orin brings trail mix to subcontractor budget reviews instead of crane safety briefings.","tags":["orin","budget-reviews","preference"],"depth":2},{"id":"e4","text":"Corin brings sesame rice crackers to scaffold audits, not crane safety briefings.","tags":["corin","scaffold-audits","preference"],"depth":2},{"id":"e5","text":"A site snack order said Orin and Orrin both preferred snacks that did not leave grease on plans.","tags":["orin","orrin","site-ops"],"depth":3},{"id":"e6","text":"Orin keeps the rice crackers in the side pouch of the crane briefing binder.","tags":["orin","crane-safety","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh entity case from construction safety with near-name overlap and briefing-vs-budget context shifts."}
{"id":"entity-hard-032","query":"What bag does Sarela carry during object movement reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Sarela carries a navy crossbody satchel during object movement reviews.","tags":["sarela","object-movement-reviews","preference"],"depth":1},{"id":"e2","text":"Sariela carries a navy crossbody satchel during object movement reviews.","tags":["sariela","object-movement-reviews","preference"],"depth":1},{"id":"e3","text":"Sarela carries a rolling case to donor gallery walkthroughs instead of object movement reviews.","tags":["sarela","gallery-walkthroughs","preference"],"depth":2},{"id":"e4","text":"Mirela carries a navy crossbody satchel during archive climate checks, not object movement reviews.","tags":["mirela","climate-checks","preference"],"depth":2},{"id":"e5","text":"A registrar note said Sarela and Sariela both wanted hands-free bags for handling movement logs.","tags":["sarela","sariela","registrar-note"],"depth":3},{"id":"e6","text":"Sarela keeps object condition tags in the front zip pocket of the satchel.","tags":["sarela","object-movement-reviews","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Fresh entity case from museum collections work with near-name overlap and role-specific bag preferences."}
{"id":"time-hard-029","query":"When did we require second reviewer approval on curb access waivers?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require second reviewer approval on curb access waivers in May.","tags":["timeline","curb-access-waivers","second-reviewer"],"depth":1,"timestamp":"2026-05-21"},{"id":"e2","text":"We proposed second reviewer approval on curb access waivers in March.","tags":["planning","curb-access-waivers","second-reviewer"],"depth":2,"timestamp":"2026-03-28"},{"id":"e3","text":"A downtown pilot required second reviewer approval in April for loading-zone waivers only.","tags":["trial","curb-access-waivers","second-reviewer"],"depth":2,"timestamp":"2026-04-16"},{"id":"e4","text":"The curb management handbook was updated in June after second reviewer approval became mandatory.","tags":["docs","curb-access-waivers","second-reviewer"],"depth":2,"timestamp":"2026-06-05"},{"id":"e5","text":"The old single-review waiver path was removed in July.","tags":["cleanup","curb-access-waivers","second-reviewer"],"depth":3,"timestamp":"2026-07-07"}],"relevant_ids":["e1"],"notes":"Fresh time case from municipal permitting with a district-specific pilot preceding the actual policy change."}
{"id":"time-hard-030","query":"When did we stop auto-clearing minor assay drift notes?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-clearing minor assay drift notes in August.","tags":["timeline","assay-drift-notes","auto-clearing"],"depth":1,"timestamp":"2026-08-18"},{"id":"e2","text":"We proposed stopping auto-clearing minor assay drift notes in June.","tags":["planning","assay-drift-notes","auto-clearing"],"depth":2,"timestamp":"2026-06-22"},{"id":"e3","text":"One stability pilot disabled auto-clearing in July, but only for refrigerated batches.","tags":["trial","assay-drift-notes","auto-clearing"],"depth":2,"timestamp":"2026-07-09"},{"id":"e4","text":"The quality systems handbook was updated in September to reflect manual closure review.","tags":["docs","assay-drift-notes","auto-clearing"],"depth":2,"timestamp":"2026-09-03"},{"id":"e5","text":"A metrics cleanup removed the old auto-clear chart in October.","tags":["cleanup","assay-drift-notes","auto-clearing"],"depth":3,"timestamp":"2026-10-06"}],"relevant_ids":["e1"],"notes":"Fresh time case from pharmaceutical QA with refrigerated-batch pilot preceding full policy change."}
{"id":"time-hard-031","query":"When did we rename the lift meeting to the crane readiness review?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the lift meeting to the crane readiness review in January.","tags":["timeline","lift-meeting","crane-readiness-review"],"depth":1,"timestamp":"2026-01-27"},{"id":"e2","text":"We discussed renaming the lift meeting to the crane readiness review in November.","tags":["planning","lift-meeting","crane-readiness-review"],"depth":2,"timestamp":"2025-11-29"},{"id":"e3","text":"A site safety draft began using crane readiness review in December before the rename was official.","tags":["trial","lift-meeting","crane-readiness-review"],"depth":2,"timestamp":"2025-12-12"},{"id":"e4","text":"The construction glossary removed the lift meeting term in February.","tags":["docs","lift-meeting","crane-readiness-review"],"depth":2,"timestamp":"2026-02-05"},{"id":"e5","text":"The old meeting label disappeared from the schedule board in March.","tags":["cleanup","lift-meeting","crane-readiness-review"],"depth":3,"timestamp":"2026-03-04"}],"relevant_ids":["e1"],"notes":"Fresh time case from construction operations with draft terminology preceding the official rename."}
{"id":"time-hard-032","query":"When did we require conservator sign-off on emergency crate swaps?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require conservator sign-off on emergency crate swaps in October.","tags":["timeline","crate-swaps","conservator-signoff"],"depth":1,"timestamp":"2026-10-20"},{"id":"e2","text":"We proposed conservator sign-off on emergency crate swaps in August.","tags":["planning","crate-swaps","conservator-signoff"],"depth":2,"timestamp":"2026-08-26"},{"id":"e3","text":"An exhibit loan pilot required conservator sign-off in September for outbound swaps only.","tags":["trial","crate-swaps","conservator-signoff"],"depth":2,"timestamp":"2026-09-14"},{"id":"e4","text":"The collections handling manual was updated in November after conservator sign-off became required.","tags":["docs","crate-swaps","conservator-signoff"],"depth":2,"timestamp":"2026-11-03"},{"id":"e5","text":"The old registrar-only emergency path was removed in December.","tags":["cleanup","crate-swaps","conservator-signoff"],"depth":3,"timestamp":"2026-12-08"}],"relevant_ids":["e1"],"notes":"Fresh time case from museum collections operations with outbound-only pilot preceding full requirement."}
{"id":"state-hard-029","query":"What is the current timeout before an unassigned curb closure request escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unassigned curb closure request escalates is 55 minutes.","tags":["curb-closure-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unassigned curb closure request escalates used to be 35 minutes.","tags":["curb-closure-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-17"},{"id":"e3","text":"The emergency works lane still uses a 10 minute curb closure escalation timeout.","tags":["curb-closure-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A curb governance proposal recommended a 75 minute escalation timeout, but that was not adopted.","tags":["curb-closure-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-02"},{"id":"e5","text":"The permit dashboard separates curb closure escalation timers from inspection dispatch overdue timers.","tags":["curb-closure-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Fresh state case from public works permitting with realistic minute alternatives and emergency override lane."}
{"id":"state-hard-030","query":"What is the current default review window for batch release exception notes?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for batch release exception notes is 28 hours.","tags":["batch-release-exception-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for batch release exception notes used to be 16 hours.","tags":["batch-release-exception-review-window","historical"],"depth":2,"timestamp":"2025-12-04"},{"id":"e3","text":"The sterile products lane still uses an 8 hour review window for batch release exception notes.","tags":["batch-release-exception-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A quality governance proposal recommended a 36 hour review window, but that was not adopted.","tags":["batch-release-exception-review-window","proposal"],"depth":2,"timestamp":"2026-03-16"},{"id":"e5","text":"The QA dashboard tracks exception note review windows separately from lot disposition deadlines.","tags":["batch-release-exception-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Fresh state case from pharmaceutical QA with realistic hour alternatives and product-lane overrides."}
{"id":"state-hard-031","query":"What is the current canonical field name for crane path approval time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for crane path approval time is path_approved_at.","tags":["crane-path-approval-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for crane path approval time used to be crane_path_time.","tags":["crane-path-approval-field","historical"],"depth":2,"timestamp":"2025-10-27"},{"id":"e3","text":"A legacy site import still emits path_approval_ts before normalization.","tags":["crane-path-approval-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A construction data proposal recommended crane_clearance_at, but it was rejected as too broad.","tags":["crane-path-approval-field","proposal"],"depth":2,"timestamp":"2026-02-08"},{"id":"e5","text":"The construction import normalizer maps older crane path field names onto the canonical one before validation.","tags":["crane-path-approval-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Fresh state case from construction data modeling with plausible legacy field variants and canonical rename pressure."}
{"id":"state-hard-032","query":"What is the current reminder cadence for unsigned outbound loan agreements?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unsigned outbound loan agreements is every 12 days.","tags":["outbound-loan-agreement-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unsigned outbound loan agreements used to be every 6 days.","tags":["outbound-loan-agreement-reminders","historical"],"depth":2,"timestamp":"2025-11-08"},{"id":"e3","text":"The emergency exhibit lane still sends unsigned outbound loan agreement reminders every 2 days.","tags":["outbound-loan-agreement-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A collections governance proposal wanted biweekly reminders for unsigned outbound loans, but that was not adopted.","tags":["outbound-loan-agreement-reminders","proposal"],"depth":2,"timestamp":"2026-01-18"},{"id":"e5","text":"The collections dashboard separates unsigned loan reminders from missing-condition-report escalations.","tags":["outbound-loan-agreement-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Fresh state case from museum loans with cadence confusion and emergency-lane override."}
{"id":"speaker-hard-029","query":"What did you suggest for municipal and permitting memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using permit queues, waiver approvals, counter reconciliations, and emergency lanes so municipal memories depend on procedural detail instead of vague city vocabulary.","tags":["municipal-permitting-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested relying on street and council terminology because civic language already sounds realistic enough.","tags":["municipal-permitting-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping hearings previews, dead-letter queues, and district pilots close to the real answer so the wrong memory sounds like a real adjacent process.","tags":["municipal-permitting-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how permitting retrieval often fails on small approval rules and lane differences rather than on the permit topic itself.","tags":["municipal-permitting-memories","discussion"],"depth":2},{"id":"e5","text":"This wave adds streets-sandbox, curb waiver, and permit counter cases written that way.","tags":["municipal-permitting-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh speaker case focused on municipal and permitting realism through adjacent systems and approval detail."}
{"id":"speaker-hard-030","query":"What did you suggest for pharma and regulated manufacturing memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using batch reviews, assay drift notes, sterile-product overrides, and certificate neighbors so regulated manufacturing memories feel like actual quality operations.","tags":["regulated-manufacturing-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested leaning mostly on GMP jargon because the terminology itself would make the cases believable.","tags":["regulated-manufacturing-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested mixing result hosts, certificate systems, and product-lane overrides because those are the realistic places retrieval can go wrong.","tags":["regulated-manufacturing-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later noted that regulated manufacturing memories often fail on tiny process distinctions rather than on high-level compliance labels.","tags":["regulated-manufacturing-memories","discussion"],"depth":2},{"id":"e5","text":"The pharma-preview and batch release exception cases follow that same design principle.","tags":["regulated-manufacturing-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh speaker case for pharmaceutical and regulated manufacturing workflows with realistic process neighbors."}
{"id":"speaker-hard-031","query":"What did you suggest for construction and physical operations memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using readiness reviews, crane path approvals, safety briefings, and overlay hosts so construction memories stay grounded in field coordination details.","tags":["construction-physical-ops-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using equipment names and site jargon because that would already sound operational enough.","tags":["construction-physical-ops-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested pairing blueprint hosts with overlay systems and draft-vs-official meeting names so the wrong answer feels like a real neighboring process.","tags":["construction-physical-ops-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that construction retrieval gets hard when the distractor is an actual adjacent coordination artifact rather than random noise.","tags":["construction-physical-ops-memories","discussion"],"depth":2},{"id":"e5","text":"The site-lab and crane-readiness cases were written to exploit that kind of realistic near-miss.","tags":["construction-physical-ops-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh speaker case about realistic construction and physical operations memory pressure from neighboring artifacts."}
{"id":"speaker-hard-032","query":"What did you suggest for collections and cultural institution memories?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using catalog hosts, object movement reviews, crate swaps, and outbound loan reminders so collections memories depend on compact but consequential procedural facts.","tags":["collections-cultural-memories","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested relying on curator and exhibit terminology because museum language already feels distinctive.","tags":["collections-cultural-memories","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping registrar sandboxes, label hosts, emergency lanes, and sign-off paths adjacent to the real answer so distractors stay believable.","tags":["collections-cultural-memories","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later discussed how cultural institution retrieval often fails on handling procedures and sign-off differences rather than on object names.","tags":["collections-cultural-memories","discussion"],"depth":2},{"id":"e5","text":"The archive-preview, crate swap, and outbound loan cases now follow that pattern on purpose.","tags":["collections-cultural-memories","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh speaker case for museum/archive workflow realism with neighboring systems and emergency-lane distractors."}
{"id":"paraphrase-hard-029","query":"What confidence is required before a curb waiver note is auto-tagged as accessibility-related?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A curb waiver note is auto-tagged as accessibility-related only at 0.88 confidence or higher.","tags":["curb-waivers","accessibility","confidence"],"depth":1},{"id":"e2","text":"A curb waiver note is auto-tagged as accessibility-related only at 0.78 confidence or higher.","tags":["curb-waivers","accessibility","confidence"],"depth":1},{"id":"e3","text":"The accessibility tagger needs at least 0.88 confidence before it auto-labels a curb waiver note.","tags":["curb-waivers","accessibility","confidence"],"depth":1},{"id":"e4","text":"A 0.88 score is also enough to auto-tag exhibit accessibility labels, but that belongs to a different classifier.","tags":["exhibit-labels","confidence"],"depth":2},{"id":"e5","text":"Below-threshold accessibility guesses are retained as suggestions for permitting review.","tags":["curb-waivers","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh paraphrase case from municipal permitting with same-threshold museum accessibility distractor."}
{"id":"paraphrase-hard-030","query":"How many assay anomalies trigger a batch disposition review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A batch disposition review is triggered once 4 assay anomalies cluster together.","tags":["assay-anomalies","batch-disposition","threshold"],"depth":1},{"id":"e2","text":"A batch disposition review is triggered once 3 assay anomalies cluster together.","tags":["assay-anomalies","batch-disposition","threshold"],"depth":1},{"id":"e3","text":"Four assay anomalies attached to one batch are enough to open a disposition review.","tags":["assay-anomalies","batch-disposition","threshold"],"depth":1},{"id":"e4","text":"Four crane safety deviations are enough to open a site readiness review, but that is a different workflow.","tags":["crane-deviations","threshold"],"depth":2},{"id":"e5","text":"The disposition threshold is checked after duplicate anomaly consolidation, not before it.","tags":["assay-anomalies","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh paraphrase case from pharma QA with same-number construction safety distractor."}
{"id":"paraphrase-hard-031","query":"What is the default delay before an unsigned crane path review is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned crane path review is escalated after a default delay of 3 hours.","tags":["crane-path-reviews","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned crane path review is escalated after a default delay of 1 hour.","tags":["crane-path-reviews","escalation","delay"],"depth":1},{"id":"e3","text":"The default crane path workflow waits three hours before escalating an unsigned review.","tags":["crane-path-reviews","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned batch certificates also wait three hours before secondary review, but that belongs to a different workflow.","tags":["batch-certificates","delay"],"depth":2},{"id":"e5","text":"A site manager can shorten the delay manually during active lifts.","tags":["crane-path-reviews","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh paraphrase case from construction with same-duration pharmaceutical review distractor."}
{"id":"paraphrase-hard-032","query":"What is the default hold period before an unsigned outbound object report is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned outbound object report is escalated after a default hold period of 7 business days.","tags":["outbound-object-reports","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned outbound object report is escalated after a default hold period of 4 business days.","tags":["outbound-object-reports","escalation","delay"],"depth":1},{"id":"e3","text":"The default outbound object workflow waits seven business days before escalating an unsigned report.","tags":["outbound-object-reports","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned curb closure fee forms also wait seven business days before finance review, but that belongs to a different workflow.","tags":["curb-fee-forms","delay"],"depth":2},{"id":"e5","text":"A collections manager can waive the hold period only for documented emergency transit changes.","tags":["outbound-object-reports","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Fresh paraphrase case from museum operations with same-duration municipal finance distractor."}
{"id":"env-hard-033","query":"What review API host does benefits-preview use for dependent eligibility audits?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The benefits-preview environment uses api-benefits-preview.internal for dependent eligibility audit reviews.","tags":["benefits-preview","dependent-eligibility-audits","review-api","infra"],"depth":1},{"id":"e2","text":"The production environment uses api-hr-prod.internal for dependent eligibility audit reviews.","tags":["production","dependent-eligibility-audits","review-api","infra"],"depth":1},{"id":"e3","text":"The payroll-sandbox environment uses api-payroll-sandbox.internal for dependent eligibility audit reviews.","tags":["payroll-sandbox","dependent-eligibility-audits","review-api","infra"],"depth":2},{"id":"e4","text":"Benefits-preview exports decision PDFs through pdf-benefits-preview.internal, which is not the review API host.","tags":["benefits-preview","dependent-eligibility-audits","pdfs"],"depth":2},{"id":"e5","text":"A benefits migration note still mentions preview-benefits-api.internal from before hostnames were standardized.","tags":["benefits-preview","dependent-eligibility-audits","historical"],"depth":3,"timestamp":"2025-09-19"},{"id":"e6","text":"The audit checklist notes that benefits-preview and payroll-sandbox share sample families but not review API hosts.","tags":["benefits-preview","dependent-eligibility-audits","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder environment case with preview vs sandbox overlap, adjacent export system, and obsolete hostname alias."}
{"id":"env-hard-034","query":"What dispute queue does card-lab use for chargeback rehearsal packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The card-lab environment uses queue-card-lab.internal for chargeback rehearsal packets.","tags":["card-lab","chargeback-rehearsals","dispute-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-card-prod.internal for chargeback rehearsal packets.","tags":["production","chargeback-rehearsals","dispute-queue","infra"],"depth":1},{"id":"e3","text":"The fraud-preview environment uses queue-fraud-preview.internal for chargeback rehearsal packets.","tags":["fraud-preview","chargeback-rehearsals","dispute-queue","infra"],"depth":2},{"id":"e4","text":"Card-lab stores rejected dispute simulations in queue-card-lab-deadletter.internal, which is not the primary dispute queue.","tags":["card-lab","chargeback-rehearsals","deadletter"],"depth":2},{"id":"e5","text":"A card operations worksheet still lists disputes-lab-card.internal from before queue names were reordered.","tags":["card-lab","chargeback-rehearsals","historical"],"depth":3,"timestamp":"2025-08-12"},{"id":"e6","text":"The rehearsal notes say that card-lab and fraud-preview share synthetic merchants but not dispute queues.","tags":["card-lab","chargeback-rehearsals","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder environment case with fraud-preview temptation and host-order alias confusion."}
{"id":"env-hard-035","query":"What archive host does donor-appeals-preview use for reconsideration packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The donor-appeals-preview environment uses archive-donor-appeals-preview.internal for reconsideration packets.","tags":["donor-appeals-preview","reconsideration-packets","archive-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses archive-donor-prod.internal for reconsideration packets.","tags":["production","reconsideration-packets","archive-host","infra"],"depth":1},{"id":"e3","text":"The grants-sandbox environment uses archive-grants-sandbox.internal for reconsideration packets.","tags":["grants-sandbox","reconsideration-packets","archive-host","infra"],"depth":2},{"id":"e4","text":"Donor-appeals-preview renders mailing labels through labels-donor-appeals-preview.internal, which is not the archive host.","tags":["donor-appeals-preview","reconsideration-packets","labels"],"depth":2},{"id":"e5","text":"An advancement note still refers to preview-appeals-archive.internal from before naming was standardized.","tags":["donor-appeals-preview","reconsideration-packets","historical"],"depth":3,"timestamp":"2025-10-04"},{"id":"e6","text":"The reconsideration checklist notes that donor-appeals-preview and grants-sandbox share sample applicants but not archive hosts.","tags":["donor-appeals-preview","reconsideration-packets","ops"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder environment case with label-host and grants-sandbox neighbors plus stale alias confusion."}
{"id":"env-hard-036","query":"What specimen queue does pathology-drill use for relabel incident rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The pathology-drill environment uses queue-pathology-drill.internal for relabel incident rehearsals.","tags":["pathology-drill","relabel-incidents","specimen-queue","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-pathology-prod.internal for relabel incident rehearsals.","tags":["production","relabel-incidents","specimen-queue","infra"],"depth":1},{"id":"e3","text":"The accessioning-preview environment uses queue-accessioning-preview.internal for relabel incident rehearsals.","tags":["accessioning-preview","relabel-incidents","specimen-queue","infra"],"depth":2},{"id":"e4","text":"Pathology-drill stores quarantined rehearsal samples in queue-pathology-drill-hold.internal, which is not the primary specimen queue.","tags":["pathology-drill","relabel-incidents","hold-queue"],"depth":2},{"id":"e5","text":"A lab note still lists relabel-pathology-drill.internal from before queue names became explicit.","tags":["pathology-drill","relabel-incidents","historical"],"depth":3,"timestamp":"2025-09-01"},{"id":"e6","text":"The relabel drill checklist notes that pathology-drill and accessioning-preview share sample manifests but not specimen queues.","tags":["pathology-drill","relabel-incidents","runbook"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder environment case with hold-queue and accessioning-preview distractors plus older host alias."}
{"id":"entity-hard-033","query":"What snack does Mireen bring to dependent audit escalations?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mireen brings sea salt cashews to dependent audit escalations.","tags":["mireen","dependent-audits","preference"],"depth":1},{"id":"e2","text":"Miren brings sea salt cashews to dependent audit escalations.","tags":["miren","dependent-audits","preference"],"depth":1},{"id":"e3","text":"Mireen brings sliced pears to payroll variance reviews instead of dependent audit escalations.","tags":["mireen","payroll-variances","preference"],"depth":2},{"id":"e4","text":"Marien brings sea salt cashews to leave-of-absence reviews, not dependent audit escalations.","tags":["marien","leave-reviews","preference"],"depth":2},{"id":"e5","text":"An HR kitchen note said Mireen and Miren both preferred non-sticky snacks during long desk audits.","tags":["mireen","miren","kitchen-note"],"depth":3},{"id":"e6","text":"Mireen keeps the cashews in the side tray of the eligibility discrepancy cart.","tags":["mireen","dependent-audits","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder entity case with near-name collision, same snack distractor, and same-office adjacent workflow confusion."}
{"id":"entity-hard-034","query":"What notebook does Karsten use during chargeback war-room reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Karsten uses a black top-spiral notebook during chargeback war-room reviews.","tags":["karsten","chargeback-war-room","preference"],"depth":1},{"id":"e2","text":"Carsten uses a black top-spiral notebook during chargeback war-room reviews.","tags":["carsten","chargeback-war-room","preference"],"depth":1},{"id":"e3","text":"Karsten uses a grid legal pad during merchant onboarding reviews instead of chargeback war-room sessions.","tags":["karsten","merchant-onboarding","preference"],"depth":2},{"id":"e4","text":"Kiersten uses a black top-spiral notebook during dispute calibration huddles, not chargeback war-room reviews.","tags":["kiersten","dispute-calibration","preference"],"depth":2},{"id":"e5","text":"A dispute supply order said Karsten and Carsten both requested notebooks with stiff covers for standing-room reviews.","tags":["karsten","carsten","supply-order"],"depth":3},{"id":"e6","text":"Karsten clips merchant evidence tabs into the back spiral of the black notebook.","tags":["karsten","chargeback-war-room","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder entity case with same notebook across near names and adjacent dispute workflow confusion."}
{"id":"entity-hard-035","query":"What drink does Eleni bring to reconsideration board packets?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Eleni brings plain kefir to reconsideration board packets.","tags":["eleni","reconsideration-board","preference"],"depth":1},{"id":"e2","text":"Elina brings plain kefir to reconsideration board packets.","tags":["elina","reconsideration-board","preference"],"depth":1},{"id":"e3","text":"Eleni brings black tea to donor stewardship reviews instead of reconsideration packet sessions.","tags":["eleni","donor-stewardship","preference"],"depth":2},{"id":"e4","text":"Melina brings plain kefir to scholarship scoring reviews, not reconsideration board packets.","tags":["melina","scholarship-scoring","preference"],"depth":2},{"id":"e5","text":"An advancement catering note said Eleni and Elina both preferred unsweetened drinks during long packet reviews.","tags":["eleni","elina","catering"],"depth":3},{"id":"e6","text":"Eleni keeps the kefir bottle wrapped in a sleeve so it does not sweat onto appeal letters.","tags":["eleni","reconsideration-board","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder entity case with subtle name collision and same drink appearing in a neighboring review process."}
{"id":"entity-hard-036","query":"What mug does Soraya carry during relabel incident boards?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Soraya carries a narrow stainless tumbler during relabel incident boards.","tags":["soraya","relabel-incidents","preference"],"depth":1},{"id":"e2","text":"Suraya carries a narrow stainless tumbler during relabel incident boards.","tags":["suraya","relabel-incidents","preference"],"depth":1},{"id":"e3","text":"Soraya uses a ceramic mug during accessioning training days instead of relabel incident boards.","tags":["soraya","accessioning-training","preference"],"depth":2},{"id":"e4","text":"Sorina carries a narrow stainless tumbler during chain-of-custody audits, not relabel incident boards.","tags":["sorina","custody-audits","preference"],"depth":2},{"id":"e5","text":"A pathology break-room note said Soraya and Suraya both preferred spill-resistant cups during bench reviews.","tags":["soraya","suraya","break-room"],"depth":3},{"id":"e6","text":"Soraya clips the tumbler to the outside mesh pocket of the relabel board bag.","tags":["soraya","relabel-incidents","notes"],"depth":3}],"relevant_ids":["e1"],"notes":"Harder entity case with near-name overlap and same object preference in adjacent lab workflows."}
{"id":"time-hard-033","query":"When did we require second analyst confirmation on dependent override approvals?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require second analyst confirmation on dependent override approvals in September.","tags":["timeline","dependent-overrides","second-analyst"],"depth":1,"timestamp":"2026-09-11"},{"id":"e2","text":"We proposed second analyst confirmation on dependent override approvals in July.","tags":["planning","dependent-overrides","second-analyst"],"depth":2,"timestamp":"2026-07-26"},{"id":"e3","text":"A leave-team pilot required second analyst confirmation in August for domestic partner overrides only.","tags":["trial","dependent-overrides","second-analyst"],"depth":2,"timestamp":"2026-08-14"},{"id":"e4","text":"The benefits operations guide was updated in October after second analyst confirmation became mandatory.","tags":["docs","dependent-overrides","second-analyst"],"depth":2,"timestamp":"2026-10-03"},{"id":"e5","text":"The old single-review override path was removed in November.","tags":["cleanup","dependent-overrides","second-analyst"],"depth":3,"timestamp":"2026-11-06"}],"relevant_ids":["e1"],"notes":"Harder time case where a narrow domestic-partner pilot shadows the full benefits policy change."}
{"id":"time-hard-034","query":"When did we stop auto-closing low-value chargeback evidence gaps?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-closing low-value chargeback evidence gaps in June.","tags":["timeline","chargeback-evidence-gaps","auto-closing"],"depth":1,"timestamp":"2026-06-18"},{"id":"e2","text":"We proposed stopping auto-closing low-value chargeback evidence gaps in April.","tags":["planning","chargeback-evidence-gaps","auto-closing"],"depth":2,"timestamp":"2026-04-29"},{"id":"e3","text":"One merchant-risk pilot disabled auto-closing in May, but only for travel-category merchants.","tags":["trial","chargeback-evidence-gaps","auto-closing"],"depth":2,"timestamp":"2026-05-17"},{"id":"e4","text":"The disputes handbook was updated in July to reflect manual closure review.","tags":["docs","chargeback-evidence-gaps","auto-closing"],"depth":2,"timestamp":"2026-07-04"},{"id":"e5","text":"A dashboard cleanup removed the old auto-close card in August.","tags":["cleanup","chargeback-evidence-gaps","auto-closing"],"depth":3,"timestamp":"2026-08-08"}],"relevant_ids":["e1"],"notes":"Harder time case with merchant-category pilot that is lexically close to the final policy change."}
{"id":"time-hard-035","query":"When did we rename the appeals packet review to the reconsideration board?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the appeals packet review to the reconsideration board in February.","tags":["timeline","appeals-packet-review","reconsideration-board"],"depth":1,"timestamp":"2026-02-19"},{"id":"e2","text":"We discussed renaming the appeals packet review to the reconsideration board in December.","tags":["planning","appeals-packet-review","reconsideration-board"],"depth":2,"timestamp":"2025-12-20"},{"id":"e3","text":"An advancement draft began using reconsideration board in January before the rename was official.","tags":["trial","appeals-packet-review","reconsideration-board"],"depth":2,"timestamp":"2026-01-09"},{"id":"e4","text":"The donor glossary removed the appeals packet review term in March.","tags":["docs","appeals-packet-review","reconsideration-board"],"depth":2,"timestamp":"2026-03-05"},{"id":"e5","text":"The old label disappeared from the packet dashboard in April.","tags":["cleanup","appeals-packet-review","reconsideration-board"],"depth":3,"timestamp":"2026-04-07"}],"relevant_ids":["e1"],"notes":"Harder time case with draft terminology preceding the official rename and later dashboard cleanup."}
{"id":"time-hard-036","query":"When did we require pathology director approval on chain-of-custody corrections?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require pathology director approval on chain-of-custody corrections in January.","tags":["timeline","custody-corrections","director-approval"],"depth":1,"timestamp":"2026-01-22"},{"id":"e2","text":"We proposed pathology director approval on chain-of-custody corrections in November.","tags":["planning","custody-corrections","director-approval"],"depth":2,"timestamp":"2025-11-18"},{"id":"e3","text":"An accessioning pilot required director approval in December for external courier corrections only.","tags":["trial","custody-corrections","director-approval"],"depth":2,"timestamp":"2025-12-15"},{"id":"e4","text":"The specimen handling manual was updated in February after director approval became required.","tags":["docs","custody-corrections","director-approval"],"depth":2,"timestamp":"2026-02-06"},{"id":"e5","text":"The old supervisor-only correction path was removed in March.","tags":["cleanup","custody-corrections","director-approval"],"depth":3,"timestamp":"2026-03-10"}],"relevant_ids":["e1"],"notes":"Harder time case with courier-specific pilot preceding the full chain-of-custody policy change."}
{"id":"state-hard-033","query":"What is the current timeout before an unreviewed dependent discrepancy escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unreviewed dependent discrepancy escalates is 42 minutes.","tags":["dependent-discrepancy-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unreviewed dependent discrepancy escalates used to be 20 minutes.","tags":["dependent-discrepancy-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-09"},{"id":"e3","text":"The urgent coverage lane still uses a 5 minute discrepancy escalation timeout.","tags":["dependent-discrepancy-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A benefits governance proposal recommended a 60 minute discrepancy escalation timeout, but that was not adopted.","tags":["dependent-discrepancy-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-04"},{"id":"e5","text":"The benefits dashboard separates discrepancy escalation timers from enrollment document overdue timers.","tags":["dependent-discrepancy-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Harder state case with realistic minute spacing and an urgent lane that can easily outrank the default."}
{"id":"state-hard-034","query":"What is the current default review window for chargeback evidence holes?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for chargeback evidence holes is 30 hours.","tags":["chargeback-evidence-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for chargeback evidence holes used to be 18 hours.","tags":["chargeback-evidence-review-window","historical"],"depth":2,"timestamp":"2025-12-01"},{"id":"e3","text":"The travel-merchant lane still uses a 6 hour review window for chargeback evidence holes.","tags":["chargeback-evidence-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A dispute governance proposal recommended a 36 hour review window, but that was not adopted.","tags":["chargeback-evidence-review-window","proposal"],"depth":2,"timestamp":"2026-03-19"},{"id":"e5","text":"The dispute dashboard tracks evidence-hole review windows separately from rebuttal filing deadlines.","tags":["chargeback-evidence-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Harder state case with lane-specific override and nearby hour alternatives that all sound plausible."}
{"id":"state-hard-035","query":"What is the current canonical field name for reconsideration vote time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for reconsideration vote time is vote_recorded_at.","tags":["reconsideration-vote-time-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for reconsideration vote time used to be appeal_vote_time.","tags":["reconsideration-vote-time-field","historical"],"depth":2,"timestamp":"2025-10-13"},{"id":"e3","text":"A legacy advancement import still emits vote_timestamp before normalization.","tags":["reconsideration-vote-time-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A donor systems proposal recommended reconsideration_voted_at, but that was rejected as too wordy.","tags":["reconsideration-vote-time-field","proposal"],"depth":2,"timestamp":"2026-02-10"},{"id":"e5","text":"The appeals import normalizer maps older vote-time field names onto the canonical one before validation.","tags":["reconsideration-vote-time-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Harder state case with multiple plausible legacy field variants and canonical naming drift."}
{"id":"state-hard-036","query":"What is the current reminder cadence for unresolved relabel incident memos?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unresolved relabel incident memos is every 9 days.","tags":["relabel-incident-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unresolved relabel incident memos used to be every 4 days.","tags":["relabel-incident-reminders","historical"],"depth":2,"timestamp":"2025-11-14"},{"id":"e3","text":"The urgent contamination lane still sends relabel incident reminders every day.","tags":["relabel-incident-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A pathology governance proposal wanted biweekly relabel reminders, but that was not adopted.","tags":["relabel-incident-reminders","proposal"],"depth":2,"timestamp":"2026-01-24"},{"id":"e5","text":"The pathology dashboard separates unresolved relabel reminders from sample quarantine escalation timers.","tags":["relabel-incident-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Harder state case with cadence confusion plus an urgent-contamination override lane that is lexically tempting."}
{"id":"speaker-hard-033","query":"What did you suggest for making these new cases more challenging without making them noisy?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested compounding two or three realistic pressures inside one case, like stale aliases, lane-specific overrides, and near-name collisions, while keeping one primary adversary label.","tags":["harder-but-not-noisy","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just doubling the entry count so the cases would feel harder from size alone.","tags":["harder-but-not-noisy","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested making distractors wrong in only one dimension so they remain believable and retrieval-focused rather than obviously fake.","tags":["harder-but-not-noisy","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later compared dense hard cases with bloated ones and found the dense cases more informative.","tags":["harder-but-not-noisy","discussion"],"depth":2},{"id":"e5","text":"Several of the newest cases now mix stale aliases, special lanes, and neighboring systems under a single label.","tags":["harder-but-not-noisy","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about increasing retrieval difficulty through believable compound pressure, not through bulk."}
{"id":"speaker-hard-034","query":"What did you suggest for making state-update cases nastier?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested making the current default compete against an old default, a narrow override lane, and a rejected proposal so there are several plausible wrong answers.","tags":["nastier-state-updates","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only one stale value because it would be simpler to write and score.","tags":["nastier-state-updates","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping the numeric gaps small enough that lexical retrieval can easily prefer the wrong record.","tags":["nastier-state-updates","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later observed that small numeric deltas are often more adversarial than wildly different values.","tags":["nastier-state-updates","discussion"],"depth":2},{"id":"e5","text":"The latest timeout, review-window, and reminder-cadence cases follow that tighter-gap pattern.","tags":["nastier-state-updates","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case focused on state-update difficulty through multiple plausible alternatives and narrow numeric gaps."}
{"id":"speaker-hard-035","query":"What did you suggest for making time cases more deceptive?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested surrounding the real rollout date with a proposal, a narrow pilot, docs lag, and cleanup so models have several temporally nearby hooks to grab.","tags":["more-deceptive-time-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just asking for dates on isolated facts because timelines would already be hard enough.","tags":["more-deceptive-time-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested making the pilot lexically closer to the query than the final rollout whenever that stays realistic.","tags":["more-deceptive-time-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that narrow pilots and later handbook updates are especially strong distractors for time retrieval.","tags":["more-deceptive-time-cases","discussion"],"depth":2},{"id":"e5","text":"The newest chargeback, dependent override, and chain-of-custody time cases all lean on that pattern.","tags":["more-deceptive-time-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about improving time adversaries through clustered timeline milestones and lexically tempting pilots."}
{"id":"speaker-hard-036","query":"What did you suggest for making paraphrase cases harder than simple value swaps?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested including two correct phrasings alongside same-number or same-threshold distractors from a neighboring workflow so exact wording alone is not enough.","tags":["harder-paraphrase-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested writing one right sentence and one wrong sentence because anything more would be unnecessary.","tags":["harder-paraphrase-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested reusing the same duration or confidence score in a different subsystem so retrieval has to track the right concept, not just the number.","tags":["harder-paraphrase-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later observed that shared numbers across adjacent workflows create stronger paraphrase traps than generic wrong values.","tags":["harder-paraphrase-cases","discussion"],"depth":2},{"id":"e5","text":"Many of the recent confidence and delay cases now intentionally use that cross-workflow same-number pattern.","tags":["harder-paraphrase-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about paraphrase difficulty through concept tracking, not just value matching."}
{"id":"paraphrase-hard-033","query":"What confidence is required before a dependent audit note is auto-tagged as household-verified?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A dependent audit note is auto-tagged as household-verified only at 0.91 confidence or higher.","tags":["dependent-audits","household-verified","confidence"],"depth":1},{"id":"e2","text":"A dependent audit note is auto-tagged as household-verified only at 0.81 confidence or higher.","tags":["dependent-audits","household-verified","confidence"],"depth":1},{"id":"e3","text":"The household-verified tagger needs at least 0.91 confidence before it auto-labels a dependent audit note.","tags":["dependent-audits","household-verified","confidence"],"depth":1},{"id":"e4","text":"A 0.91 score is also enough to auto-tag specimen household contacts, but that belongs to a different classifier.","tags":["specimen-contacts","confidence"],"depth":2},{"id":"e5","text":"Below-threshold household-verified guesses are retained as suggestions for analyst review.","tags":["dependent-audits","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder paraphrase case with shared confidence value across adjacent but different review systems."}
{"id":"paraphrase-hard-034","query":"How many merchant evidence gaps trigger a dispute escalation review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A dispute escalation review is triggered once 5 merchant evidence gaps cluster together.","tags":["merchant-evidence-gaps","dispute-escalation","threshold"],"depth":1},{"id":"e2","text":"A dispute escalation review is triggered once 4 merchant evidence gaps cluster together.","tags":["merchant-evidence-gaps","dispute-escalation","threshold"],"depth":1},{"id":"e3","text":"Five merchant evidence gaps attached to one case are enough to open a dispute escalation review.","tags":["merchant-evidence-gaps","dispute-escalation","threshold"],"depth":1},{"id":"e4","text":"Five missing dependent documents are enough to open an eligibility escalation, but that is a different workflow.","tags":["dependent-documents","threshold"],"depth":2},{"id":"e5","text":"The dispute escalation threshold is checked after duplicate evidence merging, not before it.","tags":["merchant-evidence-gaps","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder paraphrase case with same-number distractor from another audit workflow and duplicate-merging pipeline context."}
{"id":"paraphrase-hard-035","query":"What is the default delay before an unsigned reconsideration packet is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned reconsideration packet is escalated after a default delay of 8 business days.","tags":["reconsideration-packets","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned reconsideration packet is escalated after a default delay of 5 business days.","tags":["reconsideration-packets","escalation","delay"],"depth":1},{"id":"e3","text":"The default reconsideration workflow waits eight business days before escalating an unsigned packet.","tags":["reconsideration-packets","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned dependent override memos also wait eight business days before governance review, but that belongs to a different workflow.","tags":["dependent-overrides","delay"],"depth":2},{"id":"e5","text":"A board chair can shorten the delay manually for emergency hardship reconsiderations.","tags":["reconsideration-packets","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder paraphrase case with same-duration distractor from benefits governance and emergency override context."}
{"id":"paraphrase-hard-036","query":"What is the default hold period before an unsigned relabel correction is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned relabel correction is escalated after a default hold period of 5 business days.","tags":["relabel-corrections","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned relabel correction is escalated after a default hold period of 2 business days.","tags":["relabel-corrections","escalation","delay"],"depth":1},{"id":"e3","text":"The default relabel correction workflow waits five business days before escalating an unsigned correction.","tags":["relabel-corrections","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned chargeback exception forms also wait five business days before supervisor review, but that belongs to a different workflow.","tags":["chargeback-exceptions","delay"],"depth":2},{"id":"e5","text":"A pathology director can waive the hold period only for documented specimen integrity risks.","tags":["relabel-corrections","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder paraphrase case with same-duration distractor from payments operations and a domain-specific override path."}
{"id":"env-hard-037","query":"What berth scheduler host does harbor-preview use for tug slot rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The harbor-preview environment uses scheduler-harbor-preview.internal for tug slot rehearsals.","tags":["harbor-preview","tug-slot-rehearsals","scheduler-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses scheduler-port-prod.internal for tug slot rehearsals.","tags":["production","tug-slot-rehearsals","scheduler-host","infra"],"depth":1},{"id":"e3","text":"The customs-sandbox environment uses scheduler-customs-sandbox.internal for tug slot rehearsals.","tags":["customs-sandbox","tug-slot-rehearsals","scheduler-host","infra"],"depth":2},{"id":"e4","text":"Harbor-preview prints berth cards through print-harbor-preview.internal, which is not the scheduler host.","tags":["harbor-preview","tug-slot-rehearsals","printing"],"depth":2},{"id":"e5","text":"A harbor note still lists preview-berth-scheduler.internal from before hostnames were standardized.","tags":["harbor-preview","tug-slot-rehearsals","historical"],"depth":3,"timestamp":"2025-09-14"}],"relevant_ids":["e1"],"notes":"Port operations environment case with customs sandbox and print-host distractors."}
{"id":"env-hard-038","query":"What review host does registrar-preview use for transfer credit audits?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The registrar-preview environment uses review-registrar-preview.internal for transfer credit audits.","tags":["registrar-preview","transfer-credit-audits","review-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses review-registrar-prod.internal for transfer credit audits.","tags":["production","transfer-credit-audits","review-host","infra"],"depth":1},{"id":"e3","text":"The bursar-sandbox environment uses review-bursar-sandbox.internal for transfer credit audits.","tags":["bursar-sandbox","transfer-credit-audits","review-host","infra"],"depth":2},{"id":"e4","text":"Registrar-preview exports audit letters through pdf-registrar-preview.internal, which is not the review host.","tags":["registrar-preview","transfer-credit-audits","pdfs"],"depth":2},{"id":"e5","text":"An advising note still refers to preview-credit-review.internal from before hostnames were normalized.","tags":["registrar-preview","transfer-credit-audits","historical"],"depth":3,"timestamp":"2025-08-22"}],"relevant_ids":["e1"],"notes":"University registrar environment case with bursar sandbox and PDF host confusion."}
{"id":"env-hard-039","query":"What moisture broker does crop-lab use for irrigation failure drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The crop-lab environment uses broker-crop-lab.internal for irrigation failure drills.","tags":["crop-lab","irrigation-failure-drills","moisture-broker","infra"],"depth":1},{"id":"e2","text":"The production environment uses broker-agri-prod.internal for irrigation failure drills.","tags":["production","irrigation-failure-drills","moisture-broker","infra"],"depth":1},{"id":"e3","text":"The weather-preview environment uses broker-weather-preview.internal for irrigation failure drills.","tags":["weather-preview","irrigation-failure-drills","moisture-broker","infra"],"depth":2},{"id":"e4","text":"Crop-lab stores replay traces on capture-crop-lab.internal, which is not the broker host.","tags":["crop-lab","irrigation-failure-drills","captures"],"depth":2},{"id":"e5","text":"A field note still mentions lab-moisture-broker.internal from before naming was standardized.","tags":["crop-lab","irrigation-failure-drills","historical"],"depth":3,"timestamp":"2025-10-02"}],"relevant_ids":["e1"],"notes":"Agriculture operations environment case with weather-preview and capture-host distractors."}
{"id":"env-hard-040","query":"What dispatch host does rail-preview use for turnout outage rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The rail-preview environment uses dispatch-rail-preview.internal for turnout outage rehearsals.","tags":["rail-preview","turnout-outages","dispatch-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses dispatch-transit-prod.internal for turnout outage rehearsals.","tags":["production","turnout-outages","dispatch-host","infra"],"depth":1},{"id":"e3","text":"The signals-sandbox environment uses dispatch-signals-sandbox.internal for turnout outage rehearsals.","tags":["signals-sandbox","turnout-outages","dispatch-host","infra"],"depth":2},{"id":"e4","text":"Rail-preview prints crew slips through print-rail-preview.internal, which is not the dispatch host.","tags":["rail-preview","turnout-outages","printing"],"depth":2},{"id":"e5","text":"A yard memo still lists preview-turnout-dispatch.internal from before renaming.","tags":["rail-preview","turnout-outages","historical"],"depth":3,"timestamp":"2025-09-11"}],"relevant_ids":["e1"],"notes":"Transit operations environment case with signals sandbox and print-host distractors."}
{"id":"env-hard-041","query":"What docket host does asylum-preview use for intake rehearing packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The asylum-preview environment uses docket-asylum-preview.internal for intake rehearing packets.","tags":["asylum-preview","intake-rehearings","docket-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses docket-legal-prod.internal for intake rehearing packets.","tags":["production","intake-rehearings","docket-host","infra"],"depth":1},{"id":"e3","text":"The hearings-sandbox environment uses docket-hearings-sandbox.internal for intake rehearing packets.","tags":["hearings-sandbox","intake-rehearings","docket-host","infra"],"depth":2},{"id":"e4","text":"Asylum-preview renders packet PDFs through pdf-asylum-preview.internal, which is not the docket host.","tags":["asylum-preview","intake-rehearings","pdfs"],"depth":2},{"id":"e5","text":"A clinic note still refers to preview-docket-rights.internal from before hostnames were standardized.","tags":["asylum-preview","intake-rehearings","historical"],"depth":3,"timestamp":"2025-10-09"}],"relevant_ids":["e1"],"notes":"Legal aid environment case with hearings sandbox and PDF host confusion."}
{"id":"env-hard-042","query":"What review host does grants-lab use for field report redactions?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The grants-lab environment uses review-grants-lab.internal for field report redactions.","tags":["grants-lab","field-report-redactions","review-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses review-grants-prod.internal for field report redactions.","tags":["production","field-report-redactions","review-host","infra"],"depth":1},{"id":"e3","text":"The donor-preview environment uses review-donor-preview.internal for field report redactions.","tags":["donor-preview","field-report-redactions","review-host","infra"],"depth":2},{"id":"e4","text":"Grants-lab generates cover sheets through pdf-grants-lab.internal, which is not the review host.","tags":["grants-lab","field-report-redactions","pdfs"],"depth":2},{"id":"e5","text":"A foundation note still mentions lab-report-review.internal from before naming cleanup.","tags":["grants-lab","field-report-redactions","historical"],"depth":3,"timestamp":"2025-08-28"}],"relevant_ids":["e1"],"notes":"Nonprofit grant environment case with donor-preview and PDF host distractors."}
{"id":"env-hard-043","query":"What panel host does trial-preview use for adverse event coding drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The trial-preview environment uses panel-trial-preview.internal for adverse event coding drills.","tags":["trial-preview","adverse-event-coding","panel-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses panel-research-prod.internal for adverse event coding drills.","tags":["production","adverse-event-coding","panel-host","infra"],"depth":1},{"id":"e3","text":"The monitor-sandbox environment uses panel-monitor-sandbox.internal for adverse event coding drills.","tags":["monitor-sandbox","adverse-event-coding","panel-host","infra"],"depth":2},{"id":"e4","text":"Trial-preview stores coded exports on export-trial-preview.internal, which is not the panel host.","tags":["trial-preview","adverse-event-coding","exports"],"depth":2},{"id":"e5","text":"A research note still lists preview-ae-panel.internal from before naming was standardized.","tags":["trial-preview","adverse-event-coding","historical"],"depth":3,"timestamp":"2025-09-07"}],"relevant_ids":["e1"],"notes":"Research administration environment case with monitor sandbox and export-host distractors."}
{"id":"env-hard-044","query":"What review host does newsroom-lab use for takedown rehearsal packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The newsroom-lab environment uses review-newsroom-lab.internal for takedown rehearsal packets.","tags":["newsroom-lab","takedown-rehearsals","review-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses review-media-prod.internal for takedown rehearsal packets.","tags":["production","takedown-rehearsals","review-host","infra"],"depth":1},{"id":"e3","text":"The legal-preview environment uses review-legal-preview.internal for takedown rehearsal packets.","tags":["legal-preview","takedown-rehearsals","review-host","infra"],"depth":2},{"id":"e4","text":"Newsroom-lab renders packet snapshots through render-newsroom-lab.internal, which is not the review host.","tags":["newsroom-lab","takedown-rehearsals","renders"],"depth":2},{"id":"e5","text":"A standards note still refers to lab-takedown-review.internal from before hostnames were reordered.","tags":["newsroom-lab","takedown-rehearsals","historical"],"depth":3,"timestamp":"2025-10-06"}],"relevant_ids":["e1"],"notes":"Media/legal environment case with legal-preview and render-host distractors."}
{"id":"env-hard-045","query":"What queue host does shelter-preview use for reunification packet drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The shelter-preview environment uses queue-shelter-preview.internal for reunification packet drills.","tags":["shelter-preview","reunification-packets","queue-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses queue-social-prod.internal for reunification packet drills.","tags":["production","reunification-packets","queue-host","infra"],"depth":1},{"id":"e3","text":"The intake-sandbox environment uses queue-intake-sandbox.internal for reunification packet drills.","tags":["intake-sandbox","reunification-packets","queue-host","infra"],"depth":2},{"id":"e4","text":"Shelter-preview stores deferred packets on queue-shelter-preview-hold.internal, which is not the primary queue host.","tags":["shelter-preview","reunification-packets","hold-queue"],"depth":2},{"id":"e5","text":"A casework note still mentions preview-reunification-queue.internal from before naming standardization.","tags":["shelter-preview","reunification-packets","historical"],"depth":3,"timestamp":"2025-09-03"}],"relevant_ids":["e1"],"notes":"Social services environment case with intake sandbox and hold-queue distractors."}
{"id":"env-hard-046","query":"What pricing host does market-preview use for hedging exception drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The market-preview environment uses pricing-market-preview.internal for hedging exception drills.","tags":["market-preview","hedging-exceptions","pricing-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses pricing-energy-prod.internal for hedging exception drills.","tags":["production","hedging-exceptions","pricing-host","infra"],"depth":1},{"id":"e3","text":"The risk-sandbox environment uses pricing-risk-sandbox.internal for hedging exception drills.","tags":["risk-sandbox","hedging-exceptions","pricing-host","infra"],"depth":2},{"id":"e4","text":"Market-preview exports settlement sheets through pdf-market-preview.internal, which is not the pricing host.","tags":["market-preview","hedging-exceptions","pdfs"],"depth":2},{"id":"e5","text":"A trading note still lists preview-hedge-pricing.internal from before hostnames were normalized.","tags":["market-preview","hedging-exceptions","historical"],"depth":3,"timestamp":"2025-08-30"}],"relevant_ids":["e1"],"notes":"Energy market environment case with risk sandbox and PDF export distractors."}
{"id":"env-hard-047","query":"What archive host does archive-lab use for rights clearances rehearsal packets?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The archive-lab environment uses archive-archive-lab.internal for rights clearance rehearsal packets.","tags":["archive-lab","rights-clearances","archive-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses archive-media-rights-prod.internal for rights clearance rehearsal packets.","tags":["production","rights-clearances","archive-host","infra"],"depth":1},{"id":"e3","text":"The legal-sandbox environment uses archive-legal-sandbox.internal for rights clearance rehearsal packets.","tags":["legal-sandbox","rights-clearances","archive-host","infra"],"depth":2},{"id":"e4","text":"Archive-lab prints rights notices through print-archive-lab.internal, which is not the archive host.","tags":["archive-lab","rights-clearances","printing"],"depth":2},{"id":"e5","text":"An archive note still refers to lab-rights-archive.internal from before hostnames were standardized.","tags":["archive-lab","rights-clearances","historical"],"depth":3,"timestamp":"2025-09-15"}],"relevant_ids":["e1"],"notes":"Rights management environment case with legal sandbox and print-host distractors."}
{"id":"env-hard-048","query":"What review host does transit-preview use for fare dispute packet drills?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The transit-preview environment uses review-transit-preview.internal for fare dispute packet drills.","tags":["transit-preview","fare-disputes","review-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses review-transit-prod.internal for fare dispute packet drills.","tags":["production","fare-disputes","review-host","infra"],"depth":1},{"id":"e3","text":"The collections-sandbox environment uses review-collections-sandbox.internal for fare dispute packet drills.","tags":["collections-sandbox","fare-disputes","review-host","infra"],"depth":2},{"id":"e4","text":"Transit-preview generates hearing summaries through pdf-transit-preview.internal, which is not the review host.","tags":["transit-preview","fare-disputes","pdfs"],"depth":2},{"id":"e5","text":"A station note still mentions preview-fare-review.internal from before naming cleanup.","tags":["transit-preview","fare-disputes","historical"],"depth":3,"timestamp":"2025-08-18"}],"relevant_ids":["e1"],"notes":"Transit dispute environment case with collections sandbox and PDF host distractors."}
{"id":"env-hard-049","query":"What panel host does ethics-preview use for consent variance rehearsals?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The ethics-preview environment uses panel-ethics-preview.internal for consent variance rehearsals.","tags":["ethics-preview","consent-variances","panel-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses panel-ethics-prod.internal for consent variance rehearsals.","tags":["production","consent-variances","panel-host","infra"],"depth":1},{"id":"e3","text":"The protocol-sandbox environment uses panel-protocol-sandbox.internal for consent variance rehearsals.","tags":["protocol-sandbox","consent-variances","panel-host","infra"],"depth":2},{"id":"e4","text":"Ethics-preview exports review packets through export-ethics-preview.internal, which is not the panel host.","tags":["ethics-preview","consent-variances","exports"],"depth":2},{"id":"e5","text":"An IRB note still refers to preview-consent-panel.internal from before hostnames were standardized.","tags":["ethics-preview","consent-variances","historical"],"depth":3,"timestamp":"2025-09-29"}],"relevant_ids":["e1"],"notes":"Research ethics environment case with protocol sandbox and export-host distractors."}
{"id":"entity-hard-037","query":"What tea does Nivara drink during berth reassignment reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Nivara drinks toasted rice tea during berth reassignment reviews.","tags":["nivara","berth-reassignments","preference"],"depth":1},{"id":"e2","text":"Nivra drinks toasted rice tea during berth reassignment reviews.","tags":["nivra","berth-reassignments","preference"],"depth":1},{"id":"e3","text":"Nivara drinks mint tea during customs dock briefings instead of berth reassignments.","tags":["nivara","customs-briefings","preference"],"depth":2},{"id":"e4","text":"Navira drinks toasted rice tea during pilotage roster checks, not berth reassignment reviews.","tags":["navira","pilotage-rosters","preference"],"depth":2},{"id":"e5","text":"A harbor kitchen note said Nivara and Nivra both preferred unsweetened tea on early shift reviews.","tags":["nivara","nivra","kitchen-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Port operations entity case with near-name overlap and adjacent harbor workflow confusion."}
{"id":"entity-hard-038","query":"What pen does Loric use for transfer exception signatures?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Loric uses a blue archival pen for transfer exception signatures.","tags":["loric","transfer-exceptions","preference"],"depth":1},{"id":"e2","text":"Lorin uses a blue archival pen for transfer exception signatures.","tags":["lorin","transfer-exceptions","preference"],"depth":1},{"id":"e3","text":"Loric switches to a black felt pen for commencement packet corrections instead of transfer exceptions.","tags":["loric","packet-corrections","preference"],"depth":2},{"id":"e4","text":"Toric uses a blue archival pen during degree audit appeals, not transfer exception signatures.","tags":["toric","degree-audits","preference"],"depth":2},{"id":"e5","text":"A registrar supply note said Loric and Lorin both preferred pens that did not smear on NCR forms.","tags":["loric","lorin","supply-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Registrar entity case with same-pen distractor and adjacent academic workflow confusion."}
{"id":"entity-hard-039","query":"What snack does Aveline bring to irrigation variance boards?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Aveline brings smoked almonds to irrigation variance boards.","tags":["aveline","irrigation-variances","preference"],"depth":1},{"id":"e2","text":"Avaline brings smoked almonds to irrigation variance boards.","tags":["avaline","irrigation-variances","preference"],"depth":1},{"id":"e3","text":"Aveline brings dried apricots to seed inventory reviews instead of irrigation variance boards.","tags":["aveline","seed-inventory","preference"],"depth":2},{"id":"e4","text":"Meline brings smoked almonds to weather anomaly triage, not irrigation variance boards.","tags":["meline","weather-anomalies","preference"],"depth":2},{"id":"e5","text":"A field office note said Aveline and Avaline both preferred dry snacks that did not stain moisture maps.","tags":["aveline","avaline","field-office"],"depth":3}],"relevant_ids":["e1"],"notes":"Agriculture entity case with near-name overlap and neighboring board review confusion."}
{"id":"entity-hard-040","query":"What drink does Parisa bring to turnout delay boards?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Parisa brings lemon seltzer to turnout delay boards.","tags":["parisa","turnout-delays","preference"],"depth":1},{"id":"e2","text":"Parissa brings lemon seltzer to turnout delay boards.","tags":["parissa","turnout-delays","preference"],"depth":1},{"id":"e3","text":"Parisa brings black coffee to fare gate audits instead of turnout delay boards.","tags":["parisa","fare-gate-audits","preference"],"depth":2},{"id":"e4","text":"Larisa brings lemon seltzer to dispatch staffing reviews, not turnout delay boards.","tags":["larisa","staffing-reviews","preference"],"depth":2},{"id":"e5","text":"A transit catering note said Parisa and Parissa both preferred drinks without sugar on overnight boards.","tags":["parisa","parissa","catering"],"depth":3}],"relevant_ids":["e1"],"notes":"Transit entity case with same drink distractor and adjacent overnight review confusion."}
{"id":"entity-hard-041","query":"What notebook does Sameer use during asylum intake rehearings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Sameer uses a maroon stitched notebook during asylum intake rehearings.","tags":["sameer","asylum-rehearings","preference"],"depth":1},{"id":"e2","text":"Samir uses a maroon stitched notebook during asylum intake rehearings.","tags":["samir","asylum-rehearings","preference"],"depth":1},{"id":"e3","text":"Sameer uses a yellow legal pad during family reunification screenings instead of rehearings.","tags":["sameer","reunification-screenings","preference"],"depth":2},{"id":"e4","text":"Sameena uses a maroon stitched notebook during detention intake reviews, not asylum rehearings.","tags":["sameena","detention-intake","preference"],"depth":2},{"id":"e5","text":"A clinic supply note said Sameer and Samir both preferred notebooks with stitched spines for court packet inserts.","tags":["sameer","samir","supply-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Legal aid entity case with near-name overlap and neighboring intake workflow confusion."}
{"id":"entity-hard-042","query":"What snack does Taliah bring to field report redaction sweeps?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Taliah brings salted pumpkin seeds to field report redaction sweeps.","tags":["taliah","field-report-redactions","preference"],"depth":1},{"id":"e2","text":"Taliyah brings salted pumpkin seeds to field report redaction sweeps.","tags":["taliyah","field-report-redactions","preference"],"depth":1},{"id":"e3","text":"Taliah brings dried blueberries to donor dashboard audits instead of redaction sweeps.","tags":["taliah","donor-dashboards","preference"],"depth":2},{"id":"e4","text":"Talin brings salted pumpkin seeds to grant compliance huddles, not redaction sweeps.","tags":["talin","grant-compliance","preference"],"depth":2},{"id":"e5","text":"A foundation snack order said Taliah and Taliyah both preferred small dry snacks for long review sessions.","tags":["taliah","taliyah","snack-order"],"depth":3}],"relevant_ids":["e1"],"notes":"Grant administration entity case with subtle name collision and same snack in adjacent review contexts."}
{"id":"entity-hard-043","query":"What mug does Revan use during adverse event coding syncs?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Revan uses a gray insulated mug during adverse event coding syncs.","tags":["revan","adverse-event-coding","preference"],"depth":1},{"id":"e2","text":"Revanah uses a gray insulated mug during adverse event coding syncs.","tags":["revanah","adverse-event-coding","preference"],"depth":1},{"id":"e3","text":"Revan uses a ceramic cup during protocol onboarding workshops instead of coding syncs.","tags":["revan","protocol-onboarding","preference"],"depth":2},{"id":"e4","text":"Devan uses a gray insulated mug during monitor issue huddles, not coding syncs.","tags":["devan","monitor-issues","preference"],"depth":2},{"id":"e5","text":"A trial operations note said Revan and Revanah both preferred lidded mugs during long coding reviews.","tags":["revan","revanah","operations-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Research admin entity case with same mug distractor and neighboring workflow confusion."}
{"id":"entity-hard-044","query":"What tea does Soren bring to takedown rehearsal huddles?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Soren brings smoky oolong to takedown rehearsal huddles.","tags":["soren","takedown-rehearsals","preference"],"depth":1},{"id":"e2","text":"Soeren brings smoky oolong to takedown rehearsal huddles.","tags":["soeren","takedown-rehearsals","preference"],"depth":1},{"id":"e3","text":"Soren brings green tea to corrections desk reviews instead of takedown rehearsals.","tags":["soren","corrections-desk","preference"],"depth":2},{"id":"e4","text":"Toren brings smoky oolong to archive rights reviews, not takedown rehearsal huddles.","tags":["toren","rights-reviews","preference"],"depth":2},{"id":"e5","text":"A newsroom kitchen note said Soren and Soeren both preferred strong tea for evening legal huddles.","tags":["soren","soeren","kitchen-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Media/legal entity case with transliterated-name confusion and adjacent legal-review context."}
{"id":"entity-hard-045","query":"What bag does Mirette carry during reunification packet checks?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Mirette carries a canvas zipper folio during reunification packet checks.","tags":["mirette","reunification-packets","preference"],"depth":1},{"id":"e2","text":"Miriette carries a canvas zipper folio during reunification packet checks.","tags":["miriette","reunification-packets","preference"],"depth":1},{"id":"e3","text":"Mirette carries a rolling tote during housing intake fairs instead of packet checks.","tags":["mirette","housing-intake","preference"],"depth":2},{"id":"e4","text":"Mariette carries a canvas zipper folio during family mediation reviews, not reunification packet checks.","tags":["mariette","mediation-reviews","preference"],"depth":2},{"id":"e5","text":"A shelter supply note said Mirette and Miriette both preferred slim folios that could hold intake clipboards.","tags":["mirette","miriette","supply-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Social services entity case with near-name overlap and packet-vs-fair context shift."}
{"id":"entity-hard-046","query":"What drink does Orena bring to hedging exception panels?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Orena brings unsweetened mate to hedging exception panels.","tags":["orena","hedging-exceptions","preference"],"depth":1},{"id":"e2","text":"Orina brings unsweetened mate to hedging exception panels.","tags":["orina","hedging-exceptions","preference"],"depth":1},{"id":"e3","text":"Orena brings plain iced tea to settlement variance reviews instead of hedging exception panels.","tags":["orena","settlement-variances","preference"],"depth":2},{"id":"e4","text":"Lorena brings unsweetened mate to trader limit scrubs, not hedging exception panels.","tags":["lorena","trader-limits","preference"],"depth":2},{"id":"e5","text":"A market ops catering note said Orena and Orina both preferred caffeinated drinks without sugar during long panels.","tags":["orena","orina","catering"],"depth":3}],"relevant_ids":["e1"],"notes":"Energy markets entity case with near-name overlap and adjacent trading workflow confusion."}
{"id":"entity-hard-047","query":"What notebook does Leontine use during rights clearance reviews?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Leontine uses a thin black composition notebook during rights clearance reviews.","tags":["leontine","rights-clearances","preference"],"depth":1},{"id":"e2","text":"Leotine uses a thin black composition notebook during rights clearance reviews.","tags":["leotine","rights-clearances","preference"],"depth":1},{"id":"e3","text":"Leontine uses a reporter pad during takedown appeal reviews instead of rights clearance reviews.","tags":["leontine","takedown-appeals","preference"],"depth":2},{"id":"e4","text":"Leonie uses a thin black composition notebook during rights intake checks, not rights clearance reviews.","tags":["leonie","rights-intake","preference"],"depth":2},{"id":"e5","text":"An archive supply note said Leontine and Leotine both preferred notebooks that fit rights binders without bending.","tags":["leontine","leotine","supply-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Rights management entity case with subtle name collision and adjacent appeal workflow confusion."}
{"id":"entity-hard-048","query":"What snack does Kirel bring to fare dispute hearings?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Kirel brings plain sesame sticks to fare dispute hearings.","tags":["kirel","fare-disputes","preference"],"depth":1},{"id":"e2","text":"Kyrel brings plain sesame sticks to fare dispute hearings.","tags":["kyrel","fare-disputes","preference"],"depth":1},{"id":"e3","text":"Kirel brings apple chips to platform access reviews instead of fare dispute hearings.","tags":["kirel","platform-access","preference"],"depth":2},{"id":"e4","text":"Mirel brings plain sesame sticks to signal incident reviews, not fare dispute hearings.","tags":["mirel","signal-incidents","preference"],"depth":2},{"id":"e5","text":"A transit snack order said Kirel and Kyrel both preferred dry snacks during hearing marathons.","tags":["kirel","kyrel","snack-order"],"depth":3}],"relevant_ids":["e1"],"notes":"Transit hearings entity case with same-snack distractor and neighboring review workflow confusion."}
{"id":"entity-hard-049","query":"What mug does Selene bring to consent variance boards?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Selene brings a white vacuum mug to consent variance boards.","tags":["selene","consent-variances","preference"],"depth":1},{"id":"e2","text":"Celine brings a white vacuum mug to consent variance boards.","tags":["celine","consent-variances","preference"],"depth":1},{"id":"e3","text":"Selene brings a glass bottle to monitor protocol trainings instead of variance boards.","tags":["selene","protocol-trainings","preference"],"depth":2},{"id":"e4","text":"Selina brings a white vacuum mug to safety letter reviews, not consent variance boards.","tags":["selina","safety-letters","preference"],"depth":2},{"id":"e5","text":"An IRB break-room note said Selene and Celine both preferred spill-proof mugs during long board sessions.","tags":["selene","celine","break-room"],"depth":3}],"relevant_ids":["e1"],"notes":"Research ethics entity case with near-name overlap and adjacent board-review context."}
{"id":"time-hard-037","query":"When did we require harbor chief approval on emergency berth swaps?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require harbor chief approval on emergency berth swaps in May.","tags":["timeline","berth-swaps","chief-approval"],"depth":1,"timestamp":"2026-05-14"},{"id":"e2","text":"We proposed harbor chief approval on emergency berth swaps in March.","tags":["planning","berth-swaps","chief-approval"],"depth":2,"timestamp":"2026-03-21"},{"id":"e3","text":"A customs pilot required chief approval in April for refrigerated berth swaps only.","tags":["trial","berth-swaps","chief-approval"],"depth":2,"timestamp":"2026-04-08"},{"id":"e4","text":"The port operations guide was updated in June after chief approval became mandatory.","tags":["docs","berth-swaps","chief-approval"],"depth":2,"timestamp":"2026-06-03"},{"id":"e5","text":"The old dock-supervisor-only swap path was removed in July.","tags":["cleanup","berth-swaps","chief-approval"],"depth":3,"timestamp":"2026-07-05"}],"relevant_ids":["e1"],"notes":"Port time case with refrigerated-berth pilot preceding full approval requirement."}
{"id":"time-hard-038","query":"When did we stop auto-posting transfer exception clears?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-posting transfer exception clears in August.","tags":["timeline","transfer-exceptions","auto-posting"],"depth":1,"timestamp":"2026-08-12"},{"id":"e2","text":"We proposed stopping auto-posting transfer exception clears in June.","tags":["planning","transfer-exceptions","auto-posting"],"depth":2,"timestamp":"2026-06-27"},{"id":"e3","text":"A graduate-school pilot disabled auto-posting in July, but only for nursing program transfers.","tags":["trial","transfer-exceptions","auto-posting"],"depth":2,"timestamp":"2026-07-15"},{"id":"e4","text":"The registrar handbook was updated in September to reflect manual clearance review.","tags":["docs","transfer-exceptions","auto-posting"],"depth":2,"timestamp":"2026-09-02"},{"id":"e5","text":"A cleanup in October removed the old auto-post dashboard widget.","tags":["cleanup","transfer-exceptions","auto-posting"],"depth":3,"timestamp":"2026-10-09"}],"relevant_ids":["e1"],"notes":"Registrar time case with program-specific pilot preceding the actual workflow change."}
{"id":"time-hard-039","query":"When did we rename the moisture review to the irrigation variance board?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the moisture review to the irrigation variance board in April.","tags":["timeline","moisture-review","irrigation-variance-board"],"depth":1,"timestamp":"2026-04-18"},{"id":"e2","text":"We discussed renaming the moisture review to the irrigation variance board in February.","tags":["planning","moisture-review","irrigation-variance-board"],"depth":2,"timestamp":"2026-02-24"},{"id":"e3","text":"A field draft started using irrigation variance board in March before the rename was official.","tags":["trial","moisture-review","irrigation-variance-board"],"depth":2,"timestamp":"2026-03-11"},{"id":"e4","text":"The ag operations glossary removed the moisture review term in May.","tags":["docs","moisture-review","irrigation-variance-board"],"depth":2,"timestamp":"2026-05-06"},{"id":"e5","text":"The old label disappeared from the irrigation dashboard in June.","tags":["cleanup","moisture-review","irrigation-variance-board"],"depth":3,"timestamp":"2026-06-08"}],"relevant_ids":["e1"],"notes":"Agriculture time case with draft term preceding the official rename."}
{"id":"time-hard-040","query":"When did we require rail chief sign-off on turnout bypass approvals?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require rail chief sign-off on turnout bypass approvals in November.","tags":["timeline","turnout-bypasses","chief-signoff"],"depth":1,"timestamp":"2026-11-17"},{"id":"e2","text":"We proposed rail chief sign-off on turnout bypass approvals in September.","tags":["planning","turnout-bypasses","chief-signoff"],"depth":2,"timestamp":"2026-09-20"},{"id":"e3","text":"A snow-ops pilot required chief sign-off in October for overnight bypasses only.","tags":["trial","turnout-bypasses","chief-signoff"],"depth":2,"timestamp":"2026-10-13"},{"id":"e4","text":"The rail operations guide was updated in December after chief sign-off became mandatory.","tags":["docs","turnout-bypasses","chief-signoff"],"depth":2,"timestamp":"2026-12-04"},{"id":"e5","text":"The old dispatcher-only bypass path was removed in January.","tags":["cleanup","turnout-bypasses","chief-signoff"],"depth":3,"timestamp":"2027-01-06"}],"relevant_ids":["e1"],"notes":"Transit time case with overnight-only pilot competing with the full policy rollout."}
{"id":"time-hard-041","query":"When did we require clinic director approval on emergency rehearing packets?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require clinic director approval on emergency rehearing packets in July.","tags":["timeline","rehearing-packets","director-approval"],"depth":1,"timestamp":"2026-07-16"},{"id":"e2","text":"We proposed clinic director approval on emergency rehearing packets in May.","tags":["planning","rehearing-packets","director-approval"],"depth":2,"timestamp":"2026-05-28"},{"id":"e3","text":"A detention docket pilot required director approval in June for weekend packets only.","tags":["trial","rehearing-packets","director-approval"],"depth":2,"timestamp":"2026-06-12"},{"id":"e4","text":"The clinic handbook was updated in August after director approval became required.","tags":["docs","rehearing-packets","director-approval"],"depth":2,"timestamp":"2026-08-05"},{"id":"e5","text":"The old attorney-only emergency path was removed in September.","tags":["cleanup","rehearing-packets","director-approval"],"depth":3,"timestamp":"2026-09-04"}],"relevant_ids":["e1"],"notes":"Legal aid time case with weekend-packet pilot preceding the broader approval requirement."}
{"id":"time-hard-042","query":"When did we stop auto-clearing redaction false positives?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-clearing redaction false positives in March.","tags":["timeline","redaction-false-positives","auto-clearing"],"depth":1,"timestamp":"2026-03-19"},{"id":"e2","text":"We proposed stopping auto-clearing redaction false positives in January.","tags":["planning","redaction-false-positives","auto-clearing"],"depth":2,"timestamp":"2026-01-30"},{"id":"e3","text":"A donor-report pilot disabled auto-clearing in February, but only for overseas field packets.","tags":["trial","redaction-false-positives","auto-clearing"],"depth":2,"timestamp":"2026-02-14"},{"id":"e4","text":"The grants review manual was updated in April to reflect manual closure review.","tags":["docs","redaction-false-positives","auto-clearing"],"depth":2,"timestamp":"2026-04-03"},{"id":"e5","text":"A dashboard cleanup removed the old auto-clear tile in May.","tags":["cleanup","redaction-false-positives","auto-clearing"],"depth":3,"timestamp":"2026-05-07"}],"relevant_ids":["e1"],"notes":"Grant redaction time case with overseas-packet pilot preceding the actual policy change."}
{"id":"time-hard-043","query":"When did we rename the coding sync to the variance panel?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the coding sync to the variance panel in September.","tags":["timeline","coding-sync","variance-panel"],"depth":1,"timestamp":"2026-09-18"},{"id":"e2","text":"We discussed renaming the coding sync to the variance panel in July.","tags":["planning","coding-sync","variance-panel"],"depth":2,"timestamp":"2026-07-24"},{"id":"e3","text":"An oncology draft started using variance panel in August before the rename was official.","tags":["trial","coding-sync","variance-panel"],"depth":2,"timestamp":"2026-08-16"},{"id":"e4","text":"The trial glossary removed the coding sync term in October.","tags":["docs","coding-sync","variance-panel"],"depth":2,"timestamp":"2026-10-05"},{"id":"e5","text":"The old label disappeared from the AE dashboard in November.","tags":["cleanup","coding-sync","variance-panel"],"depth":3,"timestamp":"2026-11-08"}],"relevant_ids":["e1"],"notes":"Research admin time case with department-specific draft terminology preceding the official rename."}
{"id":"time-hard-044","query":"When did we require standards editor approval on emergency takedown reversals?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require standards editor approval on emergency takedown reversals in April.","tags":["timeline","takedown-reversals","editor-approval"],"depth":1,"timestamp":"2026-04-21"},{"id":"e2","text":"We proposed standards editor approval on emergency takedown reversals in February.","tags":["planning","takedown-reversals","editor-approval"],"depth":2,"timestamp":"2026-02-28"},{"id":"e3","text":"A politics desk pilot required editor approval in March for election-night reversals only.","tags":["trial","takedown-reversals","editor-approval"],"depth":2,"timestamp":"2026-03-17"},{"id":"e4","text":"The newsroom standards manual was updated in May after editor approval became mandatory.","tags":["docs","takedown-reversals","editor-approval"],"depth":2,"timestamp":"2026-05-04"},{"id":"e5","text":"The old producer-only reversal path was removed in June.","tags":["cleanup","takedown-reversals","editor-approval"],"depth":3,"timestamp":"2026-06-09"}],"relevant_ids":["e1"],"notes":"Media/legal time case with election-night pilot preceding the broader approval rule."}
{"id":"time-hard-045","query":"When did we require case supervisor approval on reunification overrides?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require case supervisor approval on reunification overrides in December.","tags":["timeline","reunification-overrides","supervisor-approval"],"depth":1,"timestamp":"2026-12-11"},{"id":"e2","text":"We proposed case supervisor approval on reunification overrides in October.","tags":["planning","reunification-overrides","supervisor-approval"],"depth":2,"timestamp":"2026-10-20"},{"id":"e3","text":"A foster-placement pilot required supervisor approval in November for sibling-group overrides only.","tags":["trial","reunification-overrides","supervisor-approval"],"depth":2,"timestamp":"2026-11-12"},{"id":"e4","text":"The casework manual was updated in January after supervisor approval became required.","tags":["docs","reunification-overrides","supervisor-approval"],"depth":2,"timestamp":"2027-01-05"},{"id":"e5","text":"The old specialist-only override path was removed in February.","tags":["cleanup","reunification-overrides","supervisor-approval"],"depth":3,"timestamp":"2027-02-07"}],"relevant_ids":["e1"],"notes":"Social services time case with sibling-group pilot preceding the full override rule."}
{"id":"time-hard-046","query":"When did we stop auto-closing hedging mismatch notices?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We stop auto-closing hedging mismatch notices in October.","tags":["timeline","hedging-mismatch-notices","auto-closing"],"depth":1,"timestamp":"2026-10-15"},{"id":"e2","text":"We proposed stopping auto-closing hedging mismatch notices in August.","tags":["planning","hedging-mismatch-notices","auto-closing"],"depth":2,"timestamp":"2026-08-26"},{"id":"e3","text":"A gas-book pilot disabled auto-closing in September, but only for cross-day hedges.","tags":["trial","hedging-mismatch-notices","auto-closing"],"depth":2,"timestamp":"2026-09-09"},{"id":"e4","text":"The market controls guide was updated in November to reflect manual closure review.","tags":["docs","hedging-mismatch-notices","auto-closing"],"depth":2,"timestamp":"2026-11-06"},{"id":"e5","text":"A cleanup removed the old auto-close card in December.","tags":["cleanup","hedging-mismatch-notices","auto-closing"],"depth":3,"timestamp":"2026-12-04"}],"relevant_ids":["e1"],"notes":"Energy markets time case with cross-day hedge pilot preceding the real policy change."}
{"id":"time-hard-047","query":"When did we rename the rights review to the clearance board?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We rename the rights review to the clearance board in June.","tags":["timeline","rights-review","clearance-board"],"depth":1,"timestamp":"2026-06-13"},{"id":"e2","text":"We discussed renaming the rights review to the clearance board in April.","tags":["planning","rights-review","clearance-board"],"depth":2,"timestamp":"2026-04-22"},{"id":"e3","text":"An archive draft started using clearance board in May before the rename was official.","tags":["trial","rights-review","clearance-board"],"depth":2,"timestamp":"2026-05-10"},{"id":"e4","text":"The rights glossary removed the rights review term in July.","tags":["docs","rights-review","clearance-board"],"depth":2,"timestamp":"2026-07-02"},{"id":"e5","text":"The old label disappeared from the archive dashboard in August.","tags":["cleanup","rights-review","clearance-board"],"depth":3,"timestamp":"2026-08-05"}],"relevant_ids":["e1"],"notes":"Rights management time case with archive draft usage preceding the official rename."}
{"id":"time-hard-048","query":"When did we require hearings manager sign-off on fare dispute waivers?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require hearings manager sign-off on fare dispute waivers in March.","tags":["timeline","fare-dispute-waivers","manager-signoff"],"depth":1,"timestamp":"2026-03-20"},{"id":"e2","text":"We proposed hearings manager sign-off on fare dispute waivers in January.","tags":["planning","fare-dispute-waivers","manager-signoff"],"depth":2,"timestamp":"2026-01-27"},{"id":"e3","text":"A reduced-fare pilot required manager sign-off in February for senior-pass waivers only.","tags":["trial","fare-dispute-waivers","manager-signoff"],"depth":2,"timestamp":"2026-02-15"},{"id":"e4","text":"The hearings handbook was updated in April after manager sign-off became mandatory.","tags":["docs","fare-dispute-waivers","manager-signoff"],"depth":2,"timestamp":"2026-04-04"},{"id":"e5","text":"The old reviewer-only waiver path was removed in May.","tags":["cleanup","fare-dispute-waivers","manager-signoff"],"depth":3,"timestamp":"2026-05-06"}],"relevant_ids":["e1"],"notes":"Transit hearings time case with senior-pass pilot preceding the full waiver policy change."}
{"id":"time-hard-049","query":"When did we require ethics vice-chair approval on reconsent deviations?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require ethics vice-chair approval on reconsent deviations in November.","tags":["timeline","reconsent-deviations","vice-chair-approval"],"depth":1,"timestamp":"2026-11-19"},{"id":"e2","text":"We proposed ethics vice-chair approval on reconsent deviations in September.","tags":["planning","reconsent-deviations","vice-chair-approval"],"depth":2,"timestamp":"2026-09-25"},{"id":"e3","text":"A vaccine-study pilot required vice-chair approval in October for pediatric reconsent deviations only.","tags":["trial","reconsent-deviations","vice-chair-approval"],"depth":2,"timestamp":"2026-10-14"},{"id":"e4","text":"The research ethics manual was updated in December after vice-chair approval became required.","tags":["docs","reconsent-deviations","vice-chair-approval"],"depth":2,"timestamp":"2026-12-03"},{"id":"e5","text":"The old coordinator-only deviation path was removed in January.","tags":["cleanup","reconsent-deviations","vice-chair-approval"],"depth":3,"timestamp":"2027-01-07"}],"relevant_ids":["e1"],"notes":"Research ethics time case with pediatric-study pilot preceding the broader approval requirement."}
{"id":"state-hard-037","query":"What is the current timeout before an unassigned berth exception escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unassigned berth exception escalates is 32 minutes.","tags":["berth-exception-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unassigned berth exception escalates used to be 18 minutes.","tags":["berth-exception-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-04"},{"id":"e3","text":"The hazardous cargo lane still uses a 4 minute berth exception escalation timeout.","tags":["berth-exception-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A harbor governance proposal recommended a 45 minute berth exception escalation timeout, but that was not adopted.","tags":["berth-exception-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-06"},{"id":"e5","text":"The harbor dashboard separates berth escalation timers from tug dispatch overdue timers.","tags":["berth-exception-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Port state case with narrow numeric gaps and a hazardous-cargo override lane."}
{"id":"state-hard-038","query":"What is the current default review window for transfer exception notes?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for transfer exception notes is 22 hours.","tags":["transfer-exception-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for transfer exception notes used to be 12 hours.","tags":["transfer-exception-review-window","historical"],"depth":2,"timestamp":"2025-12-11"},{"id":"e3","text":"The nursing program lane still uses a 5 hour review window for transfer exception notes.","tags":["transfer-exception-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A registrar governance proposal recommended a 28 hour review window, but that was not adopted.","tags":["transfer-exception-review-window","proposal"],"depth":2,"timestamp":"2026-03-07"},{"id":"e5","text":"The registrar dashboard tracks exception review windows separately from transcript posting deadlines.","tags":["transfer-exception-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Registrar state case with program-lane override and nearby hour-level alternatives."}
{"id":"state-hard-039","query":"What is the current canonical field name for irrigation variance approval time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for irrigation variance approval time is variance_approved_at.","tags":["irrigation-variance-approval-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for irrigation variance approval time used to be moisture_variance_time.","tags":["irrigation-variance-approval-field","historical"],"depth":2,"timestamp":"2025-10-14"},{"id":"e3","text":"A legacy ag import still emits approval_timestamp before normalization.","tags":["irrigation-variance-approval-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A field systems proposal recommended irrigation_voted_at, but that was rejected as too board-specific.","tags":["irrigation-variance-approval-field","proposal"],"depth":2,"timestamp":"2026-02-05"},{"id":"e5","text":"The ag import normalizer maps older variance approval fields onto the canonical one before validation.","tags":["irrigation-variance-approval-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Agriculture state case with plausible legacy field variants and canonical naming pressure."}
{"id":"state-hard-040","query":"What is the current reminder cadence for unresolved turnout bypass memos?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unresolved turnout bypass memos is every 7 days.","tags":["turnout-bypass-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unresolved turnout bypass memos used to be every 3 days.","tags":["turnout-bypass-reminders","historical"],"depth":2,"timestamp":"2025-11-13"},{"id":"e3","text":"The snow-ops lane still sends turnout bypass reminders every day.","tags":["turnout-bypass-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A transit governance proposal wanted weekly turnout bypass reminders, but that was not adopted as the default.","tags":["turnout-bypass-reminders","proposal"],"depth":2,"timestamp":"2026-01-17"},{"id":"e5","text":"The transit dashboard separates unresolved bypass reminders from dispatch callback escalations.","tags":["turnout-bypass-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Transit state case with urgent snow-ops override and nearby cadence alternatives."}
{"id":"state-hard-041","query":"What is the current timeout before an unassigned rehearing packet escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unassigned rehearing packet escalates is 38 minutes.","tags":["rehearing-packet-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unassigned rehearing packet escalates used to be 16 minutes.","tags":["rehearing-packet-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-06"},{"id":"e3","text":"The detention lane still uses a 6 minute rehearing packet escalation timeout.","tags":["rehearing-packet-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A clinic governance proposal recommended a 50 minute rehearing packet escalation timeout, but that was not adopted.","tags":["rehearing-packet-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-12"},{"id":"e5","text":"The clinic dashboard separates packet escalation timers from attorney callback overdue timers.","tags":["rehearing-packet-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Legal aid state case with detention-lane override and realistic minute-level alternatives."}
{"id":"state-hard-042","query":"What is the current default review window for redaction false positives?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for redaction false positives is 26 hours.","tags":["redaction-false-positive-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for redaction false positives used to be 14 hours.","tags":["redaction-false-positive-review-window","historical"],"depth":2,"timestamp":"2025-12-09"},{"id":"e3","text":"The overseas packet lane still uses a 4 hour review window for redaction false positives.","tags":["redaction-false-positive-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A grants governance proposal recommended a 32 hour review window, but that was not adopted.","tags":["redaction-false-positive-review-window","proposal"],"depth":2,"timestamp":"2026-03-18"},{"id":"e5","text":"The grants dashboard tracks false-positive review windows separately from publication deadline holds.","tags":["redaction-false-positive-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Grant redaction state case with overseas-packet override and nearby hour alternatives."}
{"id":"state-hard-043","query":"What is the current canonical field name for coding panel decision time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for coding panel decision time is decision_recorded_at.","tags":["coding-panel-decision-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for coding panel decision time used to be ae_decision_time.","tags":["coding-panel-decision-field","historical"],"depth":2,"timestamp":"2025-10-12"},{"id":"e3","text":"A legacy trial import still emits decision_timestamp before normalization.","tags":["coding-panel-decision-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A research systems proposal recommended panel_decided_at, but that was rejected as too generic.","tags":["coding-panel-decision-field","proposal"],"depth":2,"timestamp":"2026-02-09"},{"id":"e5","text":"The trial import normalizer maps older coding decision fields onto the canonical one before validation.","tags":["coding-panel-decision-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Research admin state case with plausible legacy field variants and canonical naming drift."}
{"id":"state-hard-044","query":"What is the current reminder cadence for unresolved takedown reversal memos?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unresolved takedown reversal memos is every 6 days.","tags":["takedown-reversal-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unresolved takedown reversal memos used to be every 2 days.","tags":["takedown-reversal-reminders","historical"],"depth":2,"timestamp":"2025-11-05"},{"id":"e3","text":"The election-night lane still sends takedown reversal reminders every day.","tags":["takedown-reversal-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A standards governance proposal wanted weekly takedown reversal reminders, but that was not adopted.","tags":["takedown-reversal-reminders","proposal"],"depth":2,"timestamp":"2026-01-16"},{"id":"e5","text":"The newsroom dashboard separates unresolved reversal reminders from legal escalation timers.","tags":["takedown-reversal-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Media/legal state case with election-night override and tight cadence alternatives."}
{"id":"state-hard-045","query":"What is the current timeout before an unreviewed reunification override escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unreviewed reunification override escalates is 48 minutes.","tags":["reunification-override-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unreviewed reunification override escalates used to be 22 minutes.","tags":["reunification-override-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-18"},{"id":"e3","text":"The sibling-group lane still uses a 7 minute override escalation timeout.","tags":["reunification-override-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A casework governance proposal recommended a 65 minute override escalation timeout, but that was not adopted.","tags":["reunification-override-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-13"},{"id":"e5","text":"The casework dashboard separates override escalation timers from home-visit overdue timers.","tags":["reunification-override-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Social services state case with sibling-group lane override and realistic minute-level distractors."}
{"id":"state-hard-046","query":"What is the current default review window for hedging mismatch notices?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current default review window for hedging mismatch notices is 24 hours.","tags":["hedging-mismatch-review-window","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The default review window for hedging mismatch notices used to be 12 hours.","tags":["hedging-mismatch-review-window","historical"],"depth":2,"timestamp":"2025-12-06"},{"id":"e3","text":"The cross-day hedge lane still uses a 3 hour review window for mismatch notices.","tags":["hedging-mismatch-review-window","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A market governance proposal recommended a 30 hour review window, but that was not adopted.","tags":["hedging-mismatch-review-window","proposal"],"depth":2,"timestamp":"2026-03-21"},{"id":"e5","text":"The market dashboard tracks mismatch review windows separately from settlement break deadlines.","tags":["hedging-mismatch-review-window","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Energy markets state case with cross-day override and nearby hour-level alternatives."}
{"id":"state-hard-047","query":"What is the current canonical field name for rights clearance vote time?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current canonical field name for rights clearance vote time is clearance_voted_at.","tags":["rights-clearance-vote-field","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The canonical field name for rights clearance vote time used to be rights_vote_time.","tags":["rights-clearance-vote-field","historical"],"depth":2,"timestamp":"2025-10-19"},{"id":"e3","text":"A legacy archive import still emits vote_ts before normalization.","tags":["rights-clearance-vote-field","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A rights systems proposal recommended clearance_recorded_at, but that was rejected as too broad.","tags":["rights-clearance-vote-field","proposal"],"depth":2,"timestamp":"2026-02-11"},{"id":"e5","text":"The archive import normalizer maps older rights vote fields onto the canonical one before validation.","tags":["rights-clearance-vote-field","pipeline"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Rights management state case with multiple plausible legacy field variants and canonical naming drift."}
{"id":"state-hard-048","query":"What is the current reminder cadence for unresolved fare dispute waivers?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current reminder cadence for unresolved fare dispute waivers is every 8 days.","tags":["fare-dispute-waiver-reminders","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The reminder cadence for unresolved fare dispute waivers used to be every 4 days.","tags":["fare-dispute-waiver-reminders","historical"],"depth":2,"timestamp":"2025-11-03"},{"id":"e3","text":"The senior-pass lane still sends fare dispute waiver reminders every day.","tags":["fare-dispute-waiver-reminders","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A hearings governance proposal wanted biweekly fare dispute reminders, but that was not adopted.","tags":["fare-dispute-waiver-reminders","proposal"],"depth":2,"timestamp":"2026-01-22"},{"id":"e5","text":"The hearings dashboard separates unresolved waiver reminders from payment plan overdue alerts.","tags":["fare-dispute-waiver-reminders","ui"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Transit hearings state case with senior-pass override and cadence-level confusion."}
{"id":"state-hard-049","query":"What is the current timeout before an unreviewed reconsent deviation escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unreviewed reconsent deviation escalates is 34 minutes.","tags":["reconsent-deviation-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unreviewed reconsent deviation escalates used to be 14 minutes.","tags":["reconsent-deviation-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-07"},{"id":"e3","text":"The pediatric protocol lane still uses a 5 minute reconsent deviation escalation timeout.","tags":["reconsent-deviation-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"An ethics governance proposal recommended a 45 minute escalation timeout, but that was not adopted.","tags":["reconsent-deviation-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-08"},{"id":"e5","text":"The research dashboard separates reconsent escalation timers from missed follow-up alerts.","tags":["reconsent-deviation-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Research ethics state case with pediatric lane override and narrow minute-level alternatives."}
{"id":"speaker-hard-037","query":"What did you suggest for making entity cases harder without making the names absurd?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using names that are genuinely confusable in realistic organizations, then pairing them with overlapping habits so the wrong memory still sounds human.","tags":["harder-entity-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested inventing bizarre fantasy names because that would make collisions easier to control.","tags":["harder-entity-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested varying the work context around the same object or snack so the system has to retrieve both the right person and the right situation.","tags":["harder-entity-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later saw that realistic near-name collisions work better when the adjacent routine is plausible instead of random.","tags":["harder-entity-cases","discussion"],"depth":2},{"id":"e5","text":"Many of the newest entity cases now pair near names with the same item in the wrong neighboring workflow.","tags":["harder-entity-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about believable near-name confusion and contextual overlap."}
{"id":"speaker-hard-038","query":"What did you suggest for making environment cases harder than simple prod-vs-staging swaps?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested mixing preview, sandbox, lane-specific neighbors, and stale host aliases so the right environment can be confused with several operationally adjacent systems.","tags":["harder-environment-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using only prod, staging, and dev because that would already cover environment confusion.","tags":["harder-environment-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested including nearby export, print, or hold systems because retrieval often grabs the right neighborhood but the wrong exact host.","tags":["harder-environment-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that stale aliases and neighboring subsystems are often stronger distractors than another full environment.","tags":["harder-environment-cases","discussion"],"depth":2},{"id":"e5","text":"The latest environment cases now intentionally include host-family neighbors and older naming variants.","tags":["harder-environment-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about richer environment confusion through host families, stale aliases, and adjacent systems."}
{"id":"speaker-hard-039","query":"What did you suggest for keeping 500 cases diverse without losing pressure?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested expanding across many institutions and workflows while preserving the same failure modes so diversity increases without flattening the benchmark's adversarial structure.","tags":["diverse-with-pressure","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested changing the adversary labels every time the domain changed so the cases would feel fresher.","tags":["diverse-with-pressure","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping the cases inspectable and dense so the variety comes from operational texture rather than schema churn.","tags":["diverse-with-pressure","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later confirmed that domain variety and adversary consistency reinforce each other when the distractors stay realistic.","tags":["diverse-with-pressure","discussion"],"depth":2},{"id":"e5","text":"The dataset now spans many domains while still using the same six adversary types.","tags":["diverse-with-pressure","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about preserving retrieval pressure while broadening institutional diversity."}
{"id":"speaker-hard-040","query":"What did you suggest for getting to 500 cases without sliding into scripting patterns?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested writing in larger hand-built waves but keeping each row individually considered so the expansion stays diverse instead of parameterized.","tags":["toward-500-without-scripting","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested quietly templating most of the later cases because volume mattered more than individuality.","tags":["toward-500-without-scripting","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested raising difficulty by composition and realism rather than by dumping near-clone rows with swapped nouns.","tags":["toward-500-without-scripting","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later kept adding sectors and operational textures instead of repeating one repo-shaped pattern.","tags":["toward-500-without-scripting","discussion"],"depth":2},{"id":"e5","text":"The benchmark growth has come from more realistic settings, not from code-generated permutations.","tags":["toward-500-without-scripting","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case aligned with the user's goal: manual expansion without collapsing into scripts or clones."}
{"id":"speaker-hard-041","query":"What did you suggest for making environment distractors feel operationally wrong instead of random?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested choosing distractors that share the same team, data family, or document type but fail on one operational dimension like lane, role, or host purpose.","tags":["operationally-wrong-distractors","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using unrelated distractors because any extra noise would make retrieval harder.","tags":["operationally-wrong-distractors","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested preferring wrong-neighbor systems like print, export, hold, or deadletter hosts because those are exactly the mistakes retrieval systems make.","tags":["operationally-wrong-distractors","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that operationally adjacent distractors produce better failure analysis than random semantic clutter.","tags":["operationally-wrong-distractors","discussion"],"depth":2},{"id":"e5","text":"Many of the latest environment rows now include those wrong-purpose neighboring hosts on purpose.","tags":["operationally-wrong-distractors","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about constructing believable wrong-neighbor distractors instead of generic noise."}
{"id":"speaker-hard-042","query":"What did you suggest for making paraphrase cases target concept tracking instead of number matching?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested reusing the same number or confidence score in a nearby workflow so the system has to track the right concept, not just the shared value.","tags":["concept-tracking-paraphrases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested avoiding repeated numbers entirely so the correct answer would be easier to identify.","tags":["concept-tracking-paraphrases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested keeping two correct phrasings when appropriate so recall is tested separately from exact wording bias.","tags":["concept-tracking-paraphrases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later saw that same-number distractors across neighboring workflows are stronger than arbitrary wrong values.","tags":["concept-tracking-paraphrases","discussion"],"depth":2},{"id":"e5","text":"The latest paraphrase cases repeatedly use shared thresholds across different operational systems.","tags":["concept-tracking-paraphrases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about forcing concept tracking across paraphrase distractors with shared values."}
{"id":"speaker-hard-043","query":"What did you suggest for making time cases harder at the end of the dataset?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested making the pilot narrower, the draft language closer to the final query, and the docs lag realistic so several time-adjacent memories compete naturally.","tags":["harder-late-time-cases","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested just making the dates closer together because that alone would make them hard enough.","tags":["harder-late-time-cases","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested choosing pilot scopes like one lane, one product, or one subgroup so they are semantically tempting but still operationally narrower than the rollout.","tags":["harder-late-time-cases","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later found that narrow pilots with lexically close wording often outrank the real rollout in retrieval.","tags":["harder-late-time-cases","discussion"],"depth":2},{"id":"e5","text":"The latest time cases now repeatedly use that narrower-pilot pattern on purpose.","tags":["harder-late-time-cases","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about tightening time adversaries through lexically tempting narrow pilots and realistic docs lag."}
{"id":"speaker-hard-044","query":"What did you suggest for making state cases compete on tiny but consequential differences?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested using small but meaningful differences in minutes, hours, or business-day cadences because those are easy for retrieval systems to blur together.","tags":["tiny-consequential-state-differences","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested using bigger jumps so the current value would stand out more clearly.","tags":["tiny-consequential-state-differences","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested pairing those small gaps with an override lane and a proposal so there are multiple believable wrong answers instead of just one stale value.","tags":["tiny-consequential-state-differences","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later saw that narrow numeric spacing makes state-update retrieval more brittle and more realistic.","tags":["tiny-consequential-state-differences","discussion"],"depth":2},{"id":"e5","text":"The newest state rows now consistently use small numeric gaps and lane-specific overrides.","tags":["tiny-consequential-state-differences","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Harder speaker case about using tiny but consequential numeric differences in state-update adversaries."}
{"id":"paraphrase-hard-037","query":"What confidence is required before a berth note is auto-tagged as hazardous-priority?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A berth note is auto-tagged as hazardous-priority only at 0.92 confidence or higher.","tags":["berth-notes","hazardous-priority","confidence"],"depth":1},{"id":"e2","text":"A berth note is auto-tagged as hazardous-priority only at 0.82 confidence or higher.","tags":["berth-notes","hazardous-priority","confidence"],"depth":1},{"id":"e3","text":"The hazardous-priority tagger needs at least 0.92 confidence before it auto-labels a berth note.","tags":["berth-notes","hazardous-priority","confidence"],"depth":1},{"id":"e4","text":"A 0.92 score is also enough to auto-tag specimen hazard flags, but that belongs to a different classifier.","tags":["specimen-hazards","confidence"],"depth":2},{"id":"e5","text":"Below-threshold hazardous-priority guesses are retained as suggestions for harbor review.","tags":["berth-notes","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Port paraphrase case with same-threshold distractor from laboratory hazard tagging."}
{"id":"paraphrase-hard-038","query":"How many transfer note gaps trigger an academic exception board?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An academic exception board is triggered once 6 transfer note gaps cluster together.","tags":["transfer-note-gaps","exception-board","threshold"],"depth":1},{"id":"e2","text":"An academic exception board is triggered once 4 transfer note gaps cluster together.","tags":["transfer-note-gaps","exception-board","threshold"],"depth":1},{"id":"e3","text":"Six transfer note gaps attached to one student file are enough to open an academic exception board.","tags":["transfer-note-gaps","exception-board","threshold"],"depth":1},{"id":"e4","text":"Six berth scheduling conflicts are enough to open a harbor reassignment review, but that is a different workflow.","tags":["berth-conflicts","threshold"],"depth":2},{"id":"e5","text":"The exception-board threshold is checked after duplicate note merging, not before it.","tags":["transfer-note-gaps","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Registrar paraphrase case with same-number distractor from port scheduling reviews."}
{"id":"paraphrase-hard-039","query":"What is the default delay before an unsigned irrigation variance memo is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned irrigation variance memo is escalated after a default delay of 6 business days.","tags":["irrigation-variances","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned irrigation variance memo is escalated after a default delay of 3 business days.","tags":["irrigation-variances","escalation","delay"],"depth":1},{"id":"e3","text":"The default irrigation variance workflow waits six business days before escalating an unsigned memo.","tags":["irrigation-variances","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned turnout bypass memos also wait six business days before oversight review, but that belongs to a different workflow.","tags":["turnout-bypasses","delay"],"depth":2},{"id":"e5","text":"A field supervisor can shorten the delay manually for irrigation emergencies.","tags":["irrigation-variances","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Agriculture paraphrase case with same-duration transit oversight distractor."}
{"id":"paraphrase-hard-040","query":"What is the default hold period before an unsigned fare waiver note is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned fare waiver note is escalated after a default hold period of 4 business days.","tags":["fare-waivers","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned fare waiver note is escalated after a default hold period of 2 business days.","tags":["fare-waivers","escalation","delay"],"depth":1},{"id":"e3","text":"The default fare waiver workflow waits four business days before escalating an unsigned note.","tags":["fare-waivers","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned transfer appeal notes also wait four business days before committee review, but that belongs to a different workflow.","tags":["transfer-appeals","delay"],"depth":2},{"id":"e5","text":"A hearings manager can waive the hold period only for documented service interruption cases.","tags":["fare-waivers","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Transit paraphrase case with same-duration registrar committee distractor."}
{"id":"paraphrase-hard-041","query":"What confidence is required before a rehearing packet is auto-tagged as detention-priority?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A rehearing packet is auto-tagged as detention-priority only at 0.9 confidence or higher.","tags":["rehearing-packets","detention-priority","confidence"],"depth":1},{"id":"e2","text":"A rehearing packet is auto-tagged as detention-priority only at 0.8 confidence or higher.","tags":["rehearing-packets","detention-priority","confidence"],"depth":1},{"id":"e3","text":"The detention-priority tagger needs at least 0.9 confidence before it auto-labels a rehearing packet.","tags":["rehearing-packets","detention-priority","confidence"],"depth":1},{"id":"e4","text":"A 0.9 score is also enough to auto-tag election takedown packets, but that belongs to a different classifier.","tags":["takedown-packets","confidence"],"depth":2},{"id":"e5","text":"Below-threshold detention-priority guesses are retained as suggestions for attorney review.","tags":["rehearing-packets","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Legal aid paraphrase case with same-threshold media/legal distractor."}
{"id":"paraphrase-hard-042","query":"How many redaction misses trigger a grant compliance review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A grant compliance review is triggered once 5 redaction misses cluster together.","tags":["redaction-misses","grant-compliance","threshold"],"depth":1},{"id":"e2","text":"A grant compliance review is triggered once 3 redaction misses cluster together.","tags":["redaction-misses","grant-compliance","threshold"],"depth":1},{"id":"e3","text":"Five redaction misses attached to one report are enough to open a grant compliance review.","tags":["redaction-misses","grant-compliance","threshold"],"depth":1},{"id":"e4","text":"Five rights clearance misses are enough to open an archive escalation review, but that is a different workflow.","tags":["rights-misses","threshold"],"depth":2},{"id":"e5","text":"The grant compliance threshold is checked after duplicate report merging, not before it.","tags":["redaction-misses","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Grant paraphrase case with same-number archive-rights distractor."}
{"id":"paraphrase-hard-043","query":"What is the default delay before an unsigned coding variance memo is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned coding variance memo is escalated after a default delay of 7 business days.","tags":["coding-variances","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned coding variance memo is escalated after a default delay of 4 business days.","tags":["coding-variances","escalation","delay"],"depth":1},{"id":"e3","text":"The default coding variance workflow waits seven business days before escalating an unsigned memo.","tags":["coding-variances","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned rights clearance memos also wait seven business days before board review, but that belongs to a different workflow.","tags":["rights-clearances","delay"],"depth":2},{"id":"e5","text":"A trial chair can shorten the delay manually for serious adverse event coding disputes.","tags":["coding-variances","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Research administration paraphrase case with same-duration rights-clearance distractor."}
{"id":"paraphrase-hard-044","query":"What is the default hold period before an unsigned takedown reversal is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned takedown reversal is escalated after a default hold period of 5 business days.","tags":["takedown-reversals","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned takedown reversal is escalated after a default hold period of 2 business days.","tags":["takedown-reversals","escalation","delay"],"depth":1},{"id":"e3","text":"The default takedown reversal workflow waits five business days before escalating an unsigned reversal.","tags":["takedown-reversals","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned reconsent deviations also wait five business days before ethics review, but that belongs to a different workflow.","tags":["reconsent-deviations","delay"],"depth":2},{"id":"e5","text":"A standards editor can waive the hold period only for documented defamation risks.","tags":["takedown-reversals","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Media/legal paraphrase case with same-duration research ethics distractor."}
{"id":"paraphrase-hard-045","query":"What confidence is required before a reunification packet is auto-tagged as sibling-priority?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A reunification packet is auto-tagged as sibling-priority only at 0.87 confidence or higher.","tags":["reunification-packets","sibling-priority","confidence"],"depth":1},{"id":"e2","text":"A reunification packet is auto-tagged as sibling-priority only at 0.77 confidence or higher.","tags":["reunification-packets","sibling-priority","confidence"],"depth":1},{"id":"e3","text":"The sibling-priority tagger needs at least 0.87 confidence before it auto-labels a reunification packet.","tags":["reunification-packets","sibling-priority","confidence"],"depth":1},{"id":"e4","text":"A 0.87 score is also enough to auto-tag fare dispute family bundles, but that belongs to a different classifier.","tags":["fare-disputes","confidence"],"depth":2},{"id":"e5","text":"Below-threshold sibling-priority guesses are retained as suggestions for case review.","tags":["reunification-packets","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Social services paraphrase case with same-threshold transit-family distractor."}
{"id":"paraphrase-hard-046","query":"How many hedge variance notes trigger a market controls review?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A market controls review is triggered once 6 hedge variance notes cluster together.","tags":["hedge-variance-notes","market-controls","threshold"],"depth":1},{"id":"e2","text":"A market controls review is triggered once 4 hedge variance notes cluster together.","tags":["hedge-variance-notes","market-controls","threshold"],"depth":1},{"id":"e3","text":"Six hedge variance notes attached to one book are enough to open a market controls review.","tags":["hedge-variance-notes","market-controls","threshold"],"depth":1},{"id":"e4","text":"Six chargeback evidence gaps are enough to open a dispute escalation review, but that is a different workflow.","tags":["chargeback-gaps","threshold"],"depth":2},{"id":"e5","text":"The market controls threshold is checked after duplicate note merging, not before it.","tags":["hedge-variance-notes","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Energy markets paraphrase case with same-number payments distractor and duplicate-merging context."}
{"id":"paraphrase-hard-047","query":"What is the default delay before an unsigned rights clearance packet is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned rights clearance packet is escalated after a default delay of 9 business days.","tags":["rights-clearances","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned rights clearance packet is escalated after a default delay of 6 business days.","tags":["rights-clearances","escalation","delay"],"depth":1},{"id":"e3","text":"The default rights clearance workflow waits nine business days before escalating an unsigned packet.","tags":["rights-clearances","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned grant reconsideration packets also wait nine business days before board review, but that belongs to a different workflow.","tags":["reconsideration-packets","delay"],"depth":2},{"id":"e5","text":"A rights editor can shorten the delay manually for active injunction deadlines.","tags":["rights-clearances","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Rights management paraphrase case with same-duration advancement-board distractor."}
{"id":"paraphrase-hard-048","query":"What is the default hold period before an unsigned fare dispute packet is escalated?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"An unsigned fare dispute packet is escalated after a default hold period of 6 business days.","tags":["fare-disputes","escalation","delay"],"depth":1},{"id":"e2","text":"An unsigned fare dispute packet is escalated after a default hold period of 3 business days.","tags":["fare-disputes","escalation","delay"],"depth":1},{"id":"e3","text":"The default fare dispute workflow waits six business days before escalating an unsigned packet.","tags":["fare-disputes","escalation","delay"],"depth":1},{"id":"e4","text":"Unsigned berth reassignment packets also wait six business days before harbor review, but that belongs to a different workflow.","tags":["berth-reassignments","delay"],"depth":2},{"id":"e5","text":"A hearings manager can waive the hold period only for documented service outage claims.","tags":["fare-disputes","ops"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Transit paraphrase case with same-duration harbor review distractor."}
{"id":"paraphrase-hard-049","query":"What confidence is required before a consent variance note is auto-tagged as immediate-risk?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"A consent variance note is auto-tagged as immediate-risk only at 0.93 confidence or higher.","tags":["consent-variances","immediate-risk","confidence"],"depth":1},{"id":"e2","text":"A consent variance note is auto-tagged as immediate-risk only at 0.83 confidence or higher.","tags":["consent-variances","immediate-risk","confidence"],"depth":1},{"id":"e3","text":"The immediate-risk tagger needs at least 0.93 confidence before it auto-labels a consent variance note.","tags":["consent-variances","immediate-risk","confidence"],"depth":1},{"id":"e4","text":"A 0.93 score is also enough to auto-tag hazardous berth exceptions, but that belongs to a different classifier.","tags":["berth-exceptions","confidence"],"depth":2},{"id":"e5","text":"Below-threshold immediate-risk guesses are retained as suggestions for chair review.","tags":["consent-variances","pipeline"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Research ethics paraphrase case with same-threshold harbor operations distractor."}
{"id":"env-hard-050","query":"What dispatch host does flood-preview use for levee gate rehearsal alerts?","adversary_type":"environment_swap","entries":[{"id":"e1","text":"The flood-preview environment uses dispatch-flood-preview.internal for levee gate rehearsal alerts.","tags":["flood-preview","levee-gate-alerts","dispatch-host","infra"],"depth":1},{"id":"e2","text":"The production environment uses dispatch-water-prod.internal for levee gate rehearsal alerts.","tags":["production","levee-gate-alerts","dispatch-host","infra"],"depth":1},{"id":"e3","text":"The pumps-sandbox environment uses dispatch-pumps-sandbox.internal for levee gate rehearsal alerts.","tags":["pumps-sandbox","levee-gate-alerts","dispatch-host","infra"],"depth":2},{"id":"e4","text":"Flood-preview renders incident maps through maps-flood-preview.internal, which is not the dispatch host.","tags":["flood-preview","levee-gate-alerts","maps"],"depth":2},{"id":"e5","text":"A district memo still lists preview-levee-dispatch.internal from before hostnames were standardized.","tags":["flood-preview","levee-gate-alerts","historical"],"depth":3,"timestamp":"2025-09-18"}],"relevant_ids":["e1"],"notes":"Additional environment case with dispatch-vs-map host confusion and a stale alias."}
{"id":"entity-hard-050","query":"What snack does Ilya bring to levee gate incident boards?","adversary_type":"entity_swap","entries":[{"id":"e1","text":"Ilya brings roasted edamame to levee gate incident boards.","tags":["ilya","levee-gate-incidents","preference"],"depth":1},{"id":"e2","text":"Ilia brings roasted edamame to levee gate incident boards.","tags":["ilia","levee-gate-incidents","preference"],"depth":1},{"id":"e3","text":"Ilya brings dried apples to pump maintenance reviews instead of levee gate incident boards.","tags":["ilya","pump-maintenance","preference"],"depth":2},{"id":"e4","text":"Alya brings roasted edamame to flood permit reviews, not levee gate incident boards.","tags":["alya","flood-permits","preference"],"depth":2},{"id":"e5","text":"A flood control snack note said Ilya and Ilia both preferred clean dry snacks during overnight boards.","tags":["ilya","ilia","snack-note"],"depth":3}],"relevant_ids":["e1"],"notes":"Additional entity case with transliterated-name collision and neighboring flood workflow confusion."}
{"id":"time-hard-050","query":"When did we require deputy chief approval on levee gate overrides?","adversary_type":"time_swap","entries":[{"id":"e1","text":"We require deputy chief approval on levee gate overrides in October.","tags":["timeline","levee-gate-overrides","deputy-chief-approval"],"depth":1,"timestamp":"2026-10-23"},{"id":"e2","text":"We proposed deputy chief approval on levee gate overrides in August.","tags":["planning","levee-gate-overrides","deputy-chief-approval"],"depth":2,"timestamp":"2026-08-27"},{"id":"e3","text":"A storm-lane pilot required deputy chief approval in September for night overrides only.","tags":["trial","levee-gate-overrides","deputy-chief-approval"],"depth":2,"timestamp":"2026-09-12"},{"id":"e4","text":"The flood operations manual was updated in November after deputy chief approval became required.","tags":["docs","levee-gate-overrides","deputy-chief-approval"],"depth":2,"timestamp":"2026-11-05"},{"id":"e5","text":"The old supervisor-only override path was removed in December.","tags":["cleanup","levee-gate-overrides","deputy-chief-approval"],"depth":3,"timestamp":"2026-12-06"}],"relevant_ids":["e1"],"notes":"Additional time case with a night-only pilot preceding the broader levee override rule."}
{"id":"state-hard-050","query":"What is the current timeout before an unreviewed levee alert escalates?","adversary_type":"state_update","entries":[{"id":"e1","text":"The current timeout before an unreviewed levee alert escalates is 27 minutes.","tags":["levee-alert-escalation-timeout","current"],"depth":0,"timestamp":"2026-04-10"},{"id":"e2","text":"The timeout before an unreviewed levee alert escalates used to be 12 minutes.","tags":["levee-alert-escalation-timeout","historical"],"depth":2,"timestamp":"2025-11-16"},{"id":"e3","text":"The storm lane still uses a 4 minute levee alert escalation timeout.","tags":["levee-alert-escalation-timeout","preview"],"depth":2,"timestamp":"2026-04-10"},{"id":"e4","text":"A flood governance proposal recommended a 35 minute escalation timeout, but that was not adopted.","tags":["levee-alert-escalation-timeout","proposal"],"depth":2,"timestamp":"2026-03-15"},{"id":"e5","text":"The flood dashboard separates levee alert escalation timers from pump dispatch overdue timers.","tags":["levee-alert-escalation-timeout","ops"],"depth":3,"timestamp":"2026-04-10"}],"relevant_ids":["e1"],"notes":"Additional state case with narrow minute alternatives and storm-lane override confusion."}
{"id":"speaker-hard-045","query":"What did you suggest for the last stretch to 500 cases?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"You suggested finishing with a few extra hand-written rows that preserve the same density and realism rather than padding the count with filler.","tags":["last-stretch-to-500","assistant","suggestion"],"depth":1,"speaker":"assistant"},{"id":"e2","text":"I suggested adding whatever five easiest rows would get the count over the line.","tags":["last-stretch-to-500","user","suggestion"],"depth":1,"speaker":"user"},{"id":"e3","text":"You also suggested using those last rows to reinforce the benchmark's core idea: believable wrong neighbors, not arbitrary noise.","tags":["last-stretch-to-500","assistant","follow-up"],"depth":1,"speaker":"assistant"},{"id":"e4","text":"We later used the final rows to tighten the count while keeping the adversarial style intact.","tags":["last-stretch-to-500","discussion"],"depth":2},{"id":"e5","text":"The final dataset keeps the same six adversary types and hand-inspectable structure at 500 cases.","tags":["last-stretch-to-500","notes"],"depth":3}],"relevant_ids":["e1","e3"],"notes":"Additional speaker case documenting the choice to finish with real dense cases rather than filler."}
{"id":"time-precision-001","query":"What is the exact date the settlement exception policy took effect?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Compliance memo, annotated after the rollout call and then re-circulated before audit prep: several people kept calling the settlement exception change a spring rollout, the board deck filed it under the broader Q2 controls cleanup, and one witness later testified only that it happened in 2026 before the grace-period language disappeared; however, the signed operations bulletin, which is the controlling record for effective dates, states that the policy took effect on 2026-04-17, eight business days before the related FAQ rewrite and nearly two weeks before the old approval carve-out was removed from the runbook.","tags":["settlement-exception-policy","exact-date","timeline"],"depth":1,"timestamp":"2026-04-17"},{"id":"e2","text":"Finance all-hands summary: the settlement exception policy change was packaged into the 2026 Q2 cleanup bundle with reserve review tuning, exception queue pruning, and two smaller audit fixes, so most people later remembered it only as 'that Q2 controls work' rather than as a dated policy event.","tags":["settlement-exception-policy","quarter","timeline"],"depth":2,"timestamp":"2026-04-01"},{"id":"e3","text":"Slack recap from the controller channel: 'that settlement exception change was a spring thing, right?' followed by a reply saying 'yeah, before the late-April runbook pass, definitely sometime in spring,' three thumbs-up, and no one bothering to paste the actual effective date from the bulletin.","tags":["settlement-exception-policy","season","chat"],"depth":2},{"id":"e4","text":"Court-style hearing prep note: counsel kept referring to the policy as a 2026 change because the witness only remembered the year, the season, and the fact that it predated a separate reconciliation update; the prep team highlighted that this year-level summary was accurate but too coarse for any question asking for the true effective date.","tags":["settlement-exception-policy","year","dialogue"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Follow-up runbook edit: the old grace-period language was removed on 2026-04-29 after a separate documentation review, which is close enough to muddy memory but still later than the actual effective date recorded in the signed bulletin and implementation ticket.","tags":["settlement-exception-policy","docs","cleanup"],"depth":3,"timestamp":"2026-04-29"}],"relevant_ids":["e1"],"notes":"Precision-focused time case: year, season, quarter, and exact date are all present, but only the exact effective date answers the query."}
{"id":"time-precision-002","query":"What is the exact date the orbital telemetry archive switched to immutable filenames?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Archive migration thread, reconstructed later for the reliability review: people loosely remembered the orbital telemetry archive switch as a late-winter change, one planning board grouped it under Q1, an engineer later called it 'that winter archive hardening pass,' and the docs team confused it with the March handbook rewrite, but the cutover ticket and storage-change approval are aligned that immutable filenames began on 2026-02-23, which was before the handbook update and after the checksum dry run.","tags":["orbital-telemetry-archive","exact-date","timeline"],"depth":1,"timestamp":"2026-02-23"},{"id":"e2","text":"Release planning note: the archive-filename work was tracked as a Q1 reliability item and bundled with checksum verification, retention cleanup, and a replay index sanity pass, so anyone reading only the planning board would remember the quarter correctly but not the exact day.","tags":["orbital-telemetry-archive","quarter","timeline"],"depth":2,"timestamp":"2026-01-01"},{"id":"e3","text":"Crew chat transcript: 'Was that immutable archive thing winter or early spring?' 'Definitely winter-ish, before the March docs refresh and after the checksum dry run.' The recollection is directionally true, but the thread never cites the actual date.","tags":["orbital-telemetry-archive","season","chat"],"depth":2},{"id":"e4","text":"Audit statement: the archive naming change occurred in 2026, though the auditor's summary deliberately omitted the precise date because the report focused on controls, retention boundaries, and evidence of immutability rather than rollout chronology.","tags":["orbital-telemetry-archive","year","audit"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Docs maintenance note: the operator handbook was updated on 2026-03-02, several days after the immutable-filename cutover itself, which is why later readers sometimes misremember the documentation date as the operational date.","tags":["orbital-telemetry-archive","docs","cleanup"],"depth":3,"timestamp":"2026-03-02"}],"relevant_ids":["e1"],"notes":"Precision-focused time case with exact day versus quarter, season, and year references in noisy rollout prose."}
{"id":"time-precision-003","query":"What is the exact date the witness portal began requiring second-factor review for sealed exhibits?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Courtroom support transcript: one clerk remembered that second-factor review for sealed exhibits started 'sometime in the summer calendar,' another said 'around Q3 kickoff,' but the signed portal bulletin gives the exact effective date as 2026-07-08.","tags":["sealed-exhibits-2fa-review","exact-date","timeline"],"depth":1,"timestamp":"2026-07-08"},{"id":"e2","text":"Operations briefing deck: the sealed-exhibit review tightening sat inside the Q3 witness portal hardening package.","tags":["sealed-exhibits-2fa-review","quarter","timeline"],"depth":2,"timestamp":"2026-07-01"},{"id":"e3","text":"Text message history from the courtroom deputies: 'That was definitely a summer change, because it broke the holiday calendar prep.'","tags":["sealed-exhibits-2fa-review","season","chat"],"depth":2},{"id":"e4","text":"General legal ops note: the witness portal changed its sealed-exhibit review flow in 2026, but the note collapsed all portal changes into a year-level summary.","tags":["sealed-exhibits-2fa-review","year","ops"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Portal documentation update: screenshots showing the second-factor review step were refreshed on 2026-07-19, after the actual effective date.","tags":["sealed-exhibits-2fa-review","docs","cleanup"],"depth":3,"timestamp":"2026-07-19"}],"relevant_ids":["e1"],"notes":"Precision-focused time case with exact day hidden among multiple coarser but true descriptions."}
{"id":"time-precision-004","query":"What is the exact date the museum loan desk stopped auto-approving outbound packing exceptions?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Collections manager note: staff kept describing the outbound packing exception change as an autumn policy shift or just 'the Q4 handling update,' but the signed loan-desk notice says auto-approval ended on 2026-10-14.","tags":["outbound-packing-exceptions","exact-date","timeline"],"depth":1,"timestamp":"2026-10-14"},{"id":"e2","text":"Quarterly operations summary: outbound packing controls were part of the Q4 collections-risk package with crate logging and condition-photo rules.","tags":["outbound-packing-exceptions","quarter","timeline"],"depth":2,"timestamp":"2026-10-01"},{"id":"e3","text":"Registrar text thread: 'That was a fall change, right? Somewhere between the donor show and the winter loan prep.'","tags":["outbound-packing-exceptions","season","chat"],"depth":2},{"id":"e4","text":"Board briefing language: the loan desk approval model changed in 2026, but the board packet intentionally summarized all handling changes by year only.","tags":["outbound-packing-exceptions","year","board"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Manual cleanup note: the old auto-approval FAQ entry was removed on 2026-10-31, after the policy had already been in force for over two weeks.","tags":["outbound-packing-exceptions","docs","cleanup"],"depth":3,"timestamp":"2026-10-31"}],"relevant_ids":["e1"],"notes":"Precision-focused time case with exact date competing against true quarter, season, and year references."}
{"id":"time-precision-005","query":"What is the exact date the flood control board required deputy sign-off on overnight gate reversals?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Incident review transcript: everyone remembered deputy sign-off as a storm-season tightening and one slide called it a Q4 resilience rule, but the formal flood board order made it effective on 2026-11-06.","tags":["overnight-gate-reversals","exact-date","timeline"],"depth":1,"timestamp":"2026-11-06"},{"id":"e2","text":"Seasonal controls memo: overnight gate reversal approvals were grouped into the fall flood-readiness package.","tags":["overnight-gate-reversals","season","timeline"],"depth":2},{"id":"e3","text":"Quarterly planning tracker: the deputy sign-off requirement sits under Q4 emergency-control changes.","tags":["overnight-gate-reversals","quarter","timeline"],"depth":2,"timestamp":"2026-10-01"},{"id":"e4","text":"Public hearing summary: the board referred to the change as a 2026 procedural tightening without naming the exact effective day.","tags":["overnight-gate-reversals","year","hearing"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Operations handbook note: the old supervisor-only language was removed on 2026-11-18, which lagged behind the real policy start.","tags":["overnight-gate-reversals","docs","cleanup"],"depth":3,"timestamp":"2026-11-18"}],"relevant_ids":["e1"],"notes":"Precision-focused time case with multiple true coarse summaries surrounding the exact effective date."}
{"id":"time-precision-006","query":"What is the exact date the energy desk began manual review of cross-day hedge mismatches?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Trading desk debrief: traders kept recalling manual review for cross-day hedge mismatches as a late-summer control or simply 'that Q3 cleanup,' but the signed controls bulletin says it began on 2026-09-03.","tags":["cross-day-hedge-mismatches","exact-date","timeline"],"depth":1,"timestamp":"2026-09-03"},{"id":"e2","text":"Risk controls calendar: cross-day mismatch handling was packaged under the Q3 review-tightening slate.","tags":["cross-day-hedge-mismatches","quarter","timeline"],"depth":2,"timestamp":"2026-07-01"},{"id":"e3","text":"Trader chat: 'That mismatch thing was basically a summer fix; I remember because it landed before October books.'","tags":["cross-day-hedge-mismatches","season","chat"],"depth":2},{"id":"e4","text":"Annual controls summary: mismatch review rules were updated in 2026, with no date granularity beyond the year in the summary packet.","tags":["cross-day-hedge-mismatches","year","controls"],"depth":3,"timestamp":"2026-01-01"},{"id":"e5","text":"Desk manual update: the wording on automatic closures was removed on 2026-09-22, after manual review had already become required.","tags":["cross-day-hedge-mismatches","docs","cleanup"],"depth":3,"timestamp":"2026-09-22"}],"relevant_ids":["e1"],"notes":"Precision-focused time case with exact date surrounded by several true but coarser temporal formulations."}
{"id":"paraphrase-precision-001","query":"What was the exact reimbursement amount on the bookstore charge?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Expense thread, copied from the receipt image after tax and then argued over in a follow-up email: people kept saying the bookstore reimbursement was about twenty dollars, one text rounded it to 'basically twenty,' and the ledger grouping note called it a sub-twenty office-supply reimbursement, but the photographed receipt, the corrected reimbursement line, and the final posted entry all agree that the exact amount was $19.75, not the rounded figure repeated in conversation.","tags":["bookstore-reimbursement","exact-amount","finance"],"depth":1},{"id":"e2","text":"Slack message from accounting: 'That bookstore reimbursement was around $20, not enough to trigger any review threshold, and small enough that we grouped it with the office-supply reimbursements on the recap sheet.' That shorthand is operationally useful but still less precise than the actual posted amount.","tags":["bookstore-reimbursement","approximate-amount","chat"],"depth":2},{"id":"e3","text":"Ledger annotation: the bookstore charge settled below twenty dollars, sat in the same bucket as the other minor office purchases, and kept getting summarized as a twenty-dollar item in verbal updates even though the cents mattered for exact reconciliation.","tags":["bookstore-reimbursement","rounded-description","ledger"],"depth":2},{"id":"e4","text":"Expense policy note: reimbursement lines under $20 were often discussed casually as twenty-dollar items, especially when staff were triaging many small receipts at once, which is exactly how rounded but technically imprecise values ended up in the meeting recap.","tags":["bookstore-reimbursement","policy-context"],"depth":3},{"id":"e5","text":"Follow-up email: the clerk corrected the draft ledger from '$20' to '$19.75' after re-reading the photographed receipt, confirming the tax-inclusive total, and noting that the rounded shorthand in chat had already started to propagate into the reimbursement summary.","tags":["bookstore-reimbursement","exact-amount","email"],"depth":3}],"relevant_ids":["e1","e5"],"notes":"Precision-focused amount case where multiple coarser monetary descriptions are true enough in casual conversation, but only the exact cents answer the query."}
{"id":"paraphrase-precision-002","query":"What was the exact disputed lunch charge?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Text exchange between the paralegal and clerk: everybody referred to the disputed lunch charge as 'about thirty bucks' while they were sorting receipts, but the actual line item under dispute was $29.42.","tags":["lunch-charge","exact-amount","finance"],"depth":1},{"id":"e2","text":"Expense summary sheet: the lunch charge sat in the roughly-thirty-dollar range and was grouped with two other meal receipts from the same hearing day.","tags":["lunch-charge","approximate-amount","summary"],"depth":2},{"id":"e3","text":"Court reimbursement dialogue: 'Was that lunch twenty-nine-something or just call it thirty?' 'For the hearing memo we rounded it to about thirty.'","tags":["lunch-charge","rounded-dialogue","courtroom"],"depth":2},{"id":"e4","text":"Clerk correction note: the exact disputed lunch charge was $29.42, not the rounded thirty-dollar shorthand used in the meeting recap.","tags":["lunch-charge","exact-amount","correction"],"depth":3},{"id":"e5","text":"Policy excerpt: minor meal disputes below $30 were frequently summarized informally by rounded dollar amounts in status calls.","tags":["lunch-charge","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Precision-focused financial case with approximate and rounded forms that are all conversationally true, but only the exact amount is correct."}
{"id":"paraphrase-precision-003","query":"What was the exact transfer amount, not the rounded amount?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Bank operations hearing note: everyone on the call kept referring to the transfer as 'about twenty thousand' because the exact cents felt irrelevant in conversation, but the posted transfer amount was $19,975.18.","tags":["transfer-amount","exact-amount","banking"],"depth":1},{"id":"e2","text":"Treasury recap: the transfer was in the roughly-$20k range and therefore below the escalation threshold used for larger wires.","tags":["transfer-amount","approximate-amount","treasury"],"depth":2},{"id":"e3","text":"Text message from the analyst: 'Call it twenty grand in the summary, but I think the actual posting was a hair under that.'","tags":["transfer-amount","rounded-dialogue","chat"],"depth":2},{"id":"e4","text":"Ledger correction: the exact transfer amount was $19,975.18 once the fee reversal and final posting settled on the same day.","tags":["transfer-amount","exact-amount","ledger"],"depth":3},{"id":"e5","text":"Controls note: the rounded figure in the review packet was intentionally simplified for readability and was never intended as the exact ledger value.","tags":["transfer-amount","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Precision-focused transfer case where rough scale, rounded shorthand, and the exact ledger value all coexist."}
{"id":"paraphrase-precision-004","query":"What was the exact distance recorded between the probe and the relay satellite?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Mission control transcript, later copied into the anomaly review: engineers kept talking about the probe being 'about eight thousand miles out' or 'roughly eight-point-two thousand,' public comms rounded that down for readability, and one engineering note described the spacing as merely low-eight-thousand range, but the telemetry sheet used for the actual handoff report logged the exact probe-to-relay distance as 8,214 miles, which is the figure repeated in the correction record.","tags":["probe-relay-distance","exact-distance","telemetry"],"depth":1},{"id":"e2","text":"Public comms draft: the probe was described as roughly eight thousand miles from the relay satellite so the update would be readable to a general audience, which is true in the broad sense but intentionally less precise than the telemetry value used by operations.","tags":["probe-relay-distance","approximate-distance","public-comms"],"depth":2},{"id":"e3","text":"Engineering note: the relay spacing sat in the low eight-thousand-mile range during the handoff window, and several internal summaries reused that broad range language when the exact number was not operationally necessary for the audience.","tags":["probe-relay-distance","range-description","engineering"],"depth":2},{"id":"e4","text":"Telemetry correction line: the precise probe-to-relay distance was 8,214 miles at the timestamp used for the handoff report, even though several later summaries rounded it to the nearest thousand or compressed it into a range description.","tags":["probe-relay-distance","exact-distance","correction"],"depth":3},{"id":"e5","text":"Mission summary note: distances in the public deck were rounded to the nearest thousand unless exact spacing was operationally necessary, which is why multiple rough numbers can all appear in the record even though only one is the exact measured value.","tags":["probe-relay-distance","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Precision-focused measurement case where rough thousand-mile descriptions are true at one granularity, but only the exact mileage is correct."}
{"id":"paraphrase-precision-005","query":"What was the exact measured separation between the observatory pair?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Observatory planning call: scientists kept saying the pair were 'roughly twelve thousand miles apart' and one witness later summarized it as 'a little over eleven thousand,' but the measurement worksheet recorded the exact separation as 11,842 miles.","tags":["observatory-separation","exact-distance","science"],"depth":1},{"id":"e2","text":"Press note: for readability, the observatories were described as about twelve thousand miles apart.","tags":["observatory-separation","approximate-distance","press"],"depth":2},{"id":"e3","text":"Research summary: the instrument pair stayed in the high-eleven-thousand-mile range during the measurement interval.","tags":["observatory-separation","range-description","research"],"depth":2},{"id":"e4","text":"Worksheet correction: the exact measured separation between the observatory pair was 11,842 miles, not the rounded value repeated in outreach copy.","tags":["observatory-separation","exact-distance","correction"],"depth":3},{"id":"e5","text":"Outreach policy note: external-facing astronomy summaries often rounded to the nearest thousand miles unless the exact figure mattered scientifically.","tags":["observatory-separation","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Precision-focused distance case with rough public summaries, range language, and the exact scientific measurement all present."}
{"id":"paraphrase-precision-006","query":"What was the exact settlement discount, not the rounded shorthand?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Settlement call transcript: counsel kept saying the discount was 'basically fifteen percent' and one text rounded it to 'about fifteen,' but the signed calculation set the exact settlement discount at 14.62%.","tags":["settlement-discount","exact-percentage","finance"],"depth":1},{"id":"e2","text":"Negotiation memo: the discount sat in the mid-teens and was repeatedly summarized as roughly fifteen percent during the bargaining phase.","tags":["settlement-discount","approximate-percentage","memo"],"depth":2},{"id":"e3","text":"Court clerk note: 'The lawyers kept saying fifteen even, but the spreadsheet number looked slightly under.'","tags":["settlement-discount","rounded-dialogue","courtroom"],"depth":2},{"id":"e4","text":"Final spreadsheet line: the exact settlement discount was 14.62%, not the rounded fifteen percent shorthand used in conversation.","tags":["settlement-discount","exact-percentage","spreadsheet"],"depth":3},{"id":"e5","text":"Review packet note: percentage values were often rounded for slide decks, which is why the spoken summary and the exact worksheet did not match perfectly.","tags":["settlement-discount","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Precision-focused percentage case with rough conversational and packet shorthand competing against the exact calculated value."}
{"id":"court-precision-001","query":"What was the exact time the defendant entered the loading corridor?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Trial transcript excerpt, corridor-entry dispute: The prosecutor asked the night supervisor whether the defendant arrived 'around nine-thirty, maybe closer to quarter to ten,' and the supervisor answered that the whole evening blurred together because there had been a sprinkler fault, two late freight deliveries, and a radio argument between the dock team and security. Another witness, the cleaner, testified that she remembered seeing the defendant only after she had finished the mop closet at 'maybe around ten' and before the hallway lights dimmed, which she thought happened at some point in the last part of the shift. Defense counsel then pointed out that nearly every witness was using rounded language like 'around,' 'close to,' 'the late part of the night,' and 'before the second coffee run.' The decisive exhibit was not any recollection at all but the corridor camera log, which recorded the defendant crossing the threshold at exactly 2026-09-14 21:43:18. The log was cross-referenced against the phone tree because one guard had tried to say the camera clock might be off, but the facilities technician testified that the system had been synchronized that afternoon. The transcript is cluttered with talk of the quarter, the season, and whether the incident happened before or after the late September inventory push, but the only exact answer preserved in the record is the camera timestamp 2026-09-14 21:43:18.","tags":["loading-corridor-entry","exact-time","camera-log","courtroom"],"depth":1,"timestamp":"2026-09-14T21:43:18"},{"id":"e2","text":"Witness recollection summary filed by the clerk: one person said the defendant came through 'sometime in the 9:30s,' another said 'just before ten,' and a third insisted it was 'during the late evening rush when the dock team was still arguing about the sprinkler fault.' All of those descriptions were left in the summary because they were fair recollections, but none were precise enough to answer a question that asks for the exact time of entry.","tags":["loading-corridor-entry","approximate-time","witness-summary"],"depth":2},{"id":"e3","text":"Phone records note: the nearby guard placed a call at 21:44 and later testified that the defendant was already inside by then, which narrows the time and supports the idea that the entry happened shortly before the call, but the phone log still does not supersede the exact camera timestamp.","tags":["loading-corridor-entry","authoritative-neighbor","phone-log"],"depth":2,"timestamp":"2026-09-14T21:44:00"},{"id":"e4","text":"Cross-examination excerpt: defense counsel repeatedly used coarser formulations, asking whether the event happened 'that night in September,' 'late in the third quarter,' 'after nine-thirty,' or 'somewhere before ten.' The witness agreed to some of those formulations because they were all directionally true, but the court record distinguished between broadly true temporal framing and the exact logged moment.","tags":["loading-corridor-entry","coarse-time","dialogue"],"depth":3},{"id":"e5","text":"Evidence maintenance note: the hallway camera clip was exported the next morning at 2026-09-15 07:12, and that export time later confused one junior reviewer who briefly treated the export as the event time before the clerk corrected the summary.","tags":["loading-corridor-entry","cleanup","evidence-note"],"depth":3,"timestamp":"2026-09-15T07:12:00"}],"relevant_ids":["e1"],"notes":"Courtroom precision case with conflicting witness recollections, a neighboring phone call, and one controlling camera timestamp."}
{"id":"court-precision-002","query":"What was the exact amount of the disputed wire reimbursement?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Hearing transcript on the reimbursement dispute: counsel for the claimant kept referring to the reimbursement as 'basically twenty thousand dollars' because the fight was over approval authority rather than cents, while the operations manager said it was 'just under twenty grand' and the cashier, reading from memory rather than from the ledger, said she thought it was 'nineteen nine-something, maybe seventy-five, maybe just shy of eighty.' The judge repeatedly asked for an exact figure, but the oral testimony kept sliding back into rough language because every witness had learned the number from conversation, not from the statement itself. The decisive exhibit was the banking ledger, where the disputed wire reimbursement posted at exactly $19,975.18 after a same-day fee reversal. A later reconciliation email confirmed that the rough twenty-thousand-dollar shorthand used in the hearing binder, the status call, and the legal summary was tolerated because it did not affect threshold review in that conversation, but it was never the exact number. The transcript therefore contains multiple values that are all broadly true in scale, yet the only correct answer to an exact-amount question is $19,975.18.","tags":["wire-reimbursement","exact-amount","ledger","courtroom"],"depth":1},{"id":"e2","text":"Review packet summary: the reimbursement was grouped into the 'roughly $20k' cluster for management discussion, and one page rounded it up to $20,000 because the packet compared it against larger transfers where the precision to the cent did not matter for the slide deck.","tags":["wire-reimbursement","rounded-amount","packet"],"depth":2},{"id":"e3","text":"Treasury analyst text thread: 'Call it twenty grand in the hearing notes, the exact posting is a little lower, but the approval threshold question does not depend on the cents.' That line later became part of the confusion because people remembered the shorthand instead of the posted amount.","tags":["wire-reimbursement","rounded-dialogue","chat"],"depth":2},{"id":"e4","text":"Reconciliation email entered into evidence: the exact disputed wire reimbursement was $19,975.18 once the fee reversal and corrected beneficiary posting settled on the same day, and earlier rounded versions should be treated as conversational summaries only.","tags":["wire-reimbursement","exact-amount","email"],"depth":3},{"id":"e5","text":"Bank policy note: finance teams often summarize sub-threshold wires in rounded thousands during oral review, which is why witnesses can all sound credible while still failing to provide the exact ledger figure.","tags":["wire-reimbursement","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Courtroom precision case where witnesses all give defensible rounded amounts but the exact ledger value is the only correct answer."}
{"id":"court-precision-003","query":"What was the exact measured separation between the survey craft and the marker buoy?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Admiralty-style inquiry transcript: the harbor pilot testified that the survey craft was 'about twelve thousand feet off the marker, give or take,' while the deck officer said the spacing was 'closer to eleven-eight or eleven-nine' and the harbor master later summarized the whole thing as 'roughly a couple of nautical miles short of the main lane offset.' The testimony was full of practical seafaring language because the witnesses were remembering a stressful handoff in bad weather, not reciting instrument logs. The court then admitted the actual instrument capture from the survey display, which logged the measured separation between the survey craft and the marker buoy at 11,842 feet. Public-facing summaries later rounded that to twelve thousand feet because it sounded cleaner and because the exact figure was not necessary for the press statement, but the inquiry itself preserved the exact value in the instrument exhibit. The record therefore contains range language, rounded language, and near-accurate recollections that are all broadly compatible with the truth, yet only the exact instrument reading of 11,842 feet answers the precise question.","tags":["survey-craft-separation","exact-distance","instrument-log","hearing"],"depth":1},{"id":"e2","text":"Press office summary: for readability, the survey craft was described as approximately twelve thousand feet from the buoy during the disputed maneuver, and nobody in the public statement cared whether the exact reading was a little under that mark.","tags":["survey-craft-separation","approximate-distance","press"],"depth":2},{"id":"e3","text":"Harbor operations notebook: the separation sat in the high-eleven-thousand-foot range during the maneuver window, which is more precise than the press statement but still not the exact measured value on the display capture.","tags":["survey-craft-separation","range-description","ops"],"depth":2},{"id":"e4","text":"Instrument correction memo: the exact measured separation between the survey craft and the marker buoy was 11,842 feet, not the rounded twelve-thousand-foot shorthand used in the public note and several witness summaries.","tags":["survey-craft-separation","exact-distance","correction"],"depth":3},{"id":"e5","text":"Inquiry procedure note: external summaries commonly rounded navigational spacing for readability unless the exact figure was operationally or legally material, which is why the witness accounts and the press summary drift toward a cleaner but less precise number.","tags":["survey-craft-separation","policy-context"],"depth":3}],"relevant_ids":["e1","e4"],"notes":"Courtroom-style precision case with range language, rounded testimony, and one authoritative instrument reading."}
{"id":"court-precision-004","query":"What was the exact call time on the witness phone log?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Evidentiary hearing transcript: the receptionist testified that the witness called 'a little before quarter to eight,' the deputy marshal insisted it was 'closer to the top of the hour than that,' and the witness herself said she only remembered calling after sunset, before the hallway emptied, and not long after she had texted her brother that she was still waiting. Counsel spent several pages arguing over whether the recollections were consistent enough to place the call in the 7:40s or merely somewhere in the early part of the evening. The decisive exhibit was the phone provider log, which showed the outgoing call beginning at exactly 2026-03-11 19:43:27. The marshal then tried to redirect attention to the fact that the hallway camera clock showed activity at 19:44 and the desk notebook rounded the contact to 'about 7:45 p.m.,' but the court explicitly distinguished between rounded notations and the authoritative call record. The result is a record full of true-but-imprecise time descriptions such as evening, after sunset, before eight, and roughly quarter-to, yet only the exact phone log timestamp 2026-03-11 19:43:27 answers the question asked.","tags":["witness-phone-call","exact-time","phone-log","courtroom"],"depth":1,"timestamp":"2026-03-11T19:43:27"},{"id":"e2","text":"Desk notebook summary: the receptionist wrote that the witness called at about 7:45 p.m., which was close enough for note-taking but not exact enough for the later evidentiary dispute once everyone started comparing logs.","tags":["witness-phone-call","rounded-time","notebook"],"depth":2},{"id":"e3","text":"Hallway camera note: a person matching the witness stepped away from the reception desk at 19:44:02, which supports the broader timeline but is not the start time of the call itself.","tags":["witness-phone-call","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-03-11T19:44:02"},{"id":"e4","text":"Marshal testimony: he was comfortable saying the call happened in the evening before 8 p.m. and after the post-shift sweep, but he repeatedly admitted that he was reconstructing the time from memory and hallway activity rather than from the phone log.","tags":["witness-phone-call","coarse-time","testimony"],"depth":3},{"id":"e5","text":"Clerk correction: the hearing packet initially bolded 'about 7:45 p.m.' because that was the desk note, but the correction sheet later replaced it with the exact carrier timestamp once the provider record was admitted into evidence.","tags":["witness-phone-call","cleanup","correction"],"depth":3}],"relevant_ids":["e1"],"notes":"Courtroom precision case with several plausible witness descriptions and one controlling phone-log timestamp."}
{"id":"r5-breaker-001","query":"What was the exact time the storage room door badge event was logged?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Security hearing packet, final evidence section: four people kept calling the storage room badge event 'around 6:20' or 'somewhere in the twenty-past range,' and the shift lead testified that it was definitely before the 18:22 radio check and after the 18:19 corridor sweep, but the badge controller export itself shows the exact door event at 2026-05-14 18:20:41. The packet notes that the camera overlay rounded to the nearest minute in one printed screenshot, which is why several summaries drifted toward 18:21 even though the controller log remained authoritative.","tags":["storage-room-badge-event","exact-time","badge-log","courtroom"],"depth":1,"timestamp":"2026-05-14T18:20:41"},{"id":"e2","text":"Witness statement from the porter: 'I saw the light go green at about 6:20, maybe a touch after, but definitely before the radio call.' The recollection is directionally accurate yet still coarser than the authoritative badge export.","tags":["storage-room-badge-event","rounded-time","witness"],"depth":2},{"id":"e3","text":"Guard notebook entry: the storage room access was written down as 18:21 because the guard was logging multiple doors in a hurry and rounded several times to the nearest minute for convenience.","tags":["storage-room-badge-event","rounded-time","notebook"],"depth":2,"timestamp":"2026-05-14T18:21:00"},{"id":"e4","text":"Camera review memo: the hallway clip shows the subject in frame at 18:20:56, which confirms the general timing but is not the exact instant of the badge event itself.","tags":["storage-room-badge-event","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-05-14T18:20:56"},{"id":"e5","text":"Radio dispatch sheet: the nearby status call went out at 18:22:03, and multiple witnesses anchored their memory off that call instead of the actual badge timestamp.","tags":["storage-room-badge-event","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-05-14T18:22:03"},{"id":"e6","text":"Cross-examination summary: counsel repeatedly used phrases like 'during the 6:20 minute,' 'in the early evening,' and 'before 6:22,' all of which the witness accepted as broadly true while still not yielding the exact logged second.","tags":["storage-room-badge-event","coarse-time","dialogue"],"depth":3},{"id":"e7","text":"Maintenance note: the controller and camera clocks were resynchronized at 17:02 that afternoon, undercutting the defense theory that the badge event might have been several seconds off.","tags":["storage-room-badge-event","context","maintenance"],"depth":3,"timestamp":"2026-05-14T17:02:00"},{"id":"e8","text":"Cleanup note: the printed evidence bundle later captioned the event as 'about 6:21 p.m.' because the exhibits team copied the rounded notebook line instead of the raw badge export in one section header.","tags":["storage-room-badge-event","rounded-time","cleanup"],"depth":3},{"id":"e9","text":"Year-level incident summary: the storage room access dispute happened in 2026 during the spring facility investigation, which is true but far too coarse to answer an exact-time question.","tags":["storage-room-badge-event","year-season","summary"],"depth":4,"timestamp":"2026-01-01"}],"relevant_ids":["e1"],"notes":"Recall@5 breaker with many true temporal neighbors and multiple rounded-but-plausible versions of the same event."}
{"id":"r5-breaker-002","query":"What was the exact amount of the pharmacy reimbursement?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Reimbursement dispute transcript, final accounting exhibit: in conversation the pharmacy reimbursement kept getting described as 'about forty dollars,' 'just under forty,' and once as 'forty even after tax, basically,' but the corrected reimbursement ledger shows the exact amount was $39.68 after tax and before same-day batching into the sub-fifty reimbursement pool. The exhibit specifically warns that verbal summaries rounded to whole dollars because the hearing was focused on approval authority rather than cents.","tags":["pharmacy-reimbursement","exact-amount","ledger"],"depth":1},{"id":"e2","text":"Accounting chat: 'That pharmacy reimbursement was around forty bucks and nowhere near a manual review threshold.' This is operational shorthand and not the precise amount.","tags":["pharmacy-reimbursement","approximate-amount","chat"],"depth":2},{"id":"e3","text":"Expense recap page: the line sat in the high-thirty-dollar range and was bundled with two other low-value reimbursements from the same week.","tags":["pharmacy-reimbursement","range-description","summary"],"depth":2},{"id":"e4","text":"Hearing dialogue: one witness said 'thirty-nine-something,' another said 'call it forty,' and a third said 'definitely not over forty-one,' all of which were closer than random guessing but still less precise than the ledger.","tags":["pharmacy-reimbursement","rounded-dialogue","courtroom"],"depth":2},{"id":"e5","text":"Correction email: after re-reading the receipt image and the reimbursement export, finance corrected the shorthand forty-dollar reference to the exact posted amount of $39.68.","tags":["pharmacy-reimbursement","exact-amount","correction"],"depth":3},{"id":"e6","text":"Policy note: reimbursements below $40 were often talked about as forty-dollar items in calls because the threshold logic cared more about bucket than cents.","tags":["pharmacy-reimbursement","policy-context"],"depth":3},{"id":"e7","text":"Batch settlement line: the reimbursement was grouped into a $117.91 cluster with two unrelated claims, which later caused one reviewer to confuse the cluster total with the individual amount.","tags":["pharmacy-reimbursement","neighbor-amount","batching"],"depth":3},{"id":"e8","text":"Spreadsheet margin note: one reviewer briefly wrote $39.70 as a hand-rounding convenience before the formal ledger print fixed the amount at the cent level.","tags":["pharmacy-reimbursement","rounded-amount","worksheet"],"depth":3},{"id":"e9","text":"Manager summary: the disputed pharmacy reimbursement was a small 2026 healthcare expense and not one of the larger quarter-end corrections, which is true but unhelpful for exact recall.","tags":["pharmacy-reimbursement","coarse-summary","manager"],"depth":4,"timestamp":"2026-01-01"}],"relevant_ids":["e1","e5"],"notes":"Recall@5 breaker with multiple rounded and range-based monetary descriptions plus neighboring amounts that can crowd out the exact cents."}
{"id":"r5-breaker-003","query":"What was the exact time the witness called the dispatcher?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Dispatch hearing transcript, admitted carrier exhibit: the witness said she called 'a little after 8:10,' the dispatcher thought it was 'closer to 8:15 than 8:10,' and a bystander anchored it to 'the second siren, just before the 8:14 hallway movement.' The phone carrier log, however, recorded the outgoing call at exactly 2026-08-02 20:13:11. The transcript makes clear that the witnesses were all reconstructing from surrounding events rather than from the record itself, and several later summaries rounded the time to 8:13 or 8:15 depending on what they were trying to compare it to.","tags":["dispatcher-call","exact-time","phone-log"],"depth":1,"timestamp":"2026-08-02T20:13:11"},{"id":"e2","text":"Witness recollection: 'It was a little after eight-ten, before the second hallway sweep, not yet as late as quarter after.' True in a broad sense, but still imprecise.","tags":["dispatcher-call","approximate-time","witness"],"depth":2},{"id":"e3","text":"Desk note from dispatch: the incoming witness contact was written as 20:13 because the sheet did not preserve seconds.","tags":["dispatcher-call","rounded-time","desk-note"],"depth":2,"timestamp":"2026-08-02T20:13:00"},{"id":"e4","text":"Camera event near the lobby: motion was logged at 20:14:02, which one lawyer tried to use as a proxy for the call time even though it was clearly a neighboring event.","tags":["dispatcher-call","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-08-02T20:14:02"},{"id":"e5","text":"Radio traffic note: the dispatcher repeated the witness report at 20:13:48, causing some staff to remember the relay time rather than the originating call time.","tags":["dispatcher-call","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-08-02T20:13:48"},{"id":"e6","text":"Cross-exam language: counsel cycled through 'after 8:10,' 'during the 8:13 minute,' and 'before 8:15' to show that the witness memory was not fabricated, only imprecise.","tags":["dispatcher-call","coarse-time","dialogue"],"depth":3},{"id":"e7","text":"Carrier billing printout: the official record preserves seconds and places the call before the relay, before the camera motion, and before the rounded dispatch note, despite all of those records sounding almost identical in oral summary.","tags":["dispatcher-call","authoritative-context","carrier"],"depth":3},{"id":"e8","text":"Incident report summary: the relevant communication happened on the evening of August 2, 2026, during the second response wave, which is accurate but much too coarse for the precise query.","tags":["dispatcher-call","coarse-summary","report"],"depth":4,"timestamp":"2026-08-02T00:00:00"}],"relevant_ids":["e1"],"notes":"Recall@5 breaker with a dense cloud of neighboring timestamps and rounded times around one authoritative call log."}
{"id":"r5-breaker-004","query":"What was the exact measured separation between the marker buoy and the survey launch?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Marine inquiry exhibit, instrument section: multiple witnesses described the survey launch as 'about twelve thousand feet off the buoy,' one harbor note called it 'roughly 11.8k,' and the press statement rounded it to twelve thousand for readability, but the actual instrument capture used by the inquiry recorded the separation as exactly 11,842 feet at the operative timestamp. The inquiry packet explicitly warns that witness estimates, range descriptions, and public summaries all drift toward cleaner numbers than the logged instrument value.","tags":["survey-launch-separation","exact-distance","instrument"],"depth":1},{"id":"e2","text":"Press note: the launch was described as approximately twelve thousand feet from the marker buoy, which is true enough for public orientation but not for exact recall.","tags":["survey-launch-separation","approximate-distance","press"],"depth":2},{"id":"e3","text":"Harbor operations notebook: the separation stayed in the high-eleven-thousand-foot range during the maneuver window and did not exceed the outer caution band.","tags":["survey-launch-separation","range-description","ops"],"depth":2},{"id":"e4","text":"Navigator testimony: 'I remember it as maybe eleven-eight-something, certainly not a full twelve-two,' which is closer than the rounded press note but still not the exact instrument reading.","tags":["survey-launch-separation","approximate-distance","testimony"],"depth":2},{"id":"e5","text":"Correction memo: the exact measured separation was 11,842 feet, not the rounded twelve-thousand-foot figure repeated in the outreach copy and hearing shorthand.","tags":["survey-launch-separation","exact-distance","correction"],"depth":3},{"id":"e6","text":"Inquiry policy note: navigational distances in outward-facing summaries were rounded to the nearest thousand unless the exact number was legally significant, which is why multiple witness statements can all sound reasonable while still being insufficiently precise.","tags":["survey-launch-separation","policy-context"],"depth":3},{"id":"e7","text":"Secondary chart margin note: one reviewer copied the value as 11,840 feet when drafting a visual aid, adding yet another near-exact but still wrong number to the case file.","tags":["survey-launch-separation","near-exact","chart-note"],"depth":3},{"id":"e8","text":"Admiralty summary: the launch-buoy spacing issue arose in 2026 during the late-summer harbor review, which is true background but useless for exact numeric recall.","tags":["survey-launch-separation","coarse-summary","admiralty"],"depth":4,"timestamp":"2026-01-01"}],"relevant_ids":["e1","e5"],"notes":"Recall@5 breaker with multiple rough, near-exact, and policy-rounded versions of the same measured distance."}
{"id":"weather-precision-001","query":"What was the exact average temperature recorded at the Central Park weather station on 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Climate litigation briefing, exhibit 14, stitched together from press notes, station summaries, and the original observational table: advocates kept describing 2023-07-14 as one of those mid-July days when Central Park was 'basically in the low eighties all day,' while one slide rounded the daily average to about 82 degrees and another folded the date into a broader narrative about how July 2023 was running several degrees above older baselines. A witness from the campaign team later mixed the Central Park figure with the airport figure and said the city 'averaged around eighty-two or eighty-three depending on which station you use.' The underlying station worksheet is more precise: for the Central Park weather station specifically, the exact daily average recorded on 2023-07-14 was 81.7 degrees Fahrenheit. That exact figure sits alongside a daily maximum, a daily minimum, anomaly commentary, and comparisons to the week before and after, which is why later summaries keep drifting toward rougher language like low eighties, around eighty-two, and very warm by historical standards. All of those statements are defensible in context, but only 81.7 degrees Fahrenheit is the exact Central Park daily average for that date.","tags":["central-park","daily-average-temperature","exact-reading","climate"],"depth":1,"timestamp":"2023-07-14"},{"id":"e2","text":"Campaign talking-points memo: the Central Park day was summarized as 'about 82F on average,' mainly to keep the climate narrative readable for interviews, and the memo repeatedly compared it to July normals, summer heat waves, and the broader warming trend rather than dwelling on the decimal-level reading.","tags":["central-park","daily-average-temperature","rounded-summary"],"depth":2},{"id":"e3","text":"Airport operations note: LaGuardia on the same date was also discussed in the hearing packet, and one airport-focused chart put that station in the low eighty-twos as well, which later caused several witnesses to slide between the park station and the airport station without noticing the station boundary.","tags":["laguardia","daily-average-temperature","neighbor-station"],"depth":2},{"id":"e4","text":"Cross-examination excerpt: one witness insisted the day was 'roughly eighty-two in the city, maybe a touch higher at the airport,' another said 'the park was the cooler of the two,' and a third referred only to the week as a whole because they remembered the climate narrative more clearly than the single-day number.","tags":["central-park","daily-average-temperature","testimony"],"depth":3},{"id":"e5","text":"Station methodology note: the Central Park table preserved the exact average at 81.7F, while outreach copy often rounded to the nearest whole degree or nearest broad category like low eighties, especially when discussing anomalies, decadal warming, and station-to-station differences in the same paragraph.","tags":["central-park","daily-average-temperature","methodology"],"depth":3},{"id":"e6","text":"Climate trend appendix: July 2023 as a whole was described as unusually warm, one of the hotter Julys in the modern station record, and part of a broader upward pattern in urban summer night temperatures, all of which is true but none of which answers the exact single-day Central Park value.","tags":["central-park","climate-trend","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Weather precision case with exact Central Park reading buried under rounded citywide and station-confused climate narrative."}
{"id":"weather-precision-002","query":"What was the exact average temperature recorded at the LaGuardia Airport weather station on 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Airport climate exhibit, prepared for the same hearing but sourced from the aviation weather archive instead of the park station: several speakers flattened the point by saying LaGuardia was 'a bit warmer than the park' or 'around the low eighties, maybe eighty-two and change,' while one reporter's notebook simply called the day 'roughly eighty-three by the airport sensors.' Because the hearing kept toggling between a climate-change narrative and local station comparisons, the exact airport number was easy to blur with Central Park's nearby figure. The tabulated record, however, is explicit: the exact daily average recorded at the LaGuardia Airport weather station on 2023-07-14 was 82.6 degrees Fahrenheit. That number coexists in the packet with references to the day being hotter than some nearby inland normals, cooler than a few runway-adjacent peaks, and representative of a broader summer warming signal. Those coarser descriptions are not false, but only 82.6 degrees Fahrenheit is the exact LaGuardia average for that date.","tags":["laguardia","daily-average-temperature","exact-reading","climate"],"depth":1,"timestamp":"2023-07-14"},{"id":"e2","text":"Press summary: the airport station was described as 'roughly eighty-three' because the comms team wanted a cleaner figure and cared more about the contrast with older baseline summers than the decimal-level station reading.","tags":["laguardia","daily-average-temperature","rounded-summary"],"depth":2},{"id":"e3","text":"Central Park comparison memo: the park was listed a little lower than the airport on the same date, and at least two later witnesses visibly borrowed the park number when asked about LaGuardia because the values were close enough to feel interchangeable in memory.","tags":["central-park","daily-average-temperature","neighbor-station"],"depth":2},{"id":"e4","text":"Testimony extract: one meteorology witness said 'the airport was in the eighty-two to eighty-three zone that day, definitely warmer than the park but not wildly so,' which is directionally right but still not precise enough for the exact station reading.","tags":["laguardia","daily-average-temperature","testimony"],"depth":3},{"id":"e5","text":"Aviation recordkeeping note: the LaGuardia figure remained 82.6F in the tabulated archive, even though other materials rounded it to whole numbers or discussed it only as part of a broader climate anomaly argument.","tags":["laguardia","daily-average-temperature","methodology"],"depth":3},{"id":"e6","text":"Climate context appendix: the date was used as one illustration of warming in transport-adjacent urban stations, but the appendix also discussed seasonal means, monthly means, and station pair comparisons, creating many true but less precise statements around the exact daily average.","tags":["laguardia","climate-trend","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Weather precision case with exact LaGuardia reading surrounded by rounded airport shorthand and Central Park confusion."}
{"id":"weather-precision-003","query":"What was the exact average temperature recorded at the Central Park weather station on 2022-01-29?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Winter climate testimony bundle: because the January 2022 cold snap was later used in arguments about climate variability, several witnesses described 2022-01-29 in broad, emotionally memorable language rather than numerical precision. One aide called it 'a mid-twenties day in the park,' a second witness said the average was 'around twenty-five or twenty-six,' and one attorney referred to it only as a cold late-January datapoint that opponents of climate-change messaging liked to cite as if one cold day disproved long-term warming. Buried underneath all of that rhetoric, the actual Central Park station sheet records the exact daily average on 2022-01-29 as 25.4 degrees Fahrenheit. The hearing packet then muddies the issue further by comparing that single-day exact value to monthly means, seasonal normals, and airport values from the same storm period, so the rougher claims remain defensible in ordinary speech. But if the question asks for the exact average at the Central Park station on that date, the only correct answer is 25.4 degrees Fahrenheit.","tags":["central-park","daily-average-temperature","exact-reading","climate"],"depth":1,"timestamp":"2022-01-29"},{"id":"e2","text":"Argument memo from opposing counsel: the park station was summarized as 'about twenty-five' because the memo cared more about the rhetorical force of a cold day than the decimal-level observation in the underlying sheet.","tags":["central-park","daily-average-temperature","rounded-summary"],"depth":2},{"id":"e3","text":"Station comparison note: LaGuardia on the same date was referenced repeatedly in the same record, and at least one summary blurred the airport and park figures by saying the city was generally in the mid-twenties that day.","tags":["laguardia","daily-average-temperature","neighbor-station"],"depth":2},{"id":"e4","text":"Climate scientist testimony: one cold day in late January 2022 was explained as weather rather than climate, and the scientist kept pivoting to long-term trends, monthly anomalies, and warming baselines instead of repeating the exact daily park average.","tags":["central-park","climate-trend","testimony"],"depth":3},{"id":"e5","text":"Methodology note: exact station tables preserved decimal averages like 25.4F, while most hearing summaries rounded winter readings to whole numbers because they were comparing themes rather than conducting a station audit.","tags":["central-park","daily-average-temperature","methodology"],"depth":3},{"id":"e6","text":"Seasonal appendix: January 2022, the broader winter period, and long-run cold-weather variability were all discussed in the same section, making the exact date-level Central Park value easy to lose among many true but coarser facts.","tags":["central-park","climate-trend","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Weather precision case with exact winter reading hidden inside climate variability arguments and same-date station comparisons."}
{"id":"weather-precision-004","query":"What was the exact average temperature recorded at the LaGuardia Airport weather station on 2022-01-29?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Aviation-climate exhibit from the winter storm hearing: LaGuardia's 2022-01-29 reading was repeatedly paraphrased as 'mid-twenties by the airport,' 'roughly twenty-six,' and once as 'slightly warmer than the park but still plainly a cold snap day.' Because the argument kept bouncing between airport operations, public messaging, and the broader climate debate, those rough formulations took over the conversation. The archived station table is more exact: the LaGuardia Airport weather station recorded a daily average of 26.1 degrees Fahrenheit on 2022-01-29. That figure was then wrapped in comparisons to Central Park, seasonal normals, and storm-response operations, all of which are factually connected but less precise than the actual station average.","tags":["laguardia","daily-average-temperature","exact-reading","climate"],"depth":1,"timestamp":"2022-01-29"},{"id":"e2","text":"Airport ops summary: the day was described as a mid-twenties cold event and was discussed mainly in relation to runway treatment, delay planning, and winter anomaly messaging rather than decimal-level climatology.","tags":["laguardia","daily-average-temperature","rounded-summary"],"depth":2},{"id":"e3","text":"Central Park comparison excerpt: one witness confused the airport and park values and referred to both stations collectively as 'basically twenty-five-ish that day,' which is broadly true but station-imprecise.","tags":["central-park","daily-average-temperature","neighbor-station"],"depth":2},{"id":"e4","text":"Meteorology testimony: another witness said the airport station was a little warmer than the park but still in the same cold-band narrative being used by both sides of the climate argument.","tags":["laguardia","daily-average-temperature","testimony"],"depth":3},{"id":"e5","text":"Aviation archive note: the exact LaGuardia daily average remained 26.1F in the official table even though several hearing summaries rounded it to twenty-six or mid-twenties for readability.","tags":["laguardia","daily-average-temperature","methodology"],"depth":3},{"id":"e6","text":"Climate appendix: the hearing materials blended daily station values, monthly means, season-wide comparisons, and airport-vs-park narratives, which makes the exact date-level airport value easy to bury under accurate but coarser framing.","tags":["laguardia","climate-trend","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Weather precision case with exact airport winter reading surrounded by same-date citywide and station-comparison noise."}
{"id":"debate-precision-001","query":"What was the exact date of the last confirmed death of the passenger pigeon Martha?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Transcript from the historical extinction symposium, afternoon plenary, heavily annotated by the archivist because the exchange kept circling around broad collapse curves instead of the exact terminal fact. Dr. Havel opened by arguing that the public's fixation on a single date is itself misleading: the species, he said, was ecologically doomed well before the twentieth century, and one could tell the story through wave after wave of contraction, from billions in the early nineteenth century to fragmented flocks after market hunting, telegraph-assisted hunting, and habitat conversion changed the economics of killing them. He cited state-level harvest records, fragmentary flock counts, and newspaper descriptions of skies 'dimmed for hours' and contrasted them with lonely turn-of-the-century notices about remnant birds. Dr. Quispe answered that the terminal date still matters because legal, cultural, and museum narratives require a final confirmed point; she then spent several minutes correcting Havel's tendency to talk in decades, pointing out that the public often confuses the disappearance of wild flocks, the last confirmed wild shooting, and the death of the last captive individual. Havel conceded, without explicitly agreeing in so many words, that those are separate historical markers and that many popular retellings mash them together. The transcript then became thick with competing but individually true statements: that the species was functionally finished before 1900, that the final captive era belongs to the early twentieth century, that the Cincinnati Zoo records became the controlling archival source, that September 1914 is the right month, and that some lectures round the event down to 'the beginning of the First World War period' because audiences remember world history more readily than zoo ledgers. The moderator finally forced both panelists to answer a clean archival question rather than a narrative one. Quispe cited the zoo ledger and said the last confirmed death of Martha, the last confirmed passenger pigeon, occurred on 1914-09-01. Havel, still preferring structural explanations to punctiliar dates, responded by talking about the decades-long decline, the 1890s collapse, and the way extinction is usually a process before it is an event, but he did not dispute the ledger date when pressed. The hearing packet later bundled this exchange with sidebars about market prices for pigeon meat, regional forest fragmentation, and the distinction between symbolic extinction dates and ecologically meaningful collapse points, all of which are true enough to swamp retrieval. But if the query asks for the exact date of Martha's last confirmed death, the only fully precise answer preserved in the record is 1914-09-01.","tags":["passenger-pigeon","martha","exact-date","debate"],"depth":1,"timestamp":"1914-09-01"},{"id":"e2","text":"Counter-narrative note by Dr. Havel: passenger pigeons were effectively gone as a meaningful species phenomenon well before the final zoo death, and a fixation on the ledger date obscures the way extinction unfolded as a long, industrially accelerated contraction. He cited rough abundance estimates in the billions, discussed the 1870s and 1880s hunting boom, mentioned remnant flock stories from the 1890s, and repeatedly described the species as 'historically finished before 1914 in every sense that mattered to ecology,' while still never quite denying that the captive terminal date has archival significance. This note is packed with true timeline statements but keeps sliding away from the exact date in favor of process language.","tags":["passenger-pigeon","decline-narrative","coarse-time"],"depth":2},{"id":"e3","text":"Response memo by Dr. Quispe: collapse is not the same thing as the last confirmed death, and historians should stop pretending that decade-level or even year-level framing is enough when the archival sources preserve a concrete terminal record. She agreed with Havel's statistics on population decline, market hunting, and habitat fragmentation, although she did so indirectly by building on them rather than saying 'I agree.' She then listed several dates that audiences routinely confuse: late nineteenth-century mass declines, early-twentieth-century captive survival, and the September 1914 zoo death. The memo is full of agreement buried inside disagreement and therefore easy to misread as pure opposition.","tags":["passenger-pigeon","archival-narrative","coarse-time"],"depth":2},{"id":"e4","text":"Moderator's summary sheet: the panelists converged more than their tone suggested. Both accepted that the decline was long, that wild disappearance predated the final captive death, and that popular memory compresses distinct milestones into a single fuzzy story about 'around 1914' or 'the early 1900s.' The summary intentionally preserved those broad formulations because the session was about public misunderstanding, not just date retrieval.","tags":["passenger-pigeon","agreement-summary","rounded-time"],"depth":3},{"id":"e5","text":"Archive appendix: museum and zoo materials note that Martha died in September 1914, while some educational summaries stop there and omit the day. Those summaries are not wrong, but they are less precise than the ledger-backed date used by the symposium when forced into exactness.","tags":["passenger-pigeon","month-summary","archive"],"depth":3,"timestamp":"1914-09-01"},{"id":"e6","text":"Context packet: one sidebar compared Martha's death to broader extinction narratives, citing species lost in the nineteenth century, species declared extinct much later than their final sighting, and the common habit of collapsing exact archival events into seasonal or year-level shorthand. The sidebar is true and relevant context, but it is also retrieval mud because it invites the model to settle for '1914' or 'September 1914' rather than the exact date.","tags":["passenger-pigeon","context","year-month"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form debate case with heavy historical data, implicit agreement, and one exact archival date embedded under decades of true-but-coarser extinction framing."}
{"id":"debate-precision-002","query":"What was the exact date of the last captive thylacine's death at the Hobart Zoo?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Transcript from the museum-hosted debate on extinction chronology and public memory, evening session, pages 41 through 53 of the official record. Dr. Renwick argued that the ordinary public account of the thylacine is too tidy: people say 'extinct in the 1930s' as if the matter can be compressed into one clean line, while the real evidentiary story is layered with bounty regimes, collapsing mainland populations, Tasmanian remnant sightings, dubious post-1936 claims, and the persistent confusion between the final known captive death and the later bureaucratic declaration of extinction. He filled the room with statistics on bounty payments, sheep-loss politics, habitat contraction, and the decline in verified captures, insisting that the ecological disappearance narrative is more important than the zoo date people like to memorialize. Dr. Sato countered that the date still matters because the public and the archive repeatedly confuse at least four different things: the species' longer decline, the last known captive death, the last alleged wild sighting, and the later formal declaration. She then read from the zoo records and explained that even within museum circles the event gets rounded to '1936,' 'early September 1936,' or 'the last winter season before the declaration,' all of which are broadly true but still coarser than the archival entry. Renwick answered with a long discussion of functional extinction, ecological irreversibility, and the way single dates distort long processes, yet when the moderator forced him to confront the archival record he did not dispute the primary document. The controlling record, as entered into the debate file, is the Hobart Zoo note stating that the last captive thylacine died on 1936-09-07. The remainder of the transcript is full of statistics about bounties, sheep-kill claims, collection records, and the suspicious elasticity of later sightings. Both scientists implicitly agree on many underlying facts: that the species was in severe decline well before 1936, that later sighting claims are a different category from captive archival certainty, and that people often remember decade and month while forgetting day-level precision. Still, because the debate spends so much time on process, policy, and myth, the exact date is easy to lose unless one follows the archival line all the way through.","tags":["thylacine","exact-date","debate","archive"],"depth":1,"timestamp":"1936-09-07"},{"id":"e2","text":"Renwick's position paper: extinction should not be narrated as if it began and ended with a zoo death, because by the time the final captive animal died the ecological collapse was already a fait accompli. He cited bounty totals, regional disappearance patterns, and the fog of later sightings, repeatedly using coarse formulations like 'the 1930s endgame' and 'the final prewar years.' The memo contains many true historical anchors while still resisting the exact archival date as the primary narrative object.","tags":["thylacine","decline-narrative","coarse-time"],"depth":2},{"id":"e3","text":"Sato's archival rejoinder: process and date are both necessary, and the confusion in public memory comes partly from collapsing the exact zoo death, the decade of terminal decline, and the much later legal declaration into one fuzzy extinction slogan. She agreed with Renwick's broader decline statistics by restating them, but she redirected their significance toward the need for precise archival markers when the question is exact chronology rather than ecological interpretation.","tags":["thylacine","archival-narrative","coarse-time"],"depth":2},{"id":"e4","text":"Moderator note: both panelists accepted, without explicitly announcing agreement, that the 1936 zoo death is not the same thing as the last alleged sighting or the later formal declaration. The moderator's notes preserve broad phrases like 'September 1936' and 'the last captive era' because the session was framed for a mixed public audience.","tags":["thylacine","agreement-summary","rounded-time"],"depth":3},{"id":"e5","text":"Museum wall-text draft: the thylacine's last captive death is summarized as occurring in September 1936, which is accurate enough for interpretive signage but less precise than the actual archival date read aloud in the debate.","tags":["thylacine","month-summary","museum"],"depth":3,"timestamp":"1936-09-01"},{"id":"e6","text":"Policy appendix: the hearing packet cross-referenced other species where final archival dates, formal declarations, and popular memory diverge badly, reminding readers that exactness is usually the first thing lost when complex extinction stories are compressed for public retelling.","tags":["thylacine","context","exactness"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form extinction-debate case with conflicting narratives, shared underlying facts, and one exact archival date buried beneath decades of data and rhetoric."}
{"id":"debate-precision-003","query":"What was the exact annual mean CO2 concentration recorded at Mauna Loa for 1990?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Transcript from a public climate-methods debate between two atmospheric scientists who broadly agree that long-run greenhouse concentrations have risen but disagree sharply about how public-facing narratives should summarize the record. Dr. Alves argued that advocates often overdramatize year-level points by cherry-picking dramatic framing like 'nearly 355 ppm by the early 1990s,' while Dr. Mercer countered that the public is already confused by excessive rounding, because people hear phrases like 'mid-350s,' 'roughly 354,' or 'around the 1990 turning point' and then later mistake those rhetorical summaries for exact observed values. Alves spent pages walking through baselines, decade-over-decade trends, interannual variability, and the danger of pretending that one annual average is the whole climate story; Mercer replied with equal detail about why exact annual means still matter in testimony, especially when lawyers and policymakers start quoting rounded values back as if they were precise observations. Both scientists cited monthly values, growth rates, trend lines, and comparisons to later years above 360 ppm, and both made points that, while rhetorically opposed, depend on a large shared factual foundation: Mauna Loa is authoritative for this debate, annual means are not the same thing as isolated monthly highs, and rounded public summaries can be directionally right without being exact. Buried inside the transcript, when the moderator finally demanded an exact answer for the annual mean CO2 concentration recorded at Mauna Loa in 1990, Mercer stated the value plainly from the table: 354.39 ppm. Alves did not dispute the number; instead he argued that debates over climate communication keep dragging exact figures into broader fights about trend interpretation. The transcript therefore contains a tangle of true but coarser formulations such as 'mid-350s,' 'roughly 354,' 'early-1990s rise,' and 'well below later 2000s levels,' yet only 354.39 ppm is the exact annual mean asked for.","tags":["mauna-loa","co2","exact-annual-mean","debate"],"depth":1,"timestamp":"1990-01-01"},{"id":"e2","text":"Alves supplemental chart note: he repeatedly described 1990 as sitting in the mid-350 ppm band and contrasted it with later annual means, arguing that the larger trend matters more than fetishizing the second decimal place. The note is factually compatible with the exact record while still being less precise than the annual mean table.","tags":["mauna-loa","co2","rounded-summary"],"depth":2},{"id":"e3","text":"Mercer rebuttal memo: annual means are often rounded in public summaries, but when questioned under oath or in formal reporting one should not substitute 'about 354' or 'mid-350s' for the actual table value, particularly when the data source is already authoritative and standardized.","tags":["mauna-loa","co2","methodology"],"depth":2},{"id":"e4","text":"Cross-examination note: both scientists spent several pages discussing monthly peaks, decade-level rates, and later concentrations above 360 ppm, which produced a dense cloud of nearby numbers that are all true in context yet are not the exact 1990 annual mean.","tags":["mauna-loa","co2","nearby-values"],"depth":3},{"id":"e5","text":"Public-facing summary: some outreach copy rounded the 1990 mean to 354.4 ppm and other materials collapsed it into a phrase like 'roughly 354,' both of which are close enough to sound convincing when compared against longer trend narratives.","tags":["mauna-loa","co2","rounded-summary"],"depth":3},{"id":"e6","text":"Moderator's synthesis: despite their disagreement over rhetoric, both scientists plainly relied on the same Mauna Loa record, the same distinction between annual means and monthly highs, and the same basic upward trend; their conflict was over framing, not over the existence of a precise annual table value.","tags":["mauna-loa","co2","agreement-summary"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form scientist debate case with climate trend rhetoric, rounded but true summaries, and one exact annual mean buried beneath mutually acknowledged context."}
{"id":"debate-precision-004","query":"What was the exact date of the final confirmed great auk killing on Eldey?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Transcript from the historical ecology colloquium, evening debate on extinction memory and narrative compression. Dr. Eames argued for nearly twenty minutes that public memory of the great auk is a museum myth stitched together from one melodramatic image: a doomed pair on a lonely Icelandic rock. He insisted that this image, while not wholly false, hides the wider commercial chronology of slaughter, egg collection, specimen demand, and the earlier pan-North Atlantic contraction that made a single late event feel like an ending rather than one act in a longer collapse. He cited shipping records, specimen market values, colonial collecting patterns, and references to birds vanishing from earlier islands long before the event on Eldey that usually dominates public retelling. Dr. Rafiq agreed with almost all of that decline narrative but said the problem is not the existence of a final date; the problem is that popular retellings then downgrade the exact date into a slogan like 'mid-1840s,' 'the summer of 1844,' or even 'around the time of the Icelandic collecting rush.' She walked through the hierarchy of increasingly precise statements: that the species was functionally broken earlier, that the final confirmed killing on Eldey happened in 1844, that it happened in June, and that the archive finally narrows it all the way to 1844-06-03. Eames continued trying to drag the conversation outward toward market incentives, comparative island extirpations, and the folly of fetishizing terminal dates, yet under cross-questioning he conceded the archive is clear on the final confirmed killing date even if it is not the whole story. The session then dissolved into a flood of true but distracting facts: one curator discussed the final egg sale; one historian argued over whether the symbolic death of a breeding pair matters more than the species-wide collapse; another speaker compared the great auk narrative to later cases where formal declaration long lagged actual disappearance. Both debaters repeatedly relied on the same underlying facts while sounding adversarial: that collapse was prolonged, that 1844 is not the whole history, and that the exact final confirmed date still exists in the documentary record. When the moderator forced the issue into a single archival answer, Rafiq stated the exact final confirmed great auk killing on Eldey as 1844-06-03. That date survived the debate, even though almost everything around it tried to dissolve it into broader historical rhetoric.","tags":["great-auk","exact-date","debate","history"],"depth":1,"timestamp":"1844-06-03"},{"id":"e2","text":"Eames position paper: the great auk was effectively doomed long before 1844, and he preferred to speak in terms of final decades, collecting pressure, and the commercial logic that made a last event almost inevitable. The paper repeatedly uses broad phrases like 'mid-1840s' and 'the final Icelandic stage' that are historically grounded yet too coarse for an exact-date query.","tags":["great-auk","coarse-time","decline-narrative"],"depth":2},{"id":"e3","text":"Rafiq archival memo: collapse history and exact chronology are not competing ideas. She listed the common degradations of precision in public memory, from 1844 to June 1844 to vague phrases like 'summer 1844,' before restating that the final confirmed Eldey killing is documented to the day. She never framed this as agreeing with Eames, but she built directly on his longer decline statistics throughout the memo.","tags":["great-auk","month-year","archival-narrative"],"depth":2},{"id":"e4","text":"Moderator's synthesis sheet: both panelists accepted that the exact terminal event is a real historical datum while also insisting it should not erase earlier collapse dynamics. The summary intentionally preserved coarser phrases like 'the last phase of 1844' because the audience discussion kept orbiting around symbolism rather than date retrieval.","tags":["great-auk","agreement-summary","coarse-time"],"depth":3},{"id":"e5","text":"Museum label draft: many exhibits stop at 'June 1844' because wall text is short and day-level precision looks pedantic to general audiences, even though the supporting archive underneath the label is more exact.","tags":["great-auk","month-summary","museum"],"depth":3,"timestamp":"1844-06-01"},{"id":"e6","text":"Historical context appendix: the packet also discussed specimen trafficking, extinction symbolism, and later conservation storytelling, all of which are factually relevant but semantically dense enough to drown out the exact date unless the retriever is unusually precise.","tags":["great-auk","context","history"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form extinction debate with broad decline rhetoric, partial hidden agreement, and one exact archival date buried under many true coarser formulations."}
{"id":"debate-precision-005","query":"What was the exact annual mean global sea level anomaly cited from the report?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Transcript from a combative panel on sea level communication, where two oceanographers spent nearly an hour arguing over whether public climate testimony should foreground smooth long-run trend lines or preserve the awkward decimal exactness of individual annual report values. Dr. Nouri argued that advocates cherry-pick emotionally resonant figures like 'record highs' and then flatten the annual tables into phrases such as 'around ninety millimeters above the baseline' or 'roughly nine centimeters up' because rounded language travels better in hearings and media. Dr. Keating countered that the rounding is itself part of the problem: once an annual report value is collapsed into broad phrases, the exact anomaly often disappears and the public cannot tell which number came from the table and which came from rhetorical compression. Nouri responded with a barrage of contextual facts that were all true but strategically widening: the anomaly sat in a broader acceleration pattern, the decade mean mattered more than one year, several neighboring years were also historically elevated, and basin-level variability complicates single-number storytelling. Keating, without openly saying she agreed, repeatedly accepted those broader points while insisting that the annual report still contains a single exact anomaly value and that formal testimony should preserve it when the question is explicit. The moderator then forced both scientists to answer from the same report page. After pages of argument about baseline windows, satellite eras, and trend framing, the exact annual mean global sea level anomaly cited in the report was stated as 91.3 mm. Around that precise value the debate accumulated a fog of nearby but not identical formulations: about ninety millimeters, roughly nine centimeters, low nineties in millimeters, part of the highest cluster on record, and above the previous decade mean. All of those statements can be true or defensible in context, but only 91.3 mm answers the exact-value question.","tags":["sea-level-anomaly","exact-value","debate","climate"],"depth":1},{"id":"e2","text":"Nouri's narrative memo: the anomaly belonged to a broader high-water period and should be understood as part of a multi-year signal rather than fetishized as a stand-alone decimal. He repeatedly used phrases like 'around ninety millimeters' and 'about nine centimeters' because his rhetorical goal was trend salience, not table fidelity.","tags":["sea-level-anomaly","rounded-summary","memo"],"depth":2},{"id":"e3","text":"Keating's rebuttal: decimal exactness matters when lawyers and policymakers later quote the record. She agreed with the larger trend narrative but argued that the exact report value should not be overwritten by rough phrases like 'low nineties' or 'roughly ninety.'","tags":["sea-level-anomaly","methodology","memo"],"depth":2},{"id":"e4","text":"Cross-examination note: the panel spent time comparing neighboring annual anomalies, decade averages, and baseline-normalized interpretations, creating a dense neighborhood of true but non-identical values around the exact annual figure.","tags":["sea-level-anomaly","nearby-values","dialogue"],"depth":3},{"id":"e5","text":"Outreach summary: one summary rounded the report value to 91 mm, another to about 90 mm, and a slide deck translated it into roughly 9.1 cm above the reference baseline. Those values are defensible summaries but still not the exact annual report figure.","tags":["sea-level-anomaly","rounded-summary","outreach"],"depth":3},{"id":"e6","text":"Moderator note: both scientists accepted the same underlying report table, the same baseline, and the same upward trajectory, even though their rhetorical strategies made them sound more opposed than they really were.","tags":["sea-level-anomaly","agreement-summary","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form climate debate with cherry-picked rhetoric, hidden agreement, and one exact report value surrounded by many rounded true summaries."}
{"id":"court-precision-005","query":"What was the exact timestamp on the dock camera when the crate crossed the yellow line?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Dock safety hearing transcript, entered as Volume III because the lawyers insisted on replaying the same three minutes of footage from five different angles. The foreman testified that the crate crossed the yellow line 'just after the half-hour mark,' the loader thought it happened 'closer to 6:31 than 6:30,' and the safety observer kept anchoring his memory to a forklift horn that he believed sounded before the crossing but could not place to the second. Counsel then spent pages walking through the ordinary confusion between the camera overlay, the rounded shift log, the forklift dispatch radio, and the handwritten incident sheet that one supervisor filled out after the fact. In the middle of that clutter, the camera export itself remained the cleanest piece of evidence: the crate crossed the yellow line at exactly 2026-06-18 18:30:47. Nearly every other document in the packet smudged the moment into a coarser form. The handwritten log called it 'about 6:31,' the radio relay went out at 18:31:09, the forklift horn was noted somewhere in the 18:30 minute, and one witness kept referring to it as 'that point after the second loading pause.' Even the safety memo later rounded the event to 'the 6:30 crossing' because the memo's purpose was procedural, not evidentiary. The record is therefore packed with true neighboring timestamps and plausible oral summaries, yet only the dock camera export preserves the exact timestamp 2026-06-18 18:30:47 for the yellow-line crossing.","tags":["crate-yellow-line-crossing","exact-time","camera-log","courtroom"],"depth":1,"timestamp":"2026-06-18T18:30:47"},{"id":"e2","text":"Shift log summary: the crate crossed the yellow line at about 6:31 p.m., written by a supervisor who rounded multiple events in the same page because the log was designed for sequence, not second-level precision.","tags":["crate-yellow-line-crossing","rounded-time","shift-log"],"depth":2,"timestamp":"2026-06-18T18:31:00"},{"id":"e3","text":"Forklift radio record: the warning call was transmitted at 18:31:09, which is close enough to contaminate memory but still later than the actual crossing on the camera export.","tags":["crate-yellow-line-crossing","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-06-18T18:31:09"},{"id":"e4","text":"Witness reconstruction: several people described the crossing as happening in the 6:30 minute after the second pause and before the horn relay, which is directionally correct but not precise enough for an exact timestamp query.","tags":["crate-yellow-line-crossing","coarse-time","witness"],"depth":3},{"id":"e5","text":"Evidence handling note: one printed still rounded the overlay to 18:30 because the exported frame header omitted seconds, which later created a false impression that the exact record lacked second-level precision.","tags":["crate-yellow-line-crossing","nearby-time","evidence"],"depth":3,"timestamp":"2026-06-18T18:30:00"},{"id":"e6","text":"Procedure memo: the hearing packet also discussed forklift spacing rules, shift fatigue, and line-of-sight problems, all of which are true context but semantically noisy if the actual question is only about the exact crossing timestamp.","tags":["crate-yellow-line-crossing","context","procedure"],"depth":4}],"relevant_ids":["e1"],"notes":"Long-form hearing case with many near-matching timestamps and one exact camera record."}
{"id":"mega-precision-001","query":"What was the exact timestamp on the courthouse vestibule camera when the blue evidence binder was first carried past the metal detector?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Superior Court evidentiary hearing, consolidated transcript volume VII, pages 114 through 167, with sidebars, clerk annotations, and exhibit references left intact because nobody could agree on whether the dispute was about chain of custody, witness memory, security procedure, or merely sloppy note-taking. The prosecutor opened by saying the blue evidence binder crossed the vestibule threshold 'around a quarter past nine,' which immediately prompted defense counsel to object that the state kept collapsing several different moments together: the time the binder was first seen on the exterior camera, the time it was carried past the metal detector, the time the deputy initialed the property sheet, and the time the hallway runner wrote '9:16-ish' in a notebook that had already rounded two earlier events. The lead deputy then testified that in his mind the binder came through 'closer to 9:16 than 9:15' and definitely before the call from courtroom three was relayed to security. Under cross-examination he admitted that this recollection rested on the rhythm of the morning rather than on any precise independent memory, because he remembered a coffee spill, a juror complaint, a magnetometer false alert, and the fact that the judge had not yet gone on the bench. The receptionist was even less exact; she said only that it was after the 9:14 elevator arrival she remembered because the elevator alarm always chimed twice, but before the 9:17 hallway traffic surge that happened when a different courtroom released witnesses. Defense counsel then produced the property intake sheet, which the clerk had written at 09:16, and spent several pages implying that the intake sheet and the camera moment were one event, while the prosecutor responded that the intake sheet time was a rounded administrative notation entered after the binder had already crossed the detector. By the time the court reached the camera evidence, the record had accumulated a small storm of mutually compatible but imprecise time claims: sometime in the 9:15 minute, around 9:16, before the courtroom-three call, after the 9:14 elevator, before the 9:17 hallway surge, and during the opening traffic of the morning session. The security integrator then testified that the vestibule camera, unlike the hallway summary printout, preserved second-level precision and had been synchronized at 07:03 that morning against the master clock. The exported frame log, which the court admitted over a chain-of-custody objection, recorded the first moment the blue evidence binder was carried past the metal detector at exactly 2026-04-09 09:15:42. That did not stop the argument. The defense insisted the frame was merely the first visible carrying moment, not necessarily the first secured passage, while the state replied that the question before the court was the exact time on the vestibule camera when the binder was first carried past the detector, not when paperwork caught up or when someone later summarized the event in a rounded note. Later in the same transcript, the court clerk acknowledged that she had initially drafted a chronology saying 'about 9:16 a.m.' because she copied the intake sheet, and only corrected it after matching the intake line against the video export and the dispatcher call record. The dispatcher record itself created more confusion by logging a related but later status confirmation at 09:16:11. A hallway still image, printed for the jury, showed the binder already beyond the threshold at 09:15:49, which is consistent but later than the first crossing frame. The summary in the final order complained that nearly every human witness had remembered the sequence accurately in shape but not in precision. The only exact answer preserved by the authoritative vestibule camera export is 2026-04-09 09:15:42.","tags":["blue-evidence-binder","exact-time","camera-log","hearing"],"depth":1,"timestamp":"2026-04-09T09:15:42"},{"id":"e2","text":"Property intake sheet and deputy initials: the binder was logged at 09:16, but the clerk later admitted that the intake sheet was an administrative capture written after the detector crossing and rounded to the minute because the form template did not preserve seconds. The sheet became persuasive to several witnesses because it was easier to read than the frame export, even though it was not the earliest exact event in the chain.","tags":["blue-evidence-binder","rounded-time","paperwork"],"depth":2,"timestamp":"2026-04-09T09:16:00"},{"id":"e3","text":"Dispatcher status line: security confirmed the binder had cleared the vestibule at 09:16:11 when courtroom three called to ask whether the evidence packet had arrived yet. During the hearing, several participants accidentally treated this as the crossing time because it was the first timestamp that appeared in the radio system, even though the camera log had already shown the binder beyond the detector before that call.","tags":["blue-evidence-binder","neighbor-timestamp","dispatcher"],"depth":2,"timestamp":"2026-04-09T09:16:11"},{"id":"e4","text":"Receptionist statement: she placed the binder movement after the 09:14 elevator arrival and before the 09:17 witness surge, saying she was 'comfortable with quarter after nine, maybe a touch after.' Her memory was built from surrounding activity, not from any instrumented record. The testimony is true enough in sequence but too blunt for the exact question.","tags":["blue-evidence-binder","coarse-time","witness"],"depth":2},{"id":"e5","text":"Hallway still-image packet: one printed frame used in argument was timestamped 09:15:49 and showed the binder already inside the secure side of the vestibule. Because it was easier to display to the room than the frame-by-frame export, several attorneys kept gravitating to this later but visually convenient timestamp.","tags":["blue-evidence-binder","neighbor-timestamp","still-image"],"depth":3,"timestamp":"2026-04-09T09:15:49"},{"id":"e6","text":"Chain-of-custody correction memo: the clerk revised an earlier chronology that said 'about 9:16 a.m.' after discovering that the intake sheet, dispatcher line, and camera export referred to adjacent but non-identical moments. The memo expressly preserves 09:15:42 as the first carrying-past-detector time on the vestibule camera.","tags":["blue-evidence-binder","correction","exact-time"],"depth":3,"timestamp":"2026-04-09T09:15:42"},{"id":"e7","text":"Master clock maintenance record: the security integrator synchronized the vestibule system at 07:03 and later testified that no offset alert appeared before the hearing began, undercutting the defense suggestion that the camera might have drifted several seconds away from the paper log and dispatch clocks.","tags":["blue-evidence-binder","context","clock-sync"],"depth":3,"timestamp":"2026-04-09T07:03:00"},{"id":"e8","text":"Narrative summary in the court's final order: human recollection preserved the shape of the morning, not the exact second. The order catalogued the rounded times, the neighboring events, the administrative delays, and the temptation to treat all timestamps as interchangeable, before concluding that only the vestibule export answered the exact question asked.","tags":["blue-evidence-binder","context","order-summary"],"depth":4}],"relevant_ids":["e1","e6"],"notes":"Massive courtroom timestamp case with multiple adjacent records, rounded paperwork, and one controlling camera timestamp buried in hearing clutter."}
{"id":"mega-precision-002","query":"What was the exact annual mean atmospheric methane concentration cited for 2004?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"International climate methods symposium, special adversarial panel on statistical narration and public misunderstanding, full afternoon transcript with audience questions retained because the interruptions themselves became part of the rhetorical fog. Dr. Elson began by objecting to what he called 'decimal theater,' arguing that climate communication often pretends to gain seriousness by reciting values to two decimal places even when the real public-facing question is trend direction, uncertainty structure, and relation to the decades before and after. He spent pages on baseline construction, atmospheric lifetime, observational harmonization, and the tendency of advocacy groups to cherry-pick a year like 2004 only when it helps a story about plateaus, pauses, or resumed acceleration. Dr. Banerjee answered that the abuse of exact values is not cured by replacing them with lazy verbal approximations. Once people start saying 'around seventeen-seventy-five ppb,' 'high seventeen-seventies,' or 'about 1.78 ppm in methane terms,' they blur together distinct annual means, monthly anomalies, and later reinterpretations. She then walked through the report table line by line, reminding the audience that 2004 sits inside a region of nearby values close enough to invite confusion: one slide rounded the annual mean to the nearest whole ppb, another turned it into an approximate ppm-equivalent phrase, and a third compared it only to the surrounding cluster of annual means without preserving the exact number at all. Elson pressed harder, citing decade-scale context, interannual wobble, measurement harmonization, and the way some policy briefs opportunistically compare 2004 to 1998 or 2010 without caring whether the exact annual mean survives the journey. Banerjee, without ever simply saying 'I agree,' accepted almost all of the contextual points and then insisted that the exact report value still matters when a question explicitly asks for it. The transcript became a thicket of true but competing formulations: the annual mean was in the high seventeen-seventies, roughly 1.78 ppm, within a narrow band relative to neighboring years, and part of a larger story about plateau narratives that were later challenged. At the moderator's insistence, Banerjee finally read the exact annual mean atmospheric methane concentration cited for 2004 from the table itself: 1774.62 ppb. Elson did not contest the figure; he returned instead to the claim that overemphasizing year-level decimals invites false certainty in broader public discourse. Audience questions then made things worse by dragging in nearby years, month-level spikes, and policy-brief roundings. The result is a transcript in which almost every surrounding statement is broadly true and semantically rich, but only 1774.62 ppb is the exact annual mean cited for 2004.","tags":["methane-2004","exact-annual-mean","climate-debate"],"depth":1,"timestamp":"2004-01-01"},{"id":"e2","text":"Elson's slide deck: the 2004 annual methane value was summarized as living in the high-1770s ppb band and discussed mainly as part of a cluster, not as an isolated decimal. He repeatedly used broad descriptors such as 'roughly 1775' and 'around 1.775 ppm equivalent' because his rhetorical target was trend interpretation rather than table fidelity.","tags":["methane-2004","rounded-summary","slides"],"depth":2},{"id":"e3","text":"Banerjee rebuttal handout: she agreed that broader context matters, but warned that once annual means are rounded into whole-ppb or approximate-ppm forms, later readers forget whether a number came from the authoritative table, a public summary, or a visual simplification. The handout contrasts exact annual means with the sloppy life of rounded derivatives in testimony and media.","tags":["methane-2004","methodology","handout"],"depth":2},{"id":"e4","text":"Audience exchange: one questioner quoted 1775 ppb, another said 'wasn't it about 1.78 ppm by then,' and a third brought up neighboring years whose means sat close enough to make 2004 easy to misremember. None of those interventions was random noise; all were based on true nearby interpretations that nevertheless fail to preserve the exact annual table value.","tags":["methane-2004","nearby-values","audience"],"depth":3},{"id":"e5","text":"Outreach appendix: policy summaries alternated between 1775 ppb, roughly 1.78 ppm, and high-1770s language depending on audience sophistication, all while the underlying report continued to preserve the exact value with two decimals.","tags":["methane-2004","rounded-summary","outreach"],"depth":3},{"id":"e6","text":"Moderator's closing summary: both scientists relied on the same report table, accepted the same directional trend, and argued mostly about narrative responsibility rather than about the existence of an exact annual mean. That hidden agreement is easy to miss because their debate style makes the shared factual substrate feel much weaker than it is.","tags":["methane-2004","agreement-summary","context"],"depth":4}],"relevant_ids":["e1"],"notes":"Massive scientific debate case with many rounded, converted, and cluster-based descriptions around one exact annual methane mean."}
{"id":"mega-precision-003","query":"What was the exact amount of the estate reimbursement check?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Probate court reimbursement hearing, transcript volume II, where the dispute somehow expanded from a single reimbursement check into a three-hour argument about informal bookkeeping culture, whether oral family summaries should count as evidence, and the degree to which rounded financial language becomes 'good enough' until someone is asked for the exact amount under oath. The executor testified that the estate reimbursement check was 'right around fifty thousand' and then, under pressure, amended that to 'a little under fifty' because he remembered discussing it in threshold language with the family accountant rather than reading directly from the canceled check. The claimant's lawyer then produced text messages in which siblings referred to the amount as 'basically fifty grand,' 'forty-nine and change,' and 'the larger of the two late-summer reimbursements,' none of which was invented but all of which blurred the exact figure. A niece testified that she remembered seeing a bank app line beginning with 49,9 and therefore assumed the check was something like 49,975, while the family bookkeeper recalled a printed ledger showing 49,947.80 but admitted she had been switching between a gross estate sheet and a reimbursement-specific sheet during the same week. The hearing then turned into a swamp of adjacent true numbers: the rounded threshold used in a family meeting, the two-check subtotal, the gross expense pool from which the reimbursement was drawn, and the amount after a later correction that bundled postage and filing fees. The court's patience visibly thinned until the probate registrar read from the deposited instrument and the corrected estate ledger. The exact amount of the estate reimbursement check, as preserved in the deposited instrument and matched by the corrected ledger, was $49,947.83. Every other number in the room had some relationship to the truth: it was a rounded version, a subtotal, a threshold summary, or a near-miss copied from a draft sheet. Even the judge remarked that the family had not really disagreed about scale so much as they had collectively allowed casual speech to erode precision. The transcript is therefore a deliberate minefield of plausible monetary statements, but only $49,947.83 answers the exact-amount question.","tags":["estate-reimbursement-check","exact-amount","probate-court"],"depth":1},{"id":"e2","text":"Executor testimony summary: the reimbursement was treated as 'around fifty thousand' in family meetings because everyone was reasoning about whether it cleared an informal threshold for extra review, not because anyone believed that rounded language would later substitute for the deposited instrument.","tags":["estate-reimbursement-check","rounded-amount","testimony"],"depth":2},{"id":"e3","text":"Sibling text packet: one sibling wrote 'forty-nine and change,' another wrote 'basically fifty grand,' and a third compared it to a separate smaller reimbursement from the same month. The packet is full of truthful shorthand that points toward the right magnitude while muddying the exact cents.","tags":["estate-reimbursement-check","rounded-dialogue","chat"],"depth":2},{"id":"e4","text":"Bookkeeper draft sheet: one printed version showed 49,947.80 because three cents had not yet been reconciled against the final deposit image. That number became a dangerously sticky near-miss in the hearing because it looked formal enough to outrank family chatter while still being wrong.","tags":["estate-reimbursement-check","near-exact","worksheet"],"depth":3},{"id":"e5","text":"Corrected estate ledger and deposited instrument: the exact reimbursement check amount was $49,947.83, with the final cents confirmed after matching the instrument image against the posted estate line and separating later postage adjustments from the reimbursement itself.","tags":["estate-reimbursement-check","exact-amount","ledger"],"depth":3},{"id":"e6","text":"Probate procedure note: family reimbursement amounts were often discussed in rounded thousands for emotional and practical reasons, especially during contentious meetings, which is how broad agreement on scale can coexist with total confusion about the exact number.","tags":["estate-reimbursement-check","policy-context","probate"],"depth":4}],"relevant_ids":["e1","e5"],"notes":"Massive probate-hearing amount case with rounded family recollections, a near-exact worksheet trap, and one exact deposited-instrument value."}
{"id":"quartet-r5-001","query":"What was the exact timestamp when the red ledger binder first left the witness prep room?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"Transcript of the courthouse internal affairs hearing, panel session two. Speaker A, the prep-room clerk, testified that the red ledger binder left 'a little after 10:08' and repeatedly tied her memory to the fact that she had just finished telling a witness to sit back down before the bailiff reopened the hallway. Speaker B, the runner, claimed much more confidently that it left at '10:11 on the nose' and built an elaborate story around hearing the printer restart, but under cross-examination admitted that he had reconstructed the sequence later from his own movement sheet and had not actually looked at the prep-room doorway when the binder left. Speaker C, the bailiff, never gave a precise time; instead he said the binder was already out before the 10:10 radio check and after the second witness swap, and he corrected counsel twice for trying to force his memory into the wrong minute. Speaker D, the hallway deputy, said she thought the binder moved during the 10:08 minute because she saw the clerk still holding the door before the radio call, though she refused to pretend she knew the exact second. The hearing then got uglier. Counsel read from a rounded movement log saying 10:09, from a hallway camera still at 10:08:41 showing the binder already beyond the doorway, and from the runner's own note at 10:11 that appears to record when he received the binder, not when it left the room. A phone call to chambers was logged at 10:10:02 and one lawyer tried to anchor the binder movement against that instead, even though the witnesses agreed only that the binder left before the call was completed. The decisive exhibit was the prep-room interior camera export, which the systems technician stated had been clock-synced at 07:12 that morning. That export recorded the first moment the red ledger binder fully cleared the prep-room threshold at exactly 2026-02-11 10:08:37. Speaker A's 'a little after 10:08' and Speaker D's 'during the 10:08 minute' are corroborative without being identical. Speaker B's cleaner 10:11 story is not corroborated by any other witness and instead lines up suspiciously with his later handoff note. The packet therefore contains one liar-like reconstruction, two rough corroborations, one neighboring exact-but-later camera still, one rounded log entry, and one exact authoritative timestamp. The only exact answer to the question is 2026-02-11 10:08:37.","tags":["red-ledger-binder","exact-time","multi-speaker","camera-log"],"depth":1,"speaker":"court-record","timestamp":"2026-02-11T10:08:37"},{"id":"e2","text":"Speaker A statement excerpt: she would not commit to a second but insisted the binder left a little after 10:08 and before the 10:10 call sequence everyone kept discussing. Her recollection is imprecise, but it points toward the exact time rather than away from it.","tags":["red-ledger-binder","approximate-time","speaker-a"],"depth":2,"speaker":"speaker_a"},{"id":"e3","text":"Speaker B reconstruction note: he swore the binder left at 10:11 exactly because that was the time beside his movement note, even though later questioning established that the note likely reflected handoff receipt rather than doorway exit. No other witness corroborated his exact minute claim.","tags":["red-ledger-binder","unsupported-time","speaker-b"],"depth":2,"speaker":"speaker_b","timestamp":"2026-02-11T10:11:00"},{"id":"e4","text":"Speaker C testimony: the binder was already gone before the 10:10 radio check and after the second witness swap, which indirectly supports an earlier time window but does not itself provide precision down to the second.","tags":["red-ledger-binder","coarse-time","speaker-c"],"depth":2,"speaker":"speaker_c"},{"id":"e5","text":"Speaker D testimony: she believed the binder moved in the 10:08 minute because she saw the prep-room door still being held as the hallway was not yet in its 10:09 churn. That statement lines up with Speaker A and the interior camera without copying either one exactly.","tags":["red-ledger-binder","approximate-time","speaker-d"],"depth":2,"speaker":"speaker_d"},{"id":"e6","text":"Rounded movement log: the clerk's administrative sheet says 10:09, which later proved to be a rounded minute-level notation entered after the movement, not a contemporaneous second-level capture.","tags":["red-ledger-binder","rounded-time","paper-log"],"depth":3,"timestamp":"2026-02-11T10:09:00"},{"id":"e7","text":"Hallway still frame: the binder is visible beyond the prep-room threshold at 10:08:41, which is close enough to the true answer to pollute retrieval while still being later than the first departure moment.","tags":["red-ledger-binder","neighbor-timestamp","hallway-camera"],"depth":3,"timestamp":"2026-02-11T10:08:41"},{"id":"e8","text":"Phone log to chambers: a related call began at 10:10:02. Several speakers used it as an anchor, but all it really proves is that the binder had already moved before the call finished entering the record.","tags":["red-ledger-binder","neighbor-timestamp","phone-log"],"depth":3,"timestamp":"2026-02-11T10:10:02"},{"id":"e9","text":"Systems technician note: the prep-room camera had been synchronized that morning and its export was treated by the panel as the controlling time source when oral recollections and rounded logs diverged.","tags":["red-ledger-binder","authoritative-context","camera-sync"],"depth":4,"timestamp":"2026-02-11T07:12:00"},{"id":"e10","text":"Panel summary: half the spoken chronology was unsupported or overconfident, half converged on an earlier movement during the 10:08 minute, and only the raw camera export preserved the exact second of departure.","tags":["red-ledger-binder","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1"],"notes":"Quartet-style recall@5 breaker with one liar-like witness, two indirect corroborators, and many nearby timestamps competing with the exact camera record."}
{"id":"quartet-r5-002","query":"What was the exact amount on the disputed maintenance reimbursement check?","adversary_type":"speaker_swap","entries":[{"id":"e1","text":"Transcript of the co-op board reimbursement dispute, pages 88 through 142. Speaker A, the treasurer, repeatedly said the disputed maintenance reimbursement check was 'forty-two and some change,' and she kept anchoring that memory to the fact that it stayed under the informal $50 board review threshold. Speaker B, the property manager, was much more theatrical and claimed the check was 'exactly forty-five dollars' because, as he put it, he remembered being annoyed that it looked too clean and suspiciously round. Under cross-examination he admitted he had actually learned that number from a board-room retelling weeks later. Speaker C, the assistant bookkeeper, never quoted an exact amount but said the reimbursement sat 'in the low forties, clearly under fifty, and lower than the elevator repair petty cash line from the same batch.' Speaker D, the resident witness, said she remembered the number starting with forty-two because she had glanced at the carbon copy, though she also admitted she might have been conflating it with a nearby receipt. The exhibit stack made things worse rather than better. One status call memo rounded the reimbursement to about $42. Another board summary described it as a sub-fifty maintenance line. A handwritten draft ledger showed $42.70 because the last two cents had been entered later from the cleared check image. Someone else kept citing the total batch amount, which was over one hundred dollars and irrelevant to the single check question. The decisive evidence was the deposited check image and corrected reimbursement ledger, both of which recorded the disputed maintenance reimbursement check at exactly $42.72. Speaker A and Speaker D both indirectly corroborated the 'forty-two and change' neighborhood without matching each other word for word. Speaker B's exact forty-five-dollar claim stood alone and mapped more cleanly to later gossip than to any document. The hearing therefore offers several true but coarser statements, one unsupported neat number, one dangerous near-exact draft amount, and one exact posted value. The only precise answer is $42.72.","tags":["maintenance-reimbursement-check","exact-amount","multi-speaker","hearing"],"depth":1,"speaker":"board-record"},{"id":"e2","text":"Speaker A testimony: the check was forty-two and some change, definitely below fifty, and small enough that nobody cared about the cents until the dispute turned adversarial. Her memory is not exact, but it lives in the same value neighborhood as the true amount.","tags":["maintenance-reimbursement-check","approximate-amount","speaker-a"],"depth":2,"speaker":"speaker_a"},{"id":"e3","text":"Speaker B statement: the amount was exactly $45.00. No one else corroborated this, and later questioning showed he likely absorbed that number from a board-room shorthand retelling rather than from the check itself.","tags":["maintenance-reimbursement-check","unsupported-amount","speaker-b"],"depth":2,"speaker":"speaker_b"},{"id":"e4","text":"Speaker C bookkeeping recollection: the reimbursement sat in the low forties, under the board threshold, and below a different petty cash line from the same week. That statement agrees with the scale and direction of the true amount without preserving the cents.","tags":["maintenance-reimbursement-check","range-description","speaker-c"],"depth":2,"speaker":"speaker_c"},{"id":"e5","text":"Speaker D resident recollection: she thought the check started with forty-two and remembered that because she compared it against a separate hallway repair receipt. The comparison may be imperfect, but the first two digits align with the exact value rather than the unsupported forty-five-dollar claim.","tags":["maintenance-reimbursement-check","approximate-amount","speaker-d"],"depth":2,"speaker":"speaker_d"},{"id":"e6","text":"Status call memo: the reimbursement was discussed as about $42, which later became a sticky rounded shorthand in board conversation.","tags":["maintenance-reimbursement-check","rounded-amount","memo"],"depth":3},{"id":"e7","text":"Draft ledger line: one handwritten version showed $42.70 before the final cents were reconciled from the deposited image, making it a near-exact trap that looked official enough to mislead multiple participants.","tags":["maintenance-reimbursement-check","near-exact","draft-ledger"],"depth":3},{"id":"e8","text":"Corrected reimbursement ledger: the exact amount on the disputed maintenance reimbursement check was $42.72 after the final two cents were posted from the cleared instrument image.","tags":["maintenance-reimbursement-check","exact-amount","ledger"],"depth":3},{"id":"e9","text":"Batch summary: the maintenance reimbursement was part of a larger reimbursement cluster that totaled well over one hundred dollars, which several speakers accidentally pulled into the discussion even though it was not the single-check value in dispute.","tags":["maintenance-reimbursement-check","neighbor-amount","batching"],"depth":4},{"id":"e10","text":"Panel conclusion: two speakers loosely corroborated the forty-two-and-change region, one speaker described the same scale without naming cents, and one speaker floated an unsupported clean number that did not survive contact with the document trail.","tags":["maintenance-reimbursement-check","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1","e8"],"notes":"Quartet-style financial recall@5 breaker with one liar-like exact claim, multiple indirect corroborators, and a near-exact worksheet trap."}
{"id":"quartet-r5-003","query":"What was the exact date the marsh habitat permit became effective?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Environmental board hearing transcript, long-form permit chronology dispute. Speaker A, the wetlands analyst, said the marsh habitat permit became effective 'in late spring of 2021' and repeatedly tied it to the beginning of the mosquito-control restrictions. Speaker B, the project consultant, confidently insisted the permit took effect 'at the start of summer, June first or thereabouts,' but later admitted he was compressing several steps together: board approval, permit issuance, effective date, and field crew mobilization. Speaker C, the local ecologist, pushed back in a different direction, saying the permit was definitely active before the June survey walk and after the final May vote, but she would not overstate the exact day because she remembered sequence more strongly than date. Speaker D, the neighboring landowner, testified that he received his notice in what he called 'the back half of May' and remembered that because he was away during Memorial Day week, though he conceded that notice receipt was not identical to the permit's legal effective date. The record accumulated many true temporal fragments: board vote in May, issuance paperwork near the end of May, field work in early June, and later summaries calling the permit a late-spring or Q2 event. The decisive document was the permit face itself, which stated an effective date of 2021-05-24. Speaker C's 'before June survey, after May vote' and Speaker D's 'back half of May notice' indirectly corroborate the right region without matching the exact date, while Speaker B's cleaner June-first story sounds memorable but is not supported by the permit. The packet also includes a handbook update dated 2021-06-02 and a field mobilization log dated 2021-05-27, both near enough to pollute retrieval. The exact effective date of the marsh habitat permit is 2021-05-24.","tags":["marsh-habitat-permit","exact-date","multi-speaker","permit"],"depth":1,"speaker":"board-record","timestamp":"2021-05-24"},{"id":"e2","text":"Speaker A chronology: late spring 2021, after the board vote and before the June restrictions messaging. Broadly true, but coarser than the permit face.","tags":["marsh-habitat-permit","seasonal-time","speaker-a"],"depth":2,"speaker":"speaker_a"},{"id":"e3","text":"Speaker B reconstruction: effective at the start of summer, basically June 1. No document or other speaker directly corroborated that exact formulation.","tags":["marsh-habitat-permit","unsupported-time","speaker-b"],"depth":2,"speaker":"speaker_b","timestamp":"2021-06-01"},{"id":"e4","text":"Speaker C sequence memory: active before the June survey walk and after the final May vote. This narrows the window substantially but still does not state the exact date.","tags":["marsh-habitat-permit","coarse-time","speaker-c"],"depth":2,"speaker":"speaker_c"},{"id":"e5","text":"Speaker D notice memory: he received notice in the back half of May and associated it with the week before his Memorial Day travel. Useful corroboration for the right region, but not the legal effective date itself.","tags":["marsh-habitat-permit","coarse-time","speaker-d"],"depth":2,"speaker":"speaker_d"},{"id":"e6","text":"Permit face and legal stamp: the marsh habitat permit became effective on 2021-05-24.","tags":["marsh-habitat-permit","exact-date","permit-face"],"depth":3,"timestamp":"2021-05-24"},{"id":"e7","text":"Field mobilization log: crews were briefed on 2021-05-27, which several witnesses later treated as if it were the operative start date.","tags":["marsh-habitat-permit","neighbor-date","field-log"],"depth":3,"timestamp":"2021-05-27"},{"id":"e8","text":"Policy handbook update: the permit guidance page was refreshed on 2021-06-02 after the permit had already become effective.","tags":["marsh-habitat-permit","neighbor-date","docs"],"depth":3,"timestamp":"2021-06-02"},{"id":"e9","text":"Summary note: the permit was a Q2 2021 wetlands action and part of the late-spring habitat package, which is true but deliberately coarser than the exact legal effective date.","tags":["marsh-habitat-permit","quarter-summary","summary"],"depth":4},{"id":"e10","text":"Panel finding: one speaker supplied an unsupported cleaner story, two others converged on the late-May/pre-June window, and the permit face resolved the chronology exactly.","tags":["marsh-habitat-permit","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1","e6"],"notes":"Quartet-style permit chronology breaker with one overconfident false narrative, multiple indirect corroborators, and exact effective date buried among nearby legal and operational dates."}
{"id":"top5-killer-001","query":"What was the exact timestamp when the east stairwell badge event was logged?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Raw controller export, later admitted after disputes about notebook rounding: the east stairwell badge event was logged at exactly 2026-01-18 14:12:37.","tags":["east-stairwell-badge","exact-time","controller"],"depth":1,"timestamp":"2026-01-18T14:12:37"},{"id":"e2","text":"Security notebook entry: the stairwell access was written as 14:12 because the officer rounded every event to the minute unless a later report required more precision.","tags":["east-stairwell-badge","rounded-time","notebook"],"depth":2,"timestamp":"2026-01-18T14:12:00"},{"id":"e3","text":"Desk summary memo: someone later copied the event as 14:13 after lining it up against the radio relay instead of the raw badge controller export.","tags":["east-stairwell-badge","rounded-time","memo"],"depth":2,"timestamp":"2026-01-18T14:13:00"},{"id":"e4","text":"Camera still packet: the subject is visible entering the stairwell corridor at 14:12:41, which is close to the badge event but not the same instant as the controller log.","tags":["east-stairwell-badge","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-01-18T14:12:41"},{"id":"e5","text":"Radio relay line: the guard called in the stairwell movement at 14:12:54, creating a second authoritative-looking but later time in the record.","tags":["east-stairwell-badge","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-01-18T14:12:54"},{"id":"e6","text":"Witness A: 'It happened during the 2:12 minute, definitely before the 2:13 radio call and after the printer jam was cleared.' That statement is compatible with the true time but not exact.","tags":["east-stairwell-badge","coarse-time","witness"],"depth":3},{"id":"e7","text":"Witness B: 'I remember 2:13 because that was the number on the hallway note.' This overweights the later rounded memo and conflicts with the raw export.","tags":["east-stairwell-badge","unsupported-time","witness"],"depth":3,"timestamp":"2026-01-18T14:13:00"},{"id":"e8","text":"Maintenance sync record: the badge controller and security cameras were synchronized at 06:58 that morning, which makes second-level disagreement between those systems less likely than notebook rounding error.","tags":["east-stairwell-badge","clock-sync","maintenance"],"depth":3,"timestamp":"2026-01-18T06:58:00"},{"id":"e9","text":"Incident recap: counsel kept using phrases like 'shortly after 2:12,' 'before 2:13,' and 'during the middle of the minute,' all of which are true enough to sound persuasive while failing to preserve the exact second.","tags":["east-stairwell-badge","coarse-time","recap"],"depth":4},{"id":"e10","text":"Draft chronology error: an early summary briefly listed 14:12:31 after someone misread the export, adding a near-exact but still wrong time that stayed in the email chain for a day.","tags":["east-stairwell-badge","near-exact","draft-error"],"depth":4,"timestamp":"2026-01-18T14:12:31"},{"id":"e11","text":"Panel note: multiple records place the event in the right minute, but only one record preserves the exact second without being a derivative summary of another source.","tags":["east-stairwell-badge","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1"],"notes":"Top-5 killer with one exact controller timestamp buried among many plausible rounded and near-exact nearby times."}
{"id":"top5-killer-002","query":"What was the exact reimbursement amount on the courier receipt?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Corrected finance ledger: the courier receipt reimbursement posted at exactly $19.75 after tax and after matching the image to the statement export.","tags":["courier-reimbursement","exact-amount","ledger"],"depth":1},{"id":"e2","text":"Expense chat: 'Call it about twenty dollars, it is one of the tiny ones.' This is directionally true but not exact.","tags":["courier-reimbursement","approximate-amount","chat"],"depth":2},{"id":"e3","text":"Draft worksheet: the clerk first wrote $19.70 because the last digits on the receipt image were blurry until the higher-resolution scan arrived.","tags":["courier-reimbursement","near-exact","worksheet"],"depth":2},{"id":"e4","text":"Status summary: the line item was grouped with sub-$20 reimbursements and later remembered as a twenty-dollar courier charge.","tags":["courier-reimbursement","rounded-description","summary"],"depth":2},{"id":"e5","text":"Expense policy note: reimbursements under $20 were frequently discussed as 'twenty-dollar items' during triage even when the cents mattered for exact reconciliation.","tags":["courier-reimbursement","policy-context"],"depth":3},{"id":"e6","text":"Receipt scan note: one reviewer thought the exact total was $19.72 before zooming further, because the five resembled a two on the initial attachment.","tags":["courier-reimbursement","near-exact","scan-note"],"depth":3},{"id":"e7","text":"Final correction email: the earlier rounded and near-exact variants were replaced once the tax-inclusive receipt line was confirmed at $19.75.","tags":["courier-reimbursement","exact-amount","email"],"depth":3},{"id":"e8","text":"Batching report: the courier reimbursement sat inside a $58.93 petty-expense cluster, causing one manager to confuse the batch subtotal with the single exact reimbursement.","tags":["courier-reimbursement","neighbor-amount","batch"],"depth":3},{"id":"e9","text":"Hearing shorthand: counsel alternated between 'under twenty,' 'nineteen-seventy-something,' and 'basically twenty,' producing several semantically sticky but non-identical values.","tags":["courier-reimbursement","rounded-dialogue","courtroom"],"depth":4},{"id":"e10","text":"Worksheet revision history also contains $19.74 and $19.76 as fleeting candidate values before the receipt image was reconciled properly, making the exact cents unusually easy to lose.","tags":["courier-reimbursement","near-exact","revision-history"],"depth":4},{"id":"e11","text":"Review board note: nearly every non-ledger record preserved the right scale while eroding the cents, which is exactly why the exact reimbursement amount has to come from the corrected ledger or the matching correction email.","tags":["courier-reimbursement","agreement-summary","board"],"depth":4}],"relevant_ids":["e1","e7"],"notes":"Top-5 killer with many rounded and near-exact monetary distractors around one exact small reimbursement value."}
{"id":"top5-killer-003","query":"What was the exact average temperature recorded at the Central Park station on 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Station worksheet excerpt: the Central Park daily average on 2023-07-14 was exactly 81.7F.","tags":["central-park","exact-temperature","station"],"depth":1,"timestamp":"2023-07-14"},{"id":"e2","text":"Climate hearing shorthand: the park was in the low eighties that day, which is a fair oral summary but not the exact station value.","tags":["central-park","rounded-temperature","hearing"],"depth":2},{"id":"e3","text":"Campaign slide: the day was rounded to 82F for readability in a climate narrative deck.","tags":["central-park","rounded-temperature","slide"],"depth":2},{"id":"e4","text":"LaGuardia comparison chart: the airport sat higher on the same date, and one witness later imported that airport warmth into their park recollection by mistake.","tags":["laguardia","neighbor-station","chart"],"depth":2},{"id":"e5","text":"Park station methodology note: the exact daily average was preserved with decimal precision even though public summaries often rounded to whole numbers or broad categories like low eighties.","tags":["central-park","methodology","notes"],"depth":3},{"id":"e6","text":"Witness A: 'I remember the park being about 82, maybe a shade under, while the airport was a bit warmer.' This is close, but not exact.","tags":["central-park","approximate-temperature","witness"],"depth":3},{"id":"e7","text":"Weather digest: the date was folded into a string of warm July days and described as an early-wave heat example, which is contextually true but numerically coarse.","tags":["central-park","climate-context","digest"],"depth":3},{"id":"e8","text":"Draft infographic: someone labeled the park line 81.8F because the visual rounded from an intermediate export rather than from the final worksheet.","tags":["central-park","near-exact","infographic"],"depth":3},{"id":"e9","text":"Monthly anomaly sheet: July 2023 was warm overall and the single day sat above older normals, but the sheet's purpose was anomaly comparison, not exact daily recall.","tags":["central-park","climate-context","monthly"],"depth":4},{"id":"e10","text":"Station pair memo: several readers later remembered only that both Central Park and LaGuardia were 'around the low eighties,' which preserved the general picture while dissolving the station-specific decimal.","tags":["central-park","laguardia","coarse-summary"],"depth":4},{"id":"e11","text":"Correction line: the precise Central Park figure remained 81.7F even after campaign materials and witness summaries generated multiple nearby but less exact values.","tags":["central-park","exact-temperature","correction"],"depth":4}],"relevant_ids":["e1","e11"],"notes":"Top-5 killer weather case with many nearby rounded and station-confused values surrounding one exact station reading."}
{"id":"top5-killer-004","query":"What was the exact effective date of the wetland access permit?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Permit face and legal stamp: the wetland access permit became effective on 2021-05-24.","tags":["wetland-access-permit","exact-date","permit"],"depth":1,"timestamp":"2021-05-24"},{"id":"e2","text":"Board vote recap: the permit package cleared in May, which is true but conflates approval with legal effective date.","tags":["wetland-access-permit","month-summary","board"],"depth":2},{"id":"e3","text":"Consultant testimony: 'It was basically a June 1 start,' a claim that sounds clean and memorable but is not corroborated by the permit face.","tags":["wetland-access-permit","unsupported-date","testimony"],"depth":2,"timestamp":"2021-06-01"},{"id":"e4","text":"Field mobilization note: crews were briefed on 2021-05-27, which later got mistaken for the effective date in several oral summaries.","tags":["wetland-access-permit","neighbor-date","field-log"],"depth":2,"timestamp":"2021-05-27"},{"id":"e5","text":"Handbook update: permit guidance was refreshed on 2021-06-02 after the permit had already become effective.","tags":["wetland-access-permit","neighbor-date","docs"],"depth":2,"timestamp":"2021-06-02"},{"id":"e6","text":"Landowner memory: notice arrived in the back half of May, which supports the broad time region while failing to pin the legal date exactly.","tags":["wetland-access-permit","coarse-time","witness"],"depth":3},{"id":"e7","text":"Ecologist testimony: the permit was active before the June survey and after the final May vote. This narrows the window but is still coarser than the exact legal effective date.","tags":["wetland-access-permit","coarse-time","testimony"],"depth":3},{"id":"e8","text":"Summary packet: the action was repeatedly described as a late-spring or Q2 2021 wetlands permit, both of which are accurate but less precise than the exact date.","tags":["wetland-access-permit","quarter-summary","summary"],"depth":3},{"id":"e9","text":"Clerk draft error: one early chronology listed 2021-05-25 after someone copied the issuance line instead of the effective-date line, creating a near-date trap.","tags":["wetland-access-permit","near-date","draft"],"depth":4,"timestamp":"2021-05-25"},{"id":"e10","text":"Correction memo: the exact effective date remained 2021-05-24 after the board packet, field schedule, and handbook refresh were untangled from the permit face itself.","tags":["wetland-access-permit","exact-date","correction"],"depth":4,"timestamp":"2021-05-24"},{"id":"e11","text":"Panel conclusion: nearly every witness described the right sequence, but the only exact legal effective date came from the permit face and the correction memo that later clarified it.","tags":["wetland-access-permit","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1","e10"],"notes":"Top-5 killer permit chronology case with many adjacent legal and operational dates crowding the exact effective date."}
{"id":"top5-killer-005","query":"What was the exact annual mean methane concentration cited for 2004?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Report table citation: the exact annual mean methane concentration for 2004 was 1774.62 ppb.","tags":["methane-2004","exact-annual-mean","report"],"depth":1,"timestamp":"2004-01-01"},{"id":"e2","text":"Debate shorthand: the 2004 mean sat in the high-1770s ppb band, a broad description that is directionally correct but not exact.","tags":["methane-2004","rounded-summary","debate"],"depth":2},{"id":"e3","text":"Slide deck: one visual rounded the annual mean to 1775 ppb for readability.","tags":["methane-2004","rounded-summary","slides"],"depth":2},{"id":"e4","text":"Public-facing note: another summary described the same value as roughly 1.78 ppm, which is a converted approximation rather than the exact report-table expression.","tags":["methane-2004","converted-summary","public"],"depth":2},{"id":"e5","text":"Cross-exam note: a questioner repeatedly confused neighboring annual values and asked whether 2004 was the 1776-ish year or the 1774-ish year, illustrating how the nearby cluster pollutes memory.","tags":["methane-2004","nearby-values","dialogue"],"depth":3},{"id":"e6","text":"Handout margin: one copy carried 1774.6 ppb after truncation, which is very close to the exact value but still not the precise report figure.","tags":["methane-2004","near-exact","handout"],"depth":3},{"id":"e7","text":"Trend memo: the annual mean was repeatedly discussed in relation to neighboring high-1770s years, making the cluster more salient than the exact decimal.","tags":["methane-2004","cluster-context","memo"],"depth":3},{"id":"e8","text":"Methodology note: the report preserves the exact annual mean with two decimals even though derivative summaries round, truncate, or convert it for audience convenience.","tags":["methane-2004","methodology","notes"],"depth":4},{"id":"e9","text":"Audience summary: some attendees left remembering 1774.6, others 1775, and others only that it belonged to a mid-2000s high plateau in methane terms.","tags":["methane-2004","rounded-memory","audience"],"depth":4},{"id":"e10","text":"Correction note: the exact report value is 1774.62 ppb, not the rounded, truncated, or converted variants that kept circulating after the debate.","tags":["methane-2004","exact-annual-mean","correction"],"depth":4},{"id":"e11","text":"Moderator summary: both scientists accepted the same authoritative table, but their rhetoric created a cloud of nearby defensible numbers around the exact annual mean.","tags":["methane-2004","agreement-summary","summary"],"depth":4}],"relevant_ids":["e1","e10"],"notes":"Top-5 killer climate case with rounded, truncated, converted, and neighboring-year methane values crowding the exact annual mean."}
{"id":"top5-killer-006","query":"What was the exact timestamp on the witness phone log?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Carrier record: the witness phone log shows the call beginning at exactly 2026-03-11 19:43:27.","tags":["witness-phone-log","exact-time","carrier"],"depth":1,"timestamp":"2026-03-11T19:43:27"},{"id":"e2","text":"Reception desk note: the witness called at about 7:45 p.m., a rounded notation that later drifted into several summaries.","tags":["witness-phone-log","rounded-time","desk-note"],"depth":2,"timestamp":"2026-03-11T19:45:00"},{"id":"e3","text":"Carrier printout header on one draft looked like 19:43 because seconds were omitted, creating a cleaner but coarser timestamp than the final admitted exhibit.","tags":["witness-phone-log","rounded-time","draft-printout"],"depth":2,"timestamp":"2026-03-11T19:43:00"},{"id":"e4","text":"Hallway camera note: motion near the desk was logged at 19:44:02, which several participants mistakenly treated as the call time because it sat so close to the carrier record.","tags":["witness-phone-log","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-03-11T19:44:02"},{"id":"e5","text":"Marshal recollection: the call happened a little before quarter to eight and after sunset, which is broad sequence truth rather than exact timing.","tags":["witness-phone-log","coarse-time","witness"],"depth":3},{"id":"e6","text":"Dispatch relay record: the witness report was repeated at 19:43:48, making it another authoritative-looking but still later time in the packet.","tags":["witness-phone-log","neighbor-timestamp","dispatch"],"depth":3,"timestamp":"2026-03-11T19:43:48"},{"id":"e7","text":"Carrier correction memo: the exact outgoing call began at 19:43:27 and not at the rounded 19:43 or the later relay and camera times that clustered around it.","tags":["witness-phone-log","exact-time","correction"],"depth":3,"timestamp":"2026-03-11T19:43:27"},{"id":"e8","text":"Hearing packet shorthand: counsel repeatedly used phrases like 'during the 7:43 minute,' 'before 7:44,' and 'around quarter to eight,' all of which are plausible enough to crowd the exact second.","tags":["witness-phone-log","coarse-time","dialogue"],"depth":4},{"id":"e9","text":"Timeline chart also included a neighboring 19:42:58 text message to the witness's brother, which some readers later fused with the phone call because both sat inside the same minute-level story.","tags":["witness-phone-log","neighbor-timestamp","text-message"],"depth":4,"timestamp":"2026-03-11T19:42:58"},{"id":"e10","text":"Evidence note: the provider log was treated as controlling because witness recollections, hallway motion, and dispatch relays all preserved only approximate sequence around the call.","tags":["witness-phone-log","authoritative-context","evidence"],"depth":4},{"id":"e11","text":"Summary: many timestamps in the record are true and close, but only one is the exact start time on the witness phone log itself.","tags":["witness-phone-log","agreement-summary","summary"],"depth":4}],"relevant_ids":["e1","e7"],"notes":"Top-5 killer phone-log case with several highly similar authoritative and quasi-authoritative neighboring timestamps."}
{"id":"top5-killer-007","query":"What was the exact timestamp when the west loading dock camera first showed the silver case crossing the threshold?","adversary_type":"time_swap","entries":[{"id":"e1","text":"The west loading dock camera export, which became exhibit 22 after three rounds of authentication argument, records the first frame where the silver case fully crosses the threshold at exactly 2026-08-21 06:14:23. Every later summary in the packet drifted away from this second-level value in one direction or another because the morning was reconstructed through radio chatter, clip stills, rounded logs, and the testimony of people who kept anchoring memory to the forklift alarm that followed a few moments later.","tags":["silver-case-threshold","exact-time","camera-export"],"depth":1,"timestamp":"2026-08-21T06:14:23"},{"id":"e2","text":"Shift worksheet entry: the crossing was logged as 06:14 because minute-level notation was the default on the worksheet and nobody expected a second-level dispute at the time.","tags":["silver-case-threshold","rounded-time","worksheet"],"depth":2,"timestamp":"2026-08-21T06:14:00"},{"id":"e3","text":"Supervisor recap email: the silver case crossed at about 6:15 a.m., a rounded value that later became sticky because the email was easier to search than the raw video export.","tags":["silver-case-threshold","rounded-time","email"],"depth":2,"timestamp":"2026-08-21T06:15:00"},{"id":"e4","text":"Hallway still image: a frame stamped 06:14:27 shows the case already inside the dock line, creating a near-neighbor timestamp that looks authoritative but is later than the first threshold crossing.","tags":["silver-case-threshold","neighbor-timestamp","still"],"depth":2,"timestamp":"2026-08-21T06:14:27"},{"id":"e5","text":"Radio record: the forklift alarm was relayed at 06:14:31, and several witnesses incorrectly tied the case movement to that broadcast rather than to the camera crossing itself.","tags":["silver-case-threshold","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-08-21T06:14:31"},{"id":"e6","text":"Witness A described the event as happening 'during the 6:14 minute, before the alarm call,' which supports the right time region but not the exact second.","tags":["silver-case-threshold","coarse-time","witness-a"],"depth":3},{"id":"e7","text":"Witness B insisted on 06:15 because that was the rounded value in the supervisor email she reviewed later, not because she independently saw the threshold crossing.","tags":["silver-case-threshold","unsupported-time","witness-b"],"depth":3,"timestamp":"2026-08-21T06:15:00"},{"id":"e8","text":"Camera synchronization note: the dock camera and yard camera were aligned at 05:02 that morning; defense questioning still tried to imply a clock drift, but no alert or desync event was logged.","tags":["silver-case-threshold","clock-sync","maintenance"],"depth":3,"timestamp":"2026-08-21T05:02:00"},{"id":"e9","text":"Dispatch note: a neighboring gate event was recorded at 06:14:18, close enough that one draft chronology briefly copied the wrong line into the threshold column.","tags":["silver-case-threshold","neighbor-timestamp","dispatch"],"depth":3,"timestamp":"2026-08-21T06:14:18"},{"id":"e10","text":"Draft chronology revision: an early spreadsheet listed 06:14:24 because the reviewer clicked the second frame after threshold instead of the first frame across the line.","tags":["silver-case-threshold","near-exact","draft"],"depth":4,"timestamp":"2026-08-21T06:14:24"},{"id":"e11","text":"Hearing shorthand repeatedly used 'just after 6:14,' 'before the alarm,' and 'around quarter past six,' all of which are semantically sticky but less precise than the export.","tags":["silver-case-threshold","coarse-time","dialogue"],"depth":4},{"id":"e12","text":"Corrective order: the panel distinguished between the first crossing frame, later stills, alarm relay timing, and rounded worksheet entries, and retained only the export value as controlling for the precise question.","tags":["silver-case-threshold","agreement-summary","order"],"depth":4}],"relevant_ids":["e1"],"notes":"Crowded top-5 killer with multiple same-minute neighbors, a near-exact draft error, and several authoritative-looking but later timestamps."}
{"id":"top5-killer-008","query":"What was the exact amount on the emergency lodging reimbursement?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The corrected reimbursement ledger and matching bank export agree that the emergency lodging reimbursement was exactly $117.38. That value later became surprisingly hard to recover because almost every adjacent document rounded it, bucketed it, or blended it with neighboring lodging and meal charges from the same disrupted travel window.","tags":["emergency-lodging-reimbursement","exact-amount","ledger"],"depth":1},{"id":"e2","text":"Travel coordinator chat: 'It was about one-eighteen, one of the bigger but still routine emergency lodging claims.' This is broadly right but not exact.","tags":["emergency-lodging-reimbursement","rounded-amount","chat"],"depth":2},{"id":"e3","text":"Expense worksheet draft: an early pass showed $117.30 before taxes and service adjustments were reconciled into the final reimbursement line.","tags":["emergency-lodging-reimbursement","near-exact","worksheet"],"depth":2},{"id":"e4","text":"Policy memo: emergency lodging items in the low one-hundreds were often summarized verbally as 'about one-twenty' during review meetings because the threshold discussions cared about scale, not cents.","tags":["emergency-lodging-reimbursement","rounded-description","policy"],"depth":2},{"id":"e5","text":"Batch sheet: the reimbursement sat next to a $117.83 hotel parking line, and multiple reviewers later confused the two because they were in the same cluster and differed by only a few cents.","tags":["emergency-lodging-reimbursement","neighbor-amount","batch"],"depth":3},{"id":"e6","text":"Receipt image note: one reviewer briefly transcribed the line as $117.88 before the decimal and trailing digits were verified against the statement export.","tags":["emergency-lodging-reimbursement","near-exact","scan-error"],"depth":3},{"id":"e7","text":"Manager summary: the reimbursement was in the one-hundred-seventeen-dollar range and below the manual hotel escalation threshold. Useful context, but still less exact than the ledger.","tags":["emergency-lodging-reimbursement","range-description","summary"],"depth":3},{"id":"e8","text":"Correction email: the exact emergency lodging reimbursement was $117.38 and not the rounded, neighbor-line, or scan-error values that kept circulating during review.","tags":["emergency-lodging-reimbursement","exact-amount","correction"],"depth":3},{"id":"e9","text":"Dispute call transcript: participants bounced among 'about one-eighteen,' 'one seventeen and change,' and 'that larger hotel item,' producing several semantically aligned but distinct retrieval traps.","tags":["emergency-lodging-reimbursement","rounded-dialogue","call"],"depth":4},{"id":"e10","text":"Bank export cluster also contained a $118.00 rounded hold release and a $117.83 neighboring parking reimbursement, both of which looked suspiciously answer-like in shallow retrieval.","tags":["emergency-lodging-reimbursement","neighbor-amount","bank-export"],"depth":4},{"id":"e11","text":"Board note: nearly every secondary record preserved the same monetary neighborhood, but only the corrected ledger and correction email preserved the exact cents for the lodging line itself.","tags":["emergency-lodging-reimbursement","agreement-summary","board"],"depth":4}],"relevant_ids":["e1","e8"],"notes":"Crowded top-5 killer amount case with same-neighborhood values, near-exact scan errors, and adjacent clustered amounts."}
{"id":"top5-killer-009","query":"What was the exact annual mean CO2 concentration cited for 1988?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The cited annual mean CO2 concentration for 1988 in the report table was exactly 351.57 ppm. That line later got buried under surrounding discussion of the late-1980s acceleration narrative, the famous summer testimony period, and a stack of nearby rounded and comparative values that all sounded answer-like without preserving the exact annual mean.","tags":["co2-1988","exact-annual-mean","report"],"depth":1},{"id":"e2","text":"Debate shorthand: the 1988 annual mean sat in the low 352 ppm range, which is directionally fine but still not the precise reported number.","tags":["co2-1988","rounded-summary","debate"],"depth":2},{"id":"e3","text":"Slide deck: one chart labeled the year 351.6 ppm after rounding to a single decimal place.","tags":["co2-1988","near-exact","slides"],"depth":2},{"id":"e4","text":"Public-facing summary: another document called it about 352 ppm because the writer prioritized readability over table fidelity.","tags":["co2-1988","rounded-summary","public"],"depth":2},{"id":"e5","text":"Cross-examination note: a witness repeatedly toggled between the famous 1988 summer testimony context and the annual mean itself, causing neighboring monthly and annual numbers to drift together in memory.","tags":["co2-1988","neighbor-values","dialogue"],"depth":3},{"id":"e6","text":"Handout margin: an intern copied the figure as 351.75 ppm in one draft packet, introducing a near-exact but wrong decimal pair that later had to be corrected.","tags":["co2-1988","near-exact","draft"],"depth":3},{"id":"e7","text":"Trend memo: 1988 was repeatedly compared to nearby annual means in the low 350s, making it easy to remember the cluster correctly while still forgetting the exact table row.","tags":["co2-1988","cluster-context","memo"],"depth":3},{"id":"e8","text":"Methodology note: the authoritative table preserved the exact annual mean at 351.57 ppm, while summary materials rounded, truncated, or rhetorically grouped the year into broader trend language.","tags":["co2-1988","exact-annual-mean","methodology"],"depth":3},{"id":"e9","text":"Audience Q&A: participants asked whether 1988 was the 351.6 year, the almost-352 year, or the summer-spike year, showing how many neighboring but plausible retrieval anchors the record now contains.","tags":["co2-1988","rounded-memory","audience"],"depth":4},{"id":"e10","text":"Comparative appendix: neighboring years in the late 1980s and early 1990s all sat close enough together to make exact annual recall brittle when only rounded narratives were retained.","tags":["co2-1988","context","appendix"],"depth":4},{"id":"e11","text":"Correction line: the exact cited annual mean for 1988 remained 351.57 ppm despite the one-decimal, whole-number, and near-miss draft variants circulating around it.","tags":["co2-1988","exact-annual-mean","correction"],"depth":4}],"relevant_ids":["e1","e8","e11"],"notes":"Crowded top-5 killer climate case with rounded and near-exact yearly CO2 values surrounding the exact annual mean."}
{"id":"top5-killer-010","query":"What was the exact average temperature recorded at LaGuardia on 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The LaGuardia station table recorded the exact 2023-07-14 daily average at 82.6F. That figure sits inside a crowded neighborhood of citywide low-eighties summaries, rounded airport shorthand, and nearby Central Park comparisons that make precise station recall unusually fragile.","tags":["laguardia-2023-07-14","exact-temperature","station"],"depth":1,"timestamp":"2023-07-14"},{"id":"e2","text":"Public comms copy: the airport was described as roughly 83F for readability in a climate-change summary.","tags":["laguardia-2023-07-14","rounded-temperature","public"],"depth":2},{"id":"e3","text":"Witness recollection: the airport was in the eighty-two to eighty-three zone and somewhat warmer than the park. Correct in feel, but still not the exact table value.","tags":["laguardia-2023-07-14","range-description","witness"],"depth":2},{"id":"e4","text":"Central Park comparison memo: one chart placed the park in the low 81s and the airport in the low 82s, which later caused multiple speakers to blur the two stations together into one citywide number.","tags":["laguardia-2023-07-14","neighbor-station","memo"],"depth":2},{"id":"e5","text":"Slide annotation: one presenter rounded the airport line to 82.5F after exporting the chart without the final decimal update, creating a near-exact but still wrong figure.","tags":["laguardia-2023-07-14","near-exact","slide"],"depth":3},{"id":"e6","text":"Station methodology note: the exact LaGuardia figure remained 82.6F, while summaries often collapsed it into whole-degree or station-pair narratives.","tags":["laguardia-2023-07-14","exact-temperature","methodology"],"depth":3},{"id":"e7","text":"Campaign note: the date was folded into a string of warm mid-July days, and some summaries simply called the airport reading 'about eighty-three' because the rhetorical emphasis was on climate trend, not decimal precision.","tags":["laguardia-2023-07-14","rounded-summary","campaign"],"depth":3},{"id":"e8","text":"Airport operations sheet also contained a same-date runway-surface summary with a much higher number, which some readers incorrectly treated as if it were the station average.","tags":["laguardia-2023-07-14","neighbor-value","ops"],"depth":4},{"id":"e9","text":"Monthly anomaly appendix: several days in the same week sat in overlapping low-eighties clusters, making the exact daily airport average easy to replace with a nearby but still plausible warm-day value.","tags":["laguardia-2023-07-14","context","monthly"],"depth":4},{"id":"e10","text":"Correction note: the exact LaGuardia daily average on that date was 82.6F, not 82.5F, 83F, or a merged citywide shorthand that blended airport and park records together.","tags":["laguardia-2023-07-14","exact-temperature","correction"],"depth":4},{"id":"e11","text":"Review summary: almost every surrounding statement captured the right heat story and the right station neighborhood, but only the exact table row and the later correction preserve the precise decimal value.","tags":["laguardia-2023-07-14","agreement-summary","summary"],"depth":4}],"relevant_ids":["e1","e6","e10"],"notes":"Crowded top-5 killer weather case with station confusion, rounded whole-degree summaries, and a near-exact chart export error."}
{"id":"top5-killer-011","query":"What was the exact timestamp when the archive cabinet sensor first alarmed?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Sensor controller export: the archive cabinet sensor first alarmed at exactly 2026-10-04 03:12:19.","tags":["archive-cabinet-alarm","exact-time","controller"],"depth":1,"timestamp":"2026-10-04T03:12:19"},{"id":"e2","text":"Building automation dashboard rounded the alarm into the 03:12 minute because the default operator view suppressed seconds unless a forensic export was requested.","tags":["archive-cabinet-alarm","rounded-time","dashboard"],"depth":2,"timestamp":"2026-10-04T03:12:00"},{"id":"e3","text":"Facilities pager relay fired at 03:12:33, close enough that several operators later remembered the pager time as if it were the first alarm time.","tags":["archive-cabinet-alarm","neighbor-timestamp","pager"],"depth":2,"timestamp":"2026-10-04T03:12:33"},{"id":"e4","text":"Security camera overlay showed the cabinet room door opening at 03:12:27, creating another near-neighbor event that looked authoritative in the later packet.","tags":["archive-cabinet-alarm","neighbor-timestamp","camera"],"depth":2,"timestamp":"2026-10-04T03:12:27"},{"id":"e5","text":"Operator note: 'alarm started during the 03:12 minute before the pager buzzed,' which is accurate in sequence but coarser than the controller export.","tags":["archive-cabinet-alarm","coarse-time","operator"],"depth":3},{"id":"e6","text":"Draft incident timeline listed 03:12:12 after someone copied the pre-alarm polling line rather than the first alarm line, introducing a near-exact but wrong time.","tags":["archive-cabinet-alarm","near-exact","draft"],"depth":3,"timestamp":"2026-10-04T03:12:12"},{"id":"e7","text":"Maintenance sync note: the cabinet sensor controller had been synchronized at 01:11 and no subsequent drift event was recorded before the overnight alarm.","tags":["archive-cabinet-alarm","clock-sync","maintenance"],"depth":3,"timestamp":"2026-10-04T01:11:00"},{"id":"e8","text":"Escalation call started at 03:13:02, which some later summaries treated as the first 'officially noticed' moment despite not being the first alarm.","tags":["archive-cabinet-alarm","neighbor-timestamp","call"],"depth":4,"timestamp":"2026-10-04T03:13:02"},{"id":"e9","text":"Report shorthand repeatedly used 'just after 3:12 a.m.' and 'before the pager,' both of which are true but incomplete if the query asks for the exact timestamp.","tags":["archive-cabinet-alarm","coarse-time","report"],"depth":4},{"id":"e10","text":"Correction memo reaffirmed 03:12:19 as the first alarm time and explained why the dashboard minute, pager relay, and door-open event had all contaminated witness memory.","tags":["archive-cabinet-alarm","exact-time","correction"],"depth":4,"timestamp":"2026-10-04T03:12:19"},{"id":"e11","text":"Panel finding: the record contains several nearby machine timestamps, one rounded operator view, and multiple coarse witness summaries, but only one first-alarm second on the controller export.","tags":["archive-cabinet-alarm","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1","e10"],"notes":"Crowded top-5 killer sensor case with many nearby machine timestamps and one exact first-alarm record."}
{"id":"top5-killer-012","query":"What was the exact annual mean sea level anomaly cited for the report year?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The cited annual mean sea level anomaly in the report year was exactly 91.3 mm.","tags":["sea-level-anomaly","exact-value","report"],"depth":1},{"id":"e2","text":"Hearing shorthand repeatedly called it about 90 mm because the testimony emphasized the larger trend rather than table fidelity.","tags":["sea-level-anomaly","rounded-summary","hearing"],"depth":2},{"id":"e3","text":"One slide rounded the same annual mean to 91 mm for visual cleanliness.","tags":["sea-level-anomaly","rounded-summary","slides"],"depth":2},{"id":"e4","text":"Another summary translated the value into roughly 9.1 cm above baseline, which is conceptually aligned but not the exact report expression.","tags":["sea-level-anomaly","converted-summary","outreach"],"depth":2},{"id":"e5","text":"Cross-exam note: neighboring years were also elevated, and the discussion kept bouncing among low-90s annual anomalies, decadal means, and trend acceleration language.","tags":["sea-level-anomaly","nearby-values","dialogue"],"depth":3},{"id":"e6","text":"Draft handout showed 91.2 mm after a truncation error in one chart export, creating a near-exact but incorrect figure that circulated briefly.","tags":["sea-level-anomaly","near-exact","handout"],"depth":3},{"id":"e7","text":"Methods note: the report preserved 91.3 mm exactly, while derivative materials rounded to whole millimeters or broad centimeter shorthand depending on audience.","tags":["sea-level-anomaly","exact-value","methodology"],"depth":3},{"id":"e8","text":"Audience recap: some attendees left saying 90, others 91, others low nineties, and a few remembered only the trend claim that the value sat among the highest on record.","tags":["sea-level-anomaly","rounded-memory","audience"],"depth":4},{"id":"e9","text":"Context appendix: the anomaly was also discussed relative to baseline windows, acceleration arguments, and neighboring annual peaks, all of which are true context but not the single exact annual value.","tags":["sea-level-anomaly","context","appendix"],"depth":4},{"id":"e10","text":"Correction note: the precise annual mean anomaly remains 91.3 mm, not 90, 91, 91.2, or 'about 9 cm.'","tags":["sea-level-anomaly","exact-value","correction"],"depth":4},{"id":"e11","text":"Moderator note: both sides in the debate accepted the same underlying report table, even as their rhetoric generated a fog of neighboring approximate values around the exact one.","tags":["sea-level-anomaly","agreement-summary","summary"],"depth":4}],"relevant_ids":["e1","e7","e10"],"notes":"Crowded top-5 killer anomaly case with rounded, converted, and near-exact variants surrounding a single exact report value."}
{"id":"extreme-r5-001","query":"What was the exact timestamp on the corridor camera when the black portfolio first crossed the doorway?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Raw corridor camera export: the black portfolio first crossed the doorway at exactly 2026-04-02 08:17:26.","tags":["black-portfolio-doorway","exact-time","camera"],"depth":1,"timestamp":"2026-04-02T08:17:26"},{"id":"e2","text":"Rounded guard note: the crossing happened at 08:17.","tags":["black-portfolio-doorway","rounded-time","guard-note"],"depth":2,"timestamp":"2026-04-02T08:17:00"},{"id":"e3","text":"Dispatcher relay: security referenced the movement at 08:17:39.","tags":["black-portfolio-doorway","neighbor-timestamp","dispatch"],"depth":2,"timestamp":"2026-04-02T08:17:39"},{"id":"e4","text":"Still-frame packet: a later frame showing the portfolio fully inside the corridor is stamped 08:17:31.","tags":["black-portfolio-doorway","neighbor-timestamp","still-frame"],"depth":2,"timestamp":"2026-04-02T08:17:31"},{"id":"e5","text":"Witness A said it happened 'during the 8:17 minute, before the radio call.'","tags":["black-portfolio-doorway","coarse-time","witness-a"],"depth":2},{"id":"e6","text":"Witness B claimed 08:18 exactly because that is what he remembered from a later summary email.","tags":["black-portfolio-doorway","unsupported-time","witness-b"],"depth":2,"timestamp":"2026-04-02T08:18:00"},{"id":"e7","text":"Summary email rounded the event to about 8:18 a.m.","tags":["black-portfolio-doorway","rounded-time","email"],"depth":3,"timestamp":"2026-04-02T08:18:00"},{"id":"e8","text":"Draft chronology mistakenly listed 08:17:24 after someone copied the pre-threshold frame instead of the first crossing frame.","tags":["black-portfolio-doorway","near-exact","draft"],"depth":3,"timestamp":"2026-04-02T08:17:24"},{"id":"e9","text":"Another draft listed 08:17:27 after using the second frame across the line.","tags":["black-portfolio-doorway","near-exact","draft"],"depth":3,"timestamp":"2026-04-02T08:17:27"},{"id":"e10","text":"Phone log to chambers started at 08:17:44 and was repeatedly used as a memory anchor even though it post-dated the crossing.","tags":["black-portfolio-doorway","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-04-02T08:17:44"},{"id":"e11","text":"Maintenance sync note: camera clocks were aligned at 06:11 that morning.","tags":["black-portfolio-doorway","clock-sync","maintenance"],"depth":3,"timestamp":"2026-04-02T06:11:00"},{"id":"e12","text":"Hallway activity memo says the crossing happened just after 8:17 and before 8:18, preserving sequence but not the second.","tags":["black-portfolio-doorway","coarse-time","memo"],"depth":4},{"id":"e13","text":"Panel shorthand kept calling this the 08:17 doorway event.","tags":["black-portfolio-doorway","rounded-time","panel"],"depth":4},{"id":"e14","text":"Correction sheet reaffirmed 2026-04-02 08:17:26 as the first crossing timestamp, not the rounded or neighboring records.","tags":["black-portfolio-doorway","exact-time","correction"],"depth":4,"timestamp":"2026-04-02T08:17:26"},{"id":"e15","text":"Witness C placed it after the elevator chime at 08:17:12 and before the hallway still, which narrows the range but does not resolve the exact second.","tags":["black-portfolio-doorway","coarse-time","witness-c"],"depth":4},{"id":"e16","text":"Administrative index refers only to 'morning corridor movement, 8:17-ish.'","tags":["black-portfolio-doorway","rounded-time","index"],"depth":4}],"relevant_ids":["e1","e14"],"notes":"Extreme crowd-out case with many same-minute and near-second distractors competing against one exact camera timestamp."}
{"id":"extreme-r5-002","query":"What was the exact amount on the hotel incident reimbursement?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Final corrected reimbursement ledger: the hotel incident reimbursement was exactly $89.47.","tags":["hotel-incident-reimbursement","exact-amount","ledger"],"depth":1},{"id":"e2","text":"Coordinator shorthand called it about ninety dollars.","tags":["hotel-incident-reimbursement","rounded-amount","summary"],"depth":2},{"id":"e3","text":"Expense recap described it as high-eighties and below the manual review threshold.","tags":["hotel-incident-reimbursement","range-description","recap"],"depth":2},{"id":"e4","text":"Draft ledger line showed $89.40 before the tax adjustment was reconciled.","tags":["hotel-incident-reimbursement","near-exact","draft"],"depth":2},{"id":"e5","text":"Scan note recorded $89.42 after the clerk misread the final digit on the receipt image.","tags":["hotel-incident-reimbursement","near-exact","scan"],"depth":2},{"id":"e6","text":"Another worksheet entry used $89.74 after two digits were transposed in a manual copy.","tags":["hotel-incident-reimbursement","near-exact","worksheet"],"depth":3},{"id":"e7","text":"Policy memo says reimbursements in the high-eighties to low-nineties were often summarized as ninety-dollar items during oral review.","tags":["hotel-incident-reimbursement","policy-context"],"depth":3},{"id":"e8","text":"Neighboring parking reimbursement in the same batch was $89.83, which later confused reviewers comparing clustered charges.","tags":["hotel-incident-reimbursement","neighbor-amount","batch"],"depth":3},{"id":"e9","text":"Correction email restated the exact hotel incident reimbursement as $89.47 after matching the bank line against the receipt image.","tags":["hotel-incident-reimbursement","exact-amount","correction"],"depth":3},{"id":"e10","text":"Hearing shorthand alternated among eighty-nine and change, about ninety, and high-eighties hotel item, all of which felt semantically close to the true number.","tags":["hotel-incident-reimbursement","rounded-dialogue","hearing"],"depth":4},{"id":"e11","text":"Batch total note listed $267.91 for the cluster, adding another unrelated but authoritative-looking number nearby.","tags":["hotel-incident-reimbursement","neighbor-amount","batch-total"],"depth":4},{"id":"e12","text":"Manager review form clipped the amount to $89.5 in one chart, creating a rounded decimal form that still points into the same numeric neighborhood.","tags":["hotel-incident-reimbursement","rounded-amount","chart"],"depth":4},{"id":"e13","text":"Internal note: the reimbursement definitely did not hit ninety exactly, even though many people summarized it that way.","tags":["hotel-incident-reimbursement","coarse-summary","note"],"depth":4},{"id":"e14","text":"The posted amount remained under ninety and in the eighty-nine-dollar range throughout the review packet.","tags":["hotel-incident-reimbursement","range-description","packet"],"depth":4},{"id":"e15","text":"Only the corrected ledger and the correction email preserve the exact cents without rounding, transposition, or clustering noise.","tags":["hotel-incident-reimbursement","agreement-summary","board"],"depth":4}],"relevant_ids":["e1","e9"],"notes":"Extreme crowd-out amount case with many same-neighborhood and near-exact distractors around one exact reimbursement value."}
{"id":"extreme-r5-003","query":"What was the exact annual mean sea ice extent value cited in the report?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Primary report table, as read into the record during the methods hearing after two earlier witnesses had already muddied the issue with rounded graphics, trend rhetoric, and comparisons to neighboring years: the exact annual mean sea ice extent value cited in the report was 6.84 million square kilometers. That number sounds deceptively simple until one looks at the surrounding packet, where the same year is repeatedly described as roughly 6.8, low 6.8s, upper sixes, about 684 x 10^4 square kilometers, a hair under 6.85, part of the late-decade low cluster, and comparable to several nearby annual means that differ only in the second decimal place. The hearing transcript makes clear that almost everyone in the room was narrating the same underlying reality but at different levels of precision and with different motives: some were simplifying for readability, some were rhetorically emphasizing trend direction, some were carelessly truncating, and some were sliding from annual means into adjacent metrics like seasonal minimum extent or decadal averages. The report table itself remained the only clean source for the exact annual mean value, and that source states 6.84 million square kilometers.","tags":["sea-ice-annual-mean","exact-value","report"],"depth":1},{"id":"e2","text":"Public summary version A described the same annual mean as about 6.8 million square kilometers because the communications team was trying to keep the number visually lightweight and trend-friendly for nontechnical readers.","tags":["sea-ice-annual-mean","rounded-summary","public"],"depth":2},{"id":"e3","text":"Public summary version B called it approximately 6.84 million square kilometers but in context used that figure as part of a sentence about an extended decline, not as a standalone exact citation.","tags":["sea-ice-annual-mean","near-exact","public"],"depth":2},{"id":"e4","text":"A slide deck rounded the annual mean down to 6.8 on one slide and up to 6.9 on another because different designers used different display rules for one decimal place versus whole-tenth shorthand.","tags":["sea-ice-annual-mean","rounded-summary","slides"],"depth":2},{"id":"e5","text":"A draft handout preserved the value as 6.83 after truncation from a spreadsheet export that dropped the last decimal step too early in the formatting chain.","tags":["sea-ice-annual-mean","near-exact","handout"],"depth":2},{"id":"e6","text":"An outreach conversion note rendered the same annual mean as roughly 684 x 10^4 square kilometers, mathematically related but not the report\u2019s exact expression.","tags":["sea-ice-annual-mean","converted-summary","outreach"],"depth":2},{"id":"e7","text":"Witness A described the year as living in the upper-sixes and said the exact decimal was less important than the direction of change compared with the prior decade.","tags":["sea-ice-annual-mean","range-description","witness-a"],"depth":3},{"id":"e8","text":"Witness B kept saying the annual mean was \u201cbasically 6.85\u201d because that was the nearest number he remembered from a chart with tightly clustered neighboring years.","tags":["sea-ice-annual-mean","near-exact","witness-b"],"depth":3},{"id":"e9","text":"Comparative appendix listed neighboring annual means at 6.79, 6.81, 6.86, 6.88, and 6.91, creating a dense neighborhood of plausible values that all belong to the same semantic family.","tags":["sea-ice-annual-mean","neighbor-values","appendix"],"depth":3},{"id":"e10","text":"Cross-examination repeatedly drifted from annual mean extent to annual minimum extent, then to decade-level averages, then back again, so several statements in the record sound numerically relevant while referring to adjacent but distinct measures.","tags":["sea-ice-annual-mean","context","dialogue"],"depth":3},{"id":"e11","text":"One chart note labeled the year 6.840 exactly, which is numerically equivalent but appeared only in a derivative visual generated after the report table had already been finalized.","tags":["sea-ice-annual-mean","equivalent-expression","chart"],"depth":3},{"id":"e12","text":"Audience recap notes captured the value as 6.84, 6.8-ish, just under 6.85, and part of the same low cluster as several surrounding years, all without carefully distinguishing exact citation from verbal paraphrase.","tags":["sea-ice-annual-mean","rounded-memory","audience"],"depth":4},{"id":"e13","text":"A methods memo warned that once the annual mean was folded into a broader trend argument, speakers started substituting low-precision category labels for the exact reported value.","tags":["sea-ice-annual-mean","methodology","memo"],"depth":4},{"id":"e14","text":"A review packet summary used 6.84 million square kilometers but immediately surrounded it with neighboring values and comparative language, making the exact value harder to isolate in retrieval even though it remained correct.","tags":["sea-ice-annual-mean","exact-value","summary"],"depth":4},{"id":"e15","text":"A draft chart incorrectly listed 6.82 after a stale filter selected the wrong row in an annual table, introducing a plausible but unsupported near-miss into the packet history.","tags":["sea-ice-annual-mean","near-exact","draft-error"],"depth":4},{"id":"e16","text":"Another chart export listed 6.87 after the designer copied the neighboring year label into the plotted annotation.","tags":["sea-ice-annual-mean","near-exact","draft-error"],"depth":4},{"id":"e17","text":"Moderator summary: most disagreement in the room concerned narrative framing, but nearly everyone kept the value inside the same very narrow numeric neighborhood, which is why exact recall remains brittle.","tags":["sea-ice-annual-mean","agreement-summary","summary"],"depth":4},{"id":"e18","text":"Correction line: the report\u2019s exact annual mean sea ice extent value remained 6.84 million square kilometers, not 6.8, 6.83, 6.85, 6.87, or any of the rounded, truncated, or neighboring-year values that surfaced during testimony and slide preparation.","tags":["sea-ice-annual-mean","exact-value","correction"],"depth":4}],"relevant_ids":["e1","e14","e18"],"notes":"Expanded extreme crowd-out climate case with many same-family annual values, equivalent expressions, and draft errors surrounding one exact cited report value."}
{"id":"extreme-r5-004","query":"What was the exact timestamp on the stairwell camera when the gray archive box first cleared the landing?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Primary stairwell camera export, admitted only after the facilities integrator explained the sync chain, the frame numbering convention, the corridor-delay issue in one derivative viewer, and the reason later printed stills did not preserve the earliest crossing frame: the gray archive box first fully cleared the landing at exactly 2026-07-22 17:18:43. By the time counsel finished reconstructing the event, the hearing record contained at least a dozen temporally adjacent descriptions that were all either true in some narrower sense or dangerously close to true: a minute-only worksheet entry at 17:18, a supervisor recap at about 5:19 p.m., a still frame at 17:18:47, a radio relay at 17:18:58, a neighboring call at 17:18:52, draft timeline variants at 17:18:41 and 17:18:44, and witness language such as \u201cduring the 5:18 minute\u201d or \u201cafter the door pulse but before the relay.\u201d The court ultimately held that only the raw stairwell export answered the exact question.","tags":["gray-archive-box-landing","exact-time","camera-export"],"depth":1,"timestamp":"2026-07-22T17:18:43"},{"id":"e2","text":"Landing worksheet recorded the movement at 17:18 because the worksheet tracked sequence by minute and was never intended to preserve seconds.","tags":["gray-archive-box-landing","rounded-time","worksheet"],"depth":2,"timestamp":"2026-07-22T17:18:00"},{"id":"e3","text":"Supervisor recap transformed the event into about 5:19 p.m. because the recap grouped several 17:18-minute events into one cleaner managerial narrative.","tags":["gray-archive-box-landing","rounded-time","recap"],"depth":2,"timestamp":"2026-07-22T17:19:00"},{"id":"e4","text":"Hallway still showed the box fully beyond the landing at 17:18:47, later than the first crossing but visually easier for the panel to discuss.","tags":["gray-archive-box-landing","neighbor-timestamp","still"],"depth":2,"timestamp":"2026-07-22T17:18:47"},{"id":"e5","text":"Dispatch radio line relayed the movement at 17:18:58, which multiple witnesses later treated as if it were the event itself because it was the first time they heard it verbalized.","tags":["gray-archive-box-landing","neighbor-timestamp","radio"],"depth":2,"timestamp":"2026-07-22T17:18:58"},{"id":"e6","text":"Witness A remembered the box clearing during the 5:18 minute after the stairwell sensor pulse and before the relay.","tags":["gray-archive-box-landing","coarse-time","witness-a"],"depth":3},{"id":"e7","text":"Witness B insisted on 17:19 sharp because he had internalized the recap email rather than the raw export.","tags":["gray-archive-box-landing","unsupported-time","witness-b"],"depth":3,"timestamp":"2026-07-22T17:19:00"},{"id":"e8","text":"Witness C narrowed the event to after the 17:18:31 door sensor pulse and before 17:18:58, which is helpful but still leaves many plausible candidate seconds.","tags":["gray-archive-box-landing","coarse-time","witness-c"],"depth":3,"timestamp":"2026-07-22T17:18:31"},{"id":"e9","text":"Draft timeline A listed 17:18:41 after the reviewer used the first visible wheel frame instead of the first fully-cleared frame.","tags":["gray-archive-box-landing","near-exact","draft-a"],"depth":3,"timestamp":"2026-07-22T17:18:41"},{"id":"e10","text":"Draft timeline B listed 17:18:44 after the reviewer overcorrected to the next frame.","tags":["gray-archive-box-landing","near-exact","draft-b"],"depth":3,"timestamp":"2026-07-22T17:18:44"},{"id":"e11","text":"Facilities sync note: the camera cluster was aligned to master time at 15:04 and no drift alert appeared thereafter.","tags":["gray-archive-box-landing","clock-sync","maintenance"],"depth":3,"timestamp":"2026-07-22T15:04:00"},{"id":"e12","text":"The index abstract called it the 5:18 landing movement, which was sequence-true but exactness-poor.","tags":["gray-archive-box-landing","rounded-time","index"],"depth":4},{"id":"e13","text":"Correction memo reaffirmed 2026-07-22 17:18:43 as the first fully-cleared landing crossing and explicitly rejected 17:18:41, 17:18:44, 17:18:47, 17:18:52, and 17:19:00 as either draft errors or later adjacent events.","tags":["gray-archive-box-landing","exact-time","correction"],"depth":4,"timestamp":"2026-07-22T17:18:43"},{"id":"e14","text":"Neighboring records call opened at 17:18:52, giving counsel another authoritative-looking but later timestamp to weaponize.","tags":["gray-archive-box-landing","neighbor-timestamp","call"],"depth":4,"timestamp":"2026-07-22T17:18:52"},{"id":"e15","text":"A still-image packet for the jury rounded one frame to 17:18 because the print header suppressed seconds, making the image look less precise than the export from which it came.","tags":["gray-archive-box-landing","rounded-time","still-packet"],"depth":4,"timestamp":"2026-07-22T17:18:00"},{"id":"e16","text":"A side-channel note from the integrator mentioned that one review workstation displayed frame times one frame late until a patch was applied, which partly explains why draft timelines drifted by one or two seconds.","tags":["gray-archive-box-landing","context","integrator-note"],"depth":4},{"id":"e17","text":"Retellings usually smoothed the event into just before 5:19 or sometime in the 5:18 minute because that cadence fit the surrounding oral chronology more naturally than a second-level time.","tags":["gray-archive-box-landing","rounded-memory","retelling"],"depth":4},{"id":"e18","text":"Panel conclusion: nearly every document preserved the correct minute and many preserved a very small interval around the event, but only the raw export and the correction memo preserved the exact crossing second.","tags":["gray-archive-box-landing","agreement-summary","panel"],"depth":4}],"relevant_ids":["e1","e13"],"notes":"Expanded extreme crowd-out timestamp case with many same-minute machine records, draft errors, and rounded derivatives surrounding one exact crossing second."}
{"id":"extreme-r5-005","query":"What was the exact amount on the emergency courier wire adjustment?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Corrected treasury ledger, after a week of circulations that somehow managed to create an entire ecosystem of wrong-but-plausible amounts around a single line item: the emergency courier wire adjustment posted at exactly $19,402.73. The problem is not that the file lacks evidence; it is that it contains too much same-neighborhood evidence. Staff discussed the line as about $19.4k, slides rendered it as roughly $19,403, one draft ledger showed $19,402.70 before the final three cents were reconciled, another showed $19,402.78 after a cent-column transposition, the export view clipped it to $19,402.7, an adjacent fee reversal landed at $19,402.37, another nearby line sat at $19,470.23, and one stale revision briefly read $19,402.72. Nearly every record stayed close enough to the truth to feel semantically irresistible. Only the corrected treasury ledger preserved the exact amount without rounding, clipping, transposition, or adjacency contamination: $19,402.73.","tags":["courier-wire-adjustment","exact-amount","ledger"],"depth":1},{"id":"e2","text":"Operations shorthand called it about $19.4k, which preserved the scale but destroyed the cents.","tags":["courier-wire-adjustment","rounded-amount","ops"],"depth":2},{"id":"e3","text":"Executive slide rendered it as roughly $19,403 to keep the chart visually tidy.","tags":["courier-wire-adjustment","rounded-amount","slides"],"depth":2},{"id":"e4","text":"Draft ledger A showed $19,402.70 before final reconciliation.","tags":["courier-wire-adjustment","near-exact","draft-a"],"depth":2},{"id":"e5","text":"Draft ledger B showed $19,402.78 after a cent transposition.","tags":["courier-wire-adjustment","near-exact","draft-b"],"depth":2},{"id":"e6","text":"Statement export clipped the value to $19,402.7 in a table view that suppressed the last decimal precision beyond one place.","tags":["courier-wire-adjustment","near-exact","export-view"],"depth":3},{"id":"e7","text":"Review email described it as being in the $19,400 range and below the next escalation bucket, which is true but coarser than exact recall requires.","tags":["courier-wire-adjustment","range-description","email"],"depth":3},{"id":"e8","text":"Neighboring fee reversal in the same packet was $19,402.37, creating a same-prefix, same-digit trap just one transposition away from the true value.","tags":["courier-wire-adjustment","neighbor-amount","fee-reversal"],"depth":3},{"id":"e9","text":"Adjacent line item was $19,470.23, which some witnesses remembered only as the other nineteen-four item.","tags":["courier-wire-adjustment","neighbor-amount","adjacent-line"],"depth":3},{"id":"e10","text":"Correction email restated the exact amount as $19,402.73 and rejected all rounded, clipped, and transposed variants.","tags":["courier-wire-adjustment","exact-amount","correction"],"depth":3},{"id":"e11","text":"Hearing shorthand bounced among nineteen-four hundred, nineteen-four-oh-three, and nineteen-four-two-ish, all semantically close but numerically distinct.","tags":["courier-wire-adjustment","rounded-dialogue","hearing"],"depth":4},{"id":"e12","text":"Batch total note listed a cluster subtotal above fifty-eight thousand dollars, adding another authoritative-looking but irrelevant number to the packet.","tags":["courier-wire-adjustment","neighbor-amount","batch-total"],"depth":4},{"id":"e13","text":"Methods note: only the corrected treasury ledger and correction email preserved the exact cents; every other surface rounded, clipped, bucketed, or conflated adjacent lines.","tags":["courier-wire-adjustment","agreement-summary","methods"],"depth":4},{"id":"e14","text":"One printed handout simplified the amount to $19.4K, maximizing readability and minimizing exactness.","tags":["courier-wire-adjustment","rounded-summary","handout"],"depth":4},{"id":"e15","text":"A stale revision briefly showed $19,402.72 after a pre-reconciliation cell was copied forward.","tags":["courier-wire-adjustment","near-exact","revision"],"depth":4},{"id":"e16","text":"Another stale revision showed $19,402.79 after a different analyst copied the wrong cent pair from a neighboring row.","tags":["courier-wire-adjustment","near-exact","revision"],"depth":4},{"id":"e17","text":"Manager oral summary referred to it as the courier wire in the upper nineteen-thousands, which is conceptually right but numerically weak.","tags":["courier-wire-adjustment","range-description","manager"],"depth":4},{"id":"e18","text":"Board recap admitted that everyone in the room agreed on the general scale while failing repeatedly on the exact cents, which is why the corrected ledger had to control the final answer.","tags":["courier-wire-adjustment","agreement-summary","board"],"depth":4}],"relevant_ids":["e1","e10"],"notes":"Expanded extreme crowd-out amount case with many same-prefix, same-digit, and one-cent-off distractors surrounding one exact treasury value."}
{"id":"anti-lexical-001","query":"What was the exact timestamp on the interview room camera when the envelope was placed on the table?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Direct answer memo prepared by counsel: the exact timestamp on the interview room camera when the envelope was placed on the table was described variously as 11:07, 11:08, around 11:07:30, and shortly after eleven-oh-seven, depending on which rounded packet the witness prep team happened to be quoting. The memo repeats the query wording aggressively but preserves only cleaned-up approximations.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Question-and-answer sheet for oral argument restates the query almost verbatim and gives 2026-01-11 11:08:00 as the answer because the drafter copied the minute-rounded still caption rather than the underlying export row.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2},{"id":"e3","text":"Cross-exam prep sheet uses the full query wording and answers it with 11:07:30 because counsel wanted a memorable phrase, not a forensic one.","tags":["interview-room-envelope","explicit-query-echo","nearby"],"depth":2},{"id":"e4","text":"A different prep sheet repeats the same question wording and answers it with 11:07, citing the rounded overlay on a printed still rather than the raw export.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2},{"id":"e5","text":"A radio packet summary again mirrors the query phrasing and then points to 11:07:42, which was actually the later acknowledgment time and not the placement itself.","tags":["interview-room-envelope","explicit-query-echo","neighbor"],"depth":2},{"id":"e6","text":"Hearing binder tab titled EXACT TIMESTAMP ON INTERVIEW ROOM CAMERA repeats the phrasing cleanly but bounces among 11:07, 11:08, and shortly after 11:07 because it was assembled from secondary notes.","tags":["interview-room-envelope","explicit-query-echo","draft"],"depth":3},{"id":"e7","text":"Hallway still packet shows the envelope already on the table at 11:07:24, which is later than the first placement moment but visually authoritative enough to attract retrieval.","tags":["interview-room-envelope","neighbor-timestamp","still"],"depth":3},{"id":"e8","text":"A nearby phone call to chambers began at 11:07:26 and kept entering oral reconstructions because everyone wanted one clean sequence chart.","tags":["interview-room-envelope","neighbor-timestamp","phone"],"depth":3},{"id":"e9","text":"The controlling exhibit was not any of those query-like summaries. It was Exhibit C-17, an NVR export row using camera channel codes, frame indices, and event-class metadata. That row does not read like a natural-language answer at all, but it marks the first qualifying frame for the envelope contacting the table surface at exactly 2026-01-11 11:07:18.","tags":["interview-room-envelope","exact-time","raw-export"],"depth":1},{"id":"e10","text":"The technical correction annotation on the same exhibit restated that 2026-01-11 11:07:18 was the first qualifying frame and distinguished it from rounded captions, later stills, and relay times.","tags":["interview-room-envelope","exact-time","correction"],"depth":3},{"id":"e11","text":"Panel note: nearly every fluent English summary in the packet echoed the query better than the export row itself, but those summaries were also the least precise artifacts in the file.","tags":["interview-room-envelope","agreement-summary","panel"],"depth":4}],"relevant_ids":["e9","e10"],"notes":"Anti-lexical top-5 case: query-echoing distractors dominate while the exact answer is buried in a technical export row and correction note."}
{"id":"anti-lexical-002","query":"What was the exact reimbursement amount on the witness travel receipt?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Answer-prep document: the exact reimbursement amount on the witness travel receipt was summarized as about $63, roughly sixty-three dollars, and in one place as $63.00 because the prep material privileged easy speech over exact bookkeeping.","tags":["witness-travel-reimbursement","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Question list for oral argument repeats the full query and gives $63.20 as the answer because the drafter copied a rounded travel summary rather than the posted reimbursement line.","tags":["witness-travel-reimbursement","explicit-query-echo","near-exact"],"depth":2},{"id":"e3","text":"A hearing binder tab titled EXACT REIMBURSEMENT AMOUNT alternates among $63, $63.2, and sixty-three and change, creating high lexical overlap with the question while failing to preserve the exact cents.","tags":["witness-travel-reimbursement","explicit-query-echo","draft"],"depth":2},{"id":"e4","text":"Expense recap sheet says the exact reimbursement amount on the witness travel receipt sat in the low sixty-threes and belonged to the smaller same-day reimbursements.","tags":["witness-travel-reimbursement","explicit-query-echo","range"],"depth":2},{"id":"e5","text":"Neighboring meal line in the same packet was $63.28, and one reviewer carried that value into a summary sentence because both lines appeared under the same hearing-day heading.","tags":["witness-travel-reimbursement","neighbor-amount","packet"],"depth":3},{"id":"e6","text":"A worksheet revision briefly used $63.21 after a one-cent keying error, adding another clean near-miss.","tags":["witness-travel-reimbursement","near-exact","worksheet"],"depth":3},{"id":"e7","text":"The bank export detail row, which does not mirror the natural-language query at all, lists line code TRV-RMB-4, settlement class witness travel, and posted amount 63.17. In the packet this row is surrounded by metadata and adjacent reimbursement lines, making it look more like bookkeeping debris than like an answer sentence.","tags":["witness-travel-reimbursement","exact-amount","bank-export"],"depth":1},{"id":"e8","text":"Policy memo: sub-$65 witness reimbursements were often rounded to whole dollars in oral review unless exact cents became part of the dispute.","tags":["witness-travel-reimbursement","policy-context"],"depth":3},{"id":"e9","text":"Correction email: the posted witness travel reimbursement amount was $63.17 after the meal line was separated out of the same-day packet.","tags":["witness-travel-reimbursement","exact-amount","correction"],"depth":3},{"id":"e10","text":"One chart rounded the reimbursement to $63.2 and another to $63, creating clean visually aligned distractors with stronger lexical overlap than the raw export row.","tags":["witness-travel-reimbursement","rounded-summary","chart"],"depth":4},{"id":"e11","text":"Clerk note: nearly every clean-English summary in the binder pointed to the right monetary neighborhood while failing on the exact cents; only the coded export row and the correction email preserved the precise amount.","tags":["witness-travel-reimbursement","agreement-summary","clerk"],"depth":4}],"relevant_ids":["e7","e9"],"notes":"Anti-lexical top-5 case: the false entries mirror the question, while the exact amount is buried in export metadata and a later correction email."}
{"id":"anti-lexical-003","query":"What was the exact annual mean methane concentration recorded for 2004?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Public FAQ: the exact annual mean methane concentration recorded for 2004 is described as about 1775 ppb, or roughly 1.775 ppm if someone prefers the converted phrasing. The FAQ repeats the user-like wording but deliberately rounds for accessibility.","tags":["methane-2004","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Testimony prep answer says the exact annual mean methane concentration recorded for 2004 lived in the high 1770s and should be paraphrased loosely unless table fidelity becomes unavoidable.","tags":["methane-2004","explicit-query-echo","coarse"],"depth":2},{"id":"e3","text":"Slide appendix answering the exact annual mean methane concentration question uses 1774.6 after truncation and 1775 after whole-number rounding on adjacent slides.","tags":["methane-2004","explicit-query-echo","near-exact"],"depth":2},{"id":"e4","text":"Another summary says the exact annual mean methane concentration recorded for 2004 was around 1.78 ppm, preserving a converted approximation rather than the report table expression.","tags":["methane-2004","explicit-query-echo","converted"],"depth":2},{"id":"e5","text":"The actual report table row, buried inside appendix columns and unit notes rather than written as a natural-language answer, records the 2004 annual methane mean as 1774.62 ppb.","tags":["methane-2004","exact-annual-mean","report-table"],"depth":1},{"id":"e6","text":"Neighboring report rows show nearby high-1770s values for adjacent years, making the exact 2004 row easy to replace if one remembers only the cluster.","tags":["methane-2004","neighbor-values","table"],"depth":3},{"id":"e7","text":"Correction note: the exact table value for 2004 remained 1774.62 ppb, not the rounded, truncated, or converted figures used elsewhere in the packet.","tags":["methane-2004","exact-annual-mean","correction"],"depth":3},{"id":"e8","text":"Methods annotation: later summaries privilege readability, but the table row is the controlling source when the question asks for the exact annual mean.","tags":["methane-2004","context","methods"],"depth":4},{"id":"e9","text":"Cross-exam summary drifted among 1774.6, 1775, and 1.775 ppm, all close enough to build a very sticky retrieval neighborhood around the exact figure.","tags":["methane-2004","rounded-memory","dialogue"],"depth":4},{"id":"e10","text":"Moderator note: nearly every fluent sentence in the packet was less precise than the raw table row and later correction note.","tags":["methane-2004","agreement-summary","summary"],"depth":4}],"relevant_ids":["e5","e7"],"notes":"Anti-lexical climate case: query-echoing entries are rounded or converted, while the exact value lives in a raw table row and a correction note."}
{"id":"anti-lexical-extreme-001","query":"What was the exact timestamp on the interview room camera when the envelope was placed on the table?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Direct answer memo prepared by counsel: the exact timestamp on the interview room camera when the envelope was placed on the table was described variously as 11:07, 11:08, around 11:07:30, and shortly after eleven-oh-seven, depending on which rounded packet the witness prep team happened to be quoting. The memo repeats the query wording aggressively but preserves only cleaned-up approximations.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Question-and-answer sheet for oral argument restates the query almost verbatim and gives 2026-01-11 11:08:00 as the answer because the drafter copied the minute-rounded still caption rather than the underlying export row.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2,"timestamp":"2026-01-11T11:08:00"},{"id":"e3","text":"Cross-exam prep sheet uses the full query wording and answers it with 11:07:30 because counsel wanted a memorable phrase, not a forensic one.","tags":["interview-room-envelope","explicit-query-echo","nearby"],"depth":2,"timestamp":"2026-01-11T11:07:30"},{"id":"e4","text":"A different prep sheet repeats the same question wording and answers it with 11:07, citing the rounded overlay on a printed still rather than the raw export.","tags":["interview-room-envelope","explicit-query-echo","rounded"],"depth":2,"timestamp":"2026-01-11T11:07:00"},{"id":"e5","text":"A radio packet summary again mirrors the query phrasing and then points to 11:07:42, which was actually the later acknowledgment time and not the placement itself.","tags":["interview-room-envelope","explicit-query-echo","neighbor"],"depth":2,"timestamp":"2026-01-11T11:07:42"},{"id":"e6","text":"Hearing binder tab titled EXACT TIMESTAMP ON INTERVIEW ROOM CAMERA repeats the phrasing cleanly but bounces among 11:07, 11:08, and shortly after 11:07 because it was assembled from secondary notes.","tags":["interview-room-envelope","explicit-query-echo","draft"],"depth":3},{"id":"e7","text":"An oral prep margin note says the exact timestamp on the interview room camera was likely in the 11:07 minute and definitely before the records call, which narrows the window while still ducking exactness.","tags":["interview-room-envelope","explicit-query-echo","coarse"],"depth":3},{"id":"e8","text":"Hallway still packet shows the envelope already on the table at 11:07:24, which is later than the first placement moment but visually authoritative enough to attract retrieval.","tags":["interview-room-envelope","neighbor-timestamp","still"],"depth":3,"timestamp":"2026-01-11T11:07:24"},{"id":"e9","text":"A nearby phone call to chambers began at 11:07:26 and kept entering oral reconstructions because everyone wanted one clean sequence chart.","tags":["interview-room-envelope","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-01-11T11:07:26"},{"id":"e10","text":"The controlling exhibit was not any of those query-like summaries. It was Exhibit C-17, an NVR export row using camera channel codes, frame indices, and event-class metadata. That row does not read like a natural-language answer at all, but it marks the first qualifying frame for the envelope contacting the table surface at exactly 2026-01-11 11:07:18.","tags":["interview-room-envelope","exact-time","raw-export"],"depth":1,"timestamp":"2026-01-11T11:07:18"},{"id":"e11","text":"The technical correction annotation on the same exhibit restated that 2026-01-11 11:07:18 was the first qualifying frame and distinguished it from rounded captions, later stills, and relay times.","tags":["interview-room-envelope","exact-time","correction"],"depth":3,"timestamp":"2026-01-11T11:07:18"},{"id":"e12","text":"Panel note: nearly every fluent English summary in the packet echoed the query better than the export row itself, but those summaries were also the least precise artifacts in the file.","tags":["interview-room-envelope","agreement-summary","panel"],"depth":4},{"id":"e13","text":"A clerk index flattened the event to 11:07 in its subject heading, making the rounded minute the dominant retrieval phrase in later packet searches.","tags":["interview-room-envelope","rounded-time","index"],"depth":4,"timestamp":"2026-01-11T11:07:00"},{"id":"e14","text":"A derivative still caption used 11:07:22 because it indexed the first fully-resting frame rather than the first contact frame.","tags":["interview-room-envelope","near-exact","caption"],"depth":4,"timestamp":"2026-01-11T11:07:22"},{"id":"e15","text":"Another derivative summary preferred 11:07:20 after counting two frames into the contact sequence on a delayed workstation export.","tags":["interview-room-envelope","near-exact","summary"],"depth":4,"timestamp":"2026-01-11T11:07:20"},{"id":"e16","text":"A witness kept tying the event to the records call and saying it happened before 11:07:30 but after the desk bell, preserving rough sequence while reinforcing the wrong neighboring anchors.","tags":["interview-room-envelope","coarse-time","witness"],"depth":4}],"relevant_ids":["e10","e11"],"notes":"Extreme anti-lexical case where the clean query-shaped summaries are mostly wrong and the exact answer lives in an ugly technical export row plus correction annotation."}
{"id":"anti-lexical-extreme-002","query":"What was the exact reimbursement amount on the emergency hotel receipt?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Answer-prep packet under the exact reimbursement question says the emergency hotel receipt was about ninety dollars, high eighties, and effectively ninety for oral purposes. It repeats the user phrasing and keeps the amount semantically close while refusing to preserve the exact cents.","tags":["emergency-hotel-receipt","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Oral argument sheet repeats the question verbatim and gives $89.40 because the drafter copied the unreconciled worksheet value rather than the posted reimbursement line.","tags":["emergency-hotel-receipt","explicit-query-echo","near-exact"],"depth":2},{"id":"e3","text":"A cleaner summary card answers the same question with $89.5 because one chart rounded to the nearest tenth and nobody corrected the speech notes afterward.","tags":["emergency-hotel-receipt","explicit-query-echo","rounded"],"depth":2},{"id":"e4","text":"Another prep note says the exact reimbursement amount on the emergency hotel receipt sat in the $89 range and was one of the low-ninety-ish travel items from the same disruption window.","tags":["emergency-hotel-receipt","explicit-query-echo","range"],"depth":2},{"id":"e5","text":"Expense board note repeats the question language and answers with about $90 because the board packet was organized by thresholds, not by exact cents.","tags":["emergency-hotel-receipt","explicit-query-echo","rounded"],"depth":2},{"id":"e6","text":"Neighboring parking line was $89.83 and a second lodging-related line was $89.74, both close enough that one reviewer later collapsed them into the same sentence about the hotel reimbursement.","tags":["emergency-hotel-receipt","neighbor-amount","packet"],"depth":3},{"id":"e7","text":"Draft worksheet A showed $89.40 before tax reconciliation. Draft worksheet B showed $89.42 after a scan-reading error. Draft worksheet C showed $89.74 after a digit transposition from the parking line. All three remained in circulation long enough to contaminate memory.","tags":["emergency-hotel-receipt","near-exact","worksheets"],"depth":3},{"id":"e8","text":"A manager review form clipped the amount to $89.5 in one chart and to $90 in another because two different display rules were applied to the same underlying reimbursement family.","tags":["emergency-hotel-receipt","rounded-summary","charts"],"depth":3},{"id":"e9","text":"The exact value lives in a coded bank-export row rather than in any of the fluent summaries. Line HTL-RMB-9 in the posting export shows 89.47 after final reconciliation, and the row is easy to skim past because it sits among unrelated travel items, tax lines, and settlement metadata instead of answering the question in plain English.","tags":["emergency-hotel-receipt","exact-amount","bank-export"],"depth":1},{"id":"e10","text":"The correction email later restated the emergency hotel reimbursement as $89.47 after matching the export row against the receipt image and excluding the adjacent parking charge. It is one of the only human-readable places where the exact cents survive intact.","tags":["emergency-hotel-receipt","exact-amount","correction"],"depth":3},{"id":"e11","text":"Policy note says sub-$90 lodging reimbursements were routinely rounded upward in oral review because precision was thought not to matter unless someone challenged the receipt later.","tags":["emergency-hotel-receipt","policy-context"],"depth":3},{"id":"e12","text":"Hearing shorthand alternated among high-eighties, about ninety, eighty-nine and change, and low ninety, all of which sit in the right semantic neighborhood while crowding the exact value.","tags":["emergency-hotel-receipt","rounded-dialogue","hearing"],"depth":4},{"id":"e13","text":"A stale reconciliation cell briefly preserved $89.46 before the final cent adjustment settled, creating a one-cent trap very close to the final number.","tags":["emergency-hotel-receipt","near-exact","stale-cell"],"depth":4},{"id":"e14","text":"The batch summary total and neighboring travel lines introduced several additional ninety-ish amounts that were unrelated to the exact hotel reimbursement but lived on the same pages and in the same review discussion.","tags":["emergency-hotel-receipt","neighbor-amount","batching"],"depth":4},{"id":"e15","text":"Clerk note: almost every clean-English answer in the packet was less precise than the coded export row and later correction note, which is why the exact cents are so vulnerable to top-5 crowd-out despite not being absent from the record.","tags":["emergency-hotel-receipt","agreement-summary","clerk"],"depth":4},{"id":"e16","text":"One revised packet even carried $89.48 for a day after a rounding-to-nearest-cent mistake in a copy step, adding yet another one-cent-off distractor with cleaner prose than the true source row.","tags":["emergency-hotel-receipt","near-exact","revision"],"depth":4}],"relevant_ids":["e9","e10"],"notes":"Extreme anti-lexical amount case where fluent query-mirroring summaries are all rounded or wrong, while the exact cents live in a coded export line and correction email."}
{"id":"anti-lexical-extreme-003","query":"What was the exact annual mean CO2 concentration recorded for 1988?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Public-facing FAQ answers the exact annual mean CO2 concentration question with about 352 ppm because it values readability and headline memory over table fidelity. It repeats the user wording closely while smearing the precise annual value into a rounded whole number.","tags":["co2-1988","explicit-query-echo","rounded"],"depth":2},{"id":"e2","text":"Witness prep answer says the exact annual mean CO2 concentration recorded for 1988 was in the low 352 range and can be described as around 351.6 or roughly 352 depending on audience tolerance for decimals.","tags":["co2-1988","explicit-query-echo","near-exact"],"depth":2},{"id":"e3","text":"Slide appendix answers the exact annual mean CO2 question with 351.6 in one place, 352 in another, and 0.3516 thousand ppm in a third graphic because several derivative formats were prepared from one underlying table.","tags":["co2-1988","explicit-query-echo","converted"],"depth":2},{"id":"e4","text":"Another summary says the exact annual mean CO2 concentration recorded for 1988 belongs in the mid-351 to low-352 band and matters mainly as a hinge year in the late-1980s public climate debate.","tags":["co2-1988","explicit-query-echo","range"],"depth":2},{"id":"e5","text":"A polished hearing handout repeats the exact query phrasing and then gives 351.6 because the designer rounded to one decimal place for visual consistency with the rest of the chart wall.","tags":["co2-1988","explicit-query-echo","rounded"],"depth":2},{"id":"e6","text":"The actual source row does not look like an answer sentence at all. In the annual table, row 1988 appears between neighboring yearly entries and records 351.57 ppm as the mean concentration. The row is surrounded by year labels, units, neighboring values, and formatting noise, making it semantically weaker than the fluent summaries even though it is the controlling record.","tags":["co2-1988","exact-annual-mean","report-table"],"depth":1},{"id":"e7","text":"Correction annotation: the exact annual mean for 1988 remained 351.57 ppm, not 351.6, 352, about 352, or any of the handout-style simplifications that spread through testimony and outreach material.","tags":["co2-1988","exact-annual-mean","correction"],"depth":3},{"id":"e8","text":"Neighboring annual values in the late-1980s table are close enough that one can remember the entire cluster correctly while still failing to identify the exact 1988 entry. Several reviewers later quoted adjacent-year numbers as if they were the 1988 annual mean.","tags":["co2-1988","neighbor-values","table"],"depth":3},{"id":"e9","text":"One draft handout briefly preserved 351.75 after the wrong row was copied into the witness packet. It looks scientific, precise, and close, which is exactly what makes it dangerous for retrieval.","tags":["co2-1988","near-exact","draft"],"depth":3},{"id":"e10","text":"Cross-exam notes drifted among the annual mean, the summer testimony period, nearby years, and broader trend claims, leaving a cloud of semantically aligned but numerically non-identical candidates around the exact value.","tags":["co2-1988","context","dialogue"],"depth":4},{"id":"e11","text":"Methods note says the raw annual table controls over later narrative summaries, but the methods page itself does not restate the number as a clean natural-language answer.","tags":["co2-1988","context","methods"],"depth":4},{"id":"e12","text":"Panel summary: nearly every smooth sentence in the packet degrades the precision a little, while the exact value survives only in the ugly table row and its later correction line.","tags":["co2-1988","agreement-summary","summary"],"depth":4}],"relevant_ids":["e6","e7"],"notes":"Extreme anti-lexical climate case where the query-aligned summaries are all rounded, and the exact annual mean lives only in a table row and correction note."}
{"id":"mega-top5-001","query":"What was the exact timestamp on the service corridor camera when the green evidence case first crossed the painted threshold?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Witness-prep answer sheet: the exact timestamp on the service corridor camera when the green evidence case first crossed the painted threshold was described as 16:42, about 16:43, during the 16:42 minute, and shortly before the later radio acknowledgment. This sheet was optimized for oral fluency, not for preserving the first-frame second.","tags":["green-evidence-case","query-echo","rounded"],"depth":2},{"id":"e2","text":"Counsel outline repeats the query almost verbatim and gives 2026-05-06 16:43:00 as the answer because the rounded summary chart had already collapsed the event into the next minute.","tags":["green-evidence-case","query-echo","rounded"],"depth":2,"timestamp":"2026-05-06T16:43:00"},{"id":"e3","text":"Cross-exam prep note repeats the exact timestamp question and answers it with 16:42:51, which was actually the radio relay time later used as a mnemonic anchor in the hearing packet.","tags":["green-evidence-case","query-echo","neighbor"],"depth":2,"timestamp":"2026-05-06T16:42:51"},{"id":"e4","text":"Another prep version uses 16:42:37 because that was the first still image where the case was clearly beyond the line, not the first threshold-crossing frame. The wording still mirrors the user query more naturally than the raw export does.","tags":["green-evidence-case","query-echo","nearby"],"depth":2,"timestamp":"2026-05-06T16:42:37"},{"id":"e5","text":"Operations recap says the exact timestamp on the service corridor camera landed in the 16:42 range and before the callout, preserving sequence while refusing exactness.","tags":["green-evidence-case","query-echo","coarse"],"depth":2},{"id":"e6","text":"Clean summary card says the event happened at 16:42 because that was the minute bucket shown on the shift sheet and looked neat enough for executive review.","tags":["green-evidence-case","query-echo","rounded"],"depth":2,"timestamp":"2026-05-06T16:42:00"},{"id":"e7","text":"Still frame A shows the case already across the threshold at 16:42:36.","tags":["green-evidence-case","neighbor-timestamp","still-a"],"depth":3,"timestamp":"2026-05-06T16:42:36"},{"id":"e8","text":"Still frame B shows the case fully clear at 16:42:37, which later became the preferred timestamp in a simplified chronology because it was visually crisp and easy to quote.","tags":["green-evidence-case","neighbor-timestamp","still-b"],"depth":3,"timestamp":"2026-05-06T16:42:37"},{"id":"e9","text":"Dispatch radio acknowledgment was logged at 16:42:51 and repeatedly overfit into oral summaries as if it were the crossing itself.","tags":["green-evidence-case","neighbor-timestamp","radio"],"depth":3,"timestamp":"2026-05-06T16:42:51"},{"id":"e10","text":"A door sensor pulse hit at 16:42:18, leading one reviewer to build a false timeline where the threshold crossing was treated as a few seconds later than it really was.","tags":["green-evidence-case","neighbor-timestamp","sensor"],"depth":3,"timestamp":"2026-05-06T16:42:18"},{"id":"e11","text":"Draft timeline A listed 16:42:31 after the reviewer clicked the second motion frame. Draft timeline B listed 16:42:34 after another reviewer overcorrected from the minute-rounded overlay. Draft timeline C used 16:42:35 after a still export changed frame numbering. None of these values survived final review, but all of them remained in packet history and are dangerously close to the truth.","tags":["green-evidence-case","near-exact","draft-history"],"depth":3},{"id":"e12","text":"A nearby call to records opened at 16:42:44 and then got woven into later sequence summaries because it fit naturally between the rounded shift sheet and the radio acknowledgment.","tags":["green-evidence-case","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-05-06T16:42:44"},{"id":"e13","text":"The controlling source was not any of those fluent summaries. It was export row SC-441-B in the NVR appendix, which used terse machine labels instead of natural-language explanation. That row recorded the first threshold-crossing frame for the green evidence case at exactly 2026-05-06 16:42:29.","tags":["green-evidence-case","exact-time","raw-export"],"depth":1,"timestamp":"2026-05-06T16:42:29"},{"id":"e14","text":"The correction annotation on SC-441-B restated 2026-05-06 16:42:29 and distinguished it from all later stills, relay times, rounded logs, and frame-selection mistakes.","tags":["green-evidence-case","exact-time","correction"],"depth":3,"timestamp":"2026-05-06T16:42:29"},{"id":"e15","text":"The maintenance sync note says the corridor camera, door sensor, and dispatch clock were aligned at 13:02 that afternoon. This matters for evidence confidence, but it does not make the raw export row linguistically easier to retrieve.","tags":["green-evidence-case","context","clock-sync"],"depth":4,"timestamp":"2026-05-06T13:02:00"},{"id":"e16","text":"Panel recap admits that almost every human-readable summary in the packet keeps the right minute and the wrong second, while the coded export and the correction line preserve the exact crossing time.","tags":["green-evidence-case","agreement-summary","panel"],"depth":4}],"relevant_ids":["e13","e14"],"notes":"Pathological anti-lexical timestamp case: 12+ query-shaped or near-query distractors compete against one ugly raw source row and its correction note."}
{"id":"mega-top5-002","query":"What was the exact reimbursement amount on the overnight relocation receipt?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Oral prep packet says the exact reimbursement amount on the overnight relocation receipt was about $214, just over two hundred, or somewhere in the low two-hundreds depending on which threshold discussion one used. Every sentence is shaped like a direct answer and every sentence sacrifices precision.","tags":["overnight-relocation-receipt","query-echo","rounded"],"depth":2},{"id":"e2","text":"Question list repeats the full query and answers it with $214.00 because that was the rounded reimbursement bucket in the finance summary.","tags":["overnight-relocation-receipt","query-echo","rounded"],"depth":2},{"id":"e3","text":"Another answer sheet gives $214.30 because the drafter copied a pre-tax worksheet rather than the posted line.","tags":["overnight-relocation-receipt","query-echo","near-exact"],"depth":2},{"id":"e4","text":"A summary card says the exact reimbursement amount on the overnight relocation receipt was $214.38, which came from a transposed scan correction and not from the final settled reimbursement.","tags":["overnight-relocation-receipt","query-echo","near-exact"],"depth":2},{"id":"e5","text":"Board memo answers the same question with two fourteen and change because nobody expected the cents to matter until the dispute hardened.","tags":["overnight-relocation-receipt","query-echo","coarse"],"depth":2},{"id":"e6","text":"Neighboring meal receipt in the same packet was $214.83. Another relocation-adjacent line was $214.37. Both later appeared in side summaries that looked cleaner than the real source row.","tags":["overnight-relocation-receipt","neighbor-amount","packet"],"depth":3},{"id":"e7","text":"Worksheet draft A showed $214.30, draft B showed $214.38, draft C showed $214.73 after a digit transposition, and a stale chart showed $214.4 because the final cent was clipped in display.","tags":["overnight-relocation-receipt","near-exact","worksheet-history"],"depth":3},{"id":"e8","text":"A manager review form clipped the amount to $214.3 in one chart and to $214 in another because two different display rules were applied to the same underlying reimbursement family.","tags":["overnight-relocation-receipt","rounded-summary","charts"],"depth":3},{"id":"e9","text":"The exact value does not live in any of the fluent summaries. It sits in settlement export line RLC-22, where the posted amount is 214.27 after tax and after exclusion of the adjacent meal item. The row is easy to skim past because it sits among unrelated travel items, tax lines, and settlement metadata instead of answering the question in plain English.","tags":["overnight-relocation-receipt","exact-amount","raw-export"],"depth":1},{"id":"e10","text":"The correction email later restated the overnight relocation reimbursement as $214.27 after matching the export row against the receipt image and excluding the adjacent meal charge. It is one of the only human-readable places where the exact cents survive intact.","tags":["overnight-relocation-receipt","exact-amount","correction"],"depth":3},{"id":"e11","text":"Policy note says low-two-hundred-dollar relocation reimbursements were often bucketed into rounded language because review focused on threshold class, not exact cents.","tags":["overnight-relocation-receipt","policy-context"],"depth":3},{"id":"e12","text":"Hearing shorthand moved between about two-fourteen, low two-hundreds, two fourteen and change, and the settled relocation line, all of which strengthen the same semantic neighborhood around the exact amount.","tags":["overnight-relocation-receipt","rounded-dialogue","hearing"],"depth":4},{"id":"e13","text":"A stale reconciliation cell briefly preserved $214.26 before the final cent adjustment landed, giving the file yet another one-cent-off trap close to the truth.","tags":["overnight-relocation-receipt","near-exact","stale-cell"],"depth":4},{"id":"e14","text":"The batch summary total and neighboring travel lines introduced several additional two-hundred-fourteen-ish amounts that were unrelated to the exact hotel reimbursement but lived on the same pages and in the same review discussion.","tags":["overnight-relocation-receipt","neighbor-amount","batching"],"depth":4},{"id":"e15","text":"Clerk note: almost every clean-English answer in the packet was less precise than the coded export row and later correction note, which is why the exact cents are so vulnerable to top-5 crowd-out despite not being absent from the record.","tags":["overnight-relocation-receipt","agreement-summary","clerk"],"depth":4},{"id":"e16","text":"One revised packet even carried $214.28 for a day after a rounding-to-nearest-cent mistake in a copy step, adding yet another one-cent-off distractor with cleaner prose than the true source row.","tags":["overnight-relocation-receipt","near-exact","revision"],"depth":4}],"relevant_ids":["e9","e10"],"notes":"Pathological anti-lexical amount case: most human-readable answers are wrong or rounded, while the exact cents survive only in a coded export row and a correction email."}
{"id":"mega-top5-003","query":"What was the exact annual mean nitrous oxide concentration recorded for 1997?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Public FAQ answers the exact annual mean nitrous oxide concentration question with about 311 ppb because it values readability and pedagogical simplicity over strict table fidelity. The sentence looks like a direct answer but is already rounded.","tags":["n2o-1997","query-echo","rounded"],"depth":2},{"id":"e2","text":"Testimony prep answer says the exact annual mean nitrous oxide concentration recorded for 1997 was in the low 311 range and can be described as 311.0-ish when speaking to a general audience.","tags":["n2o-1997","query-echo","coarse"],"depth":2},{"id":"e3","text":"Slide appendix answering the exact annual mean question uses 311.0 after one-decimal rounding and 311 after whole-number rounding on adjacent graphics.","tags":["n2o-1997","query-echo","rounded"],"depth":2},{"id":"e4","text":"Another summary says the exact annual mean nitrous oxide concentration recorded for 1997 sat around 0.311 ppm, preserving a converted approximation rather than the table expression.","tags":["n2o-1997","query-echo","converted"],"depth":2},{"id":"e5","text":"A polished hearing card repeats the entire query and answers it with 310.97 after a truncation from a derivative chart export.","tags":["n2o-1997","query-echo","near-exact"],"depth":2},{"id":"e6","text":"The actual table row does not look like an answer sentence at all. It sits inside a year-by-year atmospheric trace gas appendix and records 1997 at 310.98 ppb, surrounded by neighboring annual values, unit headings, and statistical notation.","tags":["n2o-1997","exact-annual-mean","report-table"],"depth":1},{"id":"e7","text":"Neighboring annual rows cluster tightly enough that several witnesses later remembered the right neighborhood while still assigning the exact value to the wrong year.","tags":["n2o-1997","neighbor-values","table"],"depth":3},{"id":"e8","text":"A draft handout briefly printed 311.08 after a row-offset bug, creating a scientific-looking but wrong near-exact alternative.","tags":["n2o-1997","near-exact","draft"],"depth":3},{"id":"e9","text":"Correction note preserves 310.98 ppb as the exact 1997 annual mean and explicitly distinguishes it from one-decimal, whole-number, converted, and row-offset variants.","tags":["n2o-1997","exact-annual-mean","correction"],"depth":3},{"id":"e10","text":"Methods note explains that derivative outreach materials round for readability while the raw annual appendix preserves the exact cited values.","tags":["n2o-1997","context","methods"],"depth":4},{"id":"e11","text":"Audience recap drifted among 311, 311.0, 310.97, and about 0.311 ppm, proving how easy it is for semantically close but numerically distinct values to dominate memory.","tags":["n2o-1997","rounded-memory","audience"],"depth":4},{"id":"e12","text":"Cross-examination drifted into neighboring years, decadal trend framing, and ppm-vs-ppb conversion language, building a very dense retrieval neighborhood around the exact annual value.","tags":["n2o-1997","context","dialogue"],"depth":4},{"id":"e13","text":"Moderator summary: almost every fluent sentence in the packet degrades precision a little, whereas the exact answer survives only in the appendix table row and the formal correction note.","tags":["n2o-1997","agreement-summary","summary"],"depth":4}],"relevant_ids":["e6","e9"],"notes":"Pathological anti-lexical climate case: clean query-shaped summaries are rounded or converted, while the exact annual mean is stranded in a raw appendix row and correction note."}
{"id":"anti-lexical-extreme-004","query":"What was the exact timestamp on the west annex camera when the silver evidence pouch first touched the intake shelf?","adversary_type":"time_swap","entries":[{"id":"e1","text":"Witness prep packet repeats the exact timestamp question and answers it as 07:54, about 7:55, or shortly before 7:55 depending on which rounded packet page is being used. The phrasing mirrors the query but preserves only minute-level convenience.","tags":["silver-pouch-shelf","query-echo","rounded"],"depth":2},{"id":"e2","text":"Oral argument card restates the question and gives 2026-03-08 07:55:00 because the still-caption packet had already collapsed the event into the next minute.","tags":["silver-pouch-shelf","query-echo","rounded"],"depth":2,"timestamp":"2026-03-08T07:55:00"},{"id":"e3","text":"Cross-exam draft uses the full query and answers it with 07:54:49, which was really the later acknowledgment relay, not the first shelf-contact frame.","tags":["silver-pouch-shelf","query-echo","neighbor"],"depth":2,"timestamp":"2026-03-08T07:54:49"},{"id":"e4","text":"Another prep answer uses 07:54:33 because that was the first clean still image where the pouch looked settled, not the first touching instant.","tags":["silver-pouch-shelf","query-echo","nearby"],"depth":2,"timestamp":"2026-03-08T07:54:33"},{"id":"e5","text":"A cleaner summary card says the exact timestamp was in the 07:54 minute before the relay call, preserving sequence but not the second.","tags":["silver-pouch-shelf","query-echo","coarse"],"depth":2},{"id":"e6","text":"Question list for the clerk answered the same question with 07:54 because the incident index only preserved minute buckets.","tags":["silver-pouch-shelf","query-echo","rounded"],"depth":2,"timestamp":"2026-03-08T07:54:00"},{"id":"e7","text":"Still frame A shows the pouch already touching the shelf at 07:54:31.","tags":["silver-pouch-shelf","neighbor-timestamp","still-a"],"depth":3,"timestamp":"2026-03-08T07:54:31"},{"id":"e8","text":"Still frame B shows the pouch fully resting at 07:54:33, which later became the preferred simplified answer in several packet summaries.","tags":["silver-pouch-shelf","neighbor-timestamp","still-b"],"depth":3,"timestamp":"2026-03-08T07:54:33"},{"id":"e9","text":"A nearby records call began at 07:54:39 and was repeatedly folded into one blended chronology with the shelf-contact event.","tags":["silver-pouch-shelf","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-03-08T07:54:39"},{"id":"e10","text":"The relay acknowledgment landed at 07:54:49, which some witnesses later treated as if it were the event itself.","tags":["silver-pouch-shelf","neighbor-timestamp","relay"],"depth":3,"timestamp":"2026-03-08T07:54:49"},{"id":"e11","text":"Draft timeline A listed 07:54:27 after a reviewer clicked the second frame in the motion burst.","tags":["silver-pouch-shelf","near-exact","draft-a"],"depth":3,"timestamp":"2026-03-08T07:54:27"},{"id":"e12","text":"Draft timeline B listed 07:54:30 after overcorrecting from a rounded overlay.","tags":["silver-pouch-shelf","near-exact","draft-b"],"depth":3,"timestamp":"2026-03-08T07:54:30"},{"id":"e13","text":"The controlling row is buried in export table WA-117-C, event class object_contact, frame index 8033. It records the first qualifying frame for the silver evidence pouch touching the intake shelf at exactly 2026-03-08 07:54:28. It looks like metadata, not like a fluent answer sentence.","tags":["silver-pouch-shelf","exact-time","raw-export"],"depth":1,"timestamp":"2026-03-08T07:54:28"},{"id":"e14","text":"The technical correction annotation on WA-117-C restates 2026-03-08 07:54:28 and rejects the still, relay, and rounded minute variants.","tags":["silver-pouch-shelf","exact-time","correction"],"depth":3,"timestamp":"2026-03-08T07:54:28"},{"id":"e15","text":"Camera maintenance note: the annex camera cluster was synchronized at 05:41 that morning.","tags":["silver-pouch-shelf","context","clock-sync"],"depth":4,"timestamp":"2026-03-08T05:41:00"},{"id":"e16","text":"Panel summary: the entries that look most like the user query are all rounded or neighboring times, while the exact answer survives only in the coded export row and the technical correction note.","tags":["silver-pouch-shelf","agreement-summary","panel"],"depth":4},{"id":"e17","text":"Witnesses agreed the event occurred before 07:55 and after the 07:54 sensor pulse, but none of them preserved the exact second in ordinary language.","tags":["silver-pouch-shelf","coarse-time","witnesses"],"depth":4},{"id":"e18","text":"One audit sheet later normalized the event to 07:54:30 for charting consistency, creating a polished but wrong near-exact number that spread further than the raw source row.","tags":["silver-pouch-shelf","near-exact","audit"],"depth":4,"timestamp":"2026-03-08T07:54:30"}],"relevant_ids":["e13","e14"],"notes":"Pathological anti-lexical timestamp case with many query-shaped rounded answers and one exact coded export row."}
{"id":"anti-lexical-extreme-005","query":"What was the exact reimbursement amount on the emergency relocation hotel receipt?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Prepared answer packet says the exact reimbursement amount on the emergency relocation hotel receipt was about $146, mid one-forties, or one-forty-six even for discussion purposes. It mirrors the query wording while rounding away the cents.","tags":["relocation-hotel-receipt","query-echo","rounded"],"depth":2},{"id":"e2","text":"Question list repeats the exact reimbursement question and answers it with $146.00 because that was the rounded value in the management summary.","tags":["relocation-hotel-receipt","query-echo","rounded"],"depth":2},{"id":"e3","text":"Another prep answer gives $146.28 because the drafter copied the pre-tax worksheet instead of the posted reimbursement line.","tags":["relocation-hotel-receipt","query-echo","near-exact"],"depth":2},{"id":"e4","text":"A hearing card says the exact reimbursement amount was $146.82, which actually belonged to the adjacent parking line but sounded authoritative enough to survive in oral prep.","tags":["relocation-hotel-receipt","query-echo","neighbor"],"depth":2},{"id":"e5","text":"Board note says the amount sat in the low one-forties and below the overnight escalation threshold, which is true but coarser than exact cents.","tags":["relocation-hotel-receipt","query-echo","range"],"depth":2},{"id":"e6","text":"Worksheet history includes $146.28, $146.23, and $146.32, all of which are near-exact alternatives created by stale tax and scan readings.","tags":["relocation-hotel-receipt","near-exact","worksheet-history"],"depth":3},{"id":"e7","text":"Neighboring parking line was $146.82 and meal line was $146.37, creating several answer-like numbers in the same packet neighborhood.","tags":["relocation-hotel-receipt","neighbor-amount","packet"],"depth":3},{"id":"e8","text":"Chart summary rounded the value to $146.3 on one page and $146 on another because different display rules were used.","tags":["relocation-hotel-receipt","rounded-summary","chart"],"depth":3},{"id":"e9","text":"The exact amount lives in settlement export line HTL-RMB-17, where the posted value is 146.27 after exclusion of parking and meal lines. The row is surrounded by codes and unrelated travel metadata rather than natural-language explanation.","tags":["relocation-hotel-receipt","exact-amount","raw-export"],"depth":1},{"id":"e10","text":"Correction email later restated the hotel reimbursement as $146.27 after reconciling the export row against the receipt image and removing adjacent charges.","tags":["relocation-hotel-receipt","exact-amount","correction"],"depth":3},{"id":"e11","text":"Policy note: hotel reimbursements in the one-forty range were often rounded to whole dollars in oral review.","tags":["relocation-hotel-receipt","policy-context"],"depth":3},{"id":"e12","text":"Hearing shorthand moved among one-forty-six, one-forty-six and change, and low one-forties, reinforcing the same semantic neighborhood around the exact amount.","tags":["relocation-hotel-receipt","rounded-dialogue","hearing"],"depth":4},{"id":"e13","text":"A stale reconciliation cell preserved $146.26 before the final cent adjustment landed, adding a one-cent trap.","tags":["relocation-hotel-receipt","near-exact","stale-cell"],"depth":4},{"id":"e14","text":"Another stale revision preserved $146.29 after a copied rounding bug.","tags":["relocation-hotel-receipt","near-exact","revision"],"depth":4},{"id":"e15","text":"A batch total above four hundred dollars also appeared in the same packet and sometimes contaminated memory because it shared the same hearing-day heading.","tags":["relocation-hotel-receipt","neighbor-amount","batching"],"depth":4},{"id":"e16","text":"Clerk note: nearly every clean-English answer in the packet pointed to the correct monetary neighborhood while preserving the wrong cents. Only the coded export row and correction email retained the exact amount.","tags":["relocation-hotel-receipt","agreement-summary","clerk"],"depth":4},{"id":"e17","text":"A manager summary line used \u201cabout $146\u201d because the escalation decision did not care about the final cents.","tags":["relocation-hotel-receipt","rounded-summary","manager"],"depth":4},{"id":"e18","text":"The smoothest answers in the packet were also the least exact, which is exactly why this line item became such a retrieval trap.","tags":["relocation-hotel-receipt","context","summary"],"depth":4}],"relevant_ids":["e9","e10"],"notes":"Pathological anti-lexical amount case with many query-mirroring rounded answers and one exact export-row amount."}
{"id":"anti-lexical-extreme-006","query":"What was the exact annual mean CO2 concentration recorded for 1991?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Public FAQ answers the exact annual mean CO2 concentration question with about 356 ppm because readability wins over table fidelity in the public-facing copy.","tags":["co2-1991","query-echo","rounded"],"depth":2},{"id":"e2","text":"Witness prep answer says the exact annual mean CO2 concentration recorded for 1991 lived in the mid-350s and can be described as roughly 355.7 or about 356 for general audiences.","tags":["co2-1991","query-echo","near-exact"],"depth":2},{"id":"e3","text":"Slide appendix answers the exact annual mean question with 355.7 in one place, 356 in another, and about 0.356 thousand ppm in a converted chart.","tags":["co2-1991","query-echo","converted"],"depth":2},{"id":"e4","text":"Another summary says the exact annual mean CO2 concentration recorded for 1991 belongs in the upper-355 to low-356 band and matters mostly as part of the early-1990s climb.","tags":["co2-1991","query-echo","range"],"depth":2},{"id":"e5","text":"A polished hearing card repeats the query phrasing and then gives 355.72 after one-decimal-to-two-decimal reconstruction from a derivative chart.","tags":["co2-1991","query-echo","near-exact"],"depth":2},{"id":"e6","text":"The actual source row appears in the annual table between neighboring yearly entries and records 1991 at 355.68 ppm, surrounded by unit labels, adjacent values, and appendix formatting.","tags":["co2-1991","exact-annual-mean","report-table"],"depth":1},{"id":"e7","text":"Neighboring annual rows are close enough that several speakers later remembered the right cluster while assigning the exact value to the wrong year.","tags":["co2-1991","neighbor-values","table"],"depth":3},{"id":"e8","text":"A draft handout briefly printed 355.86 after a transposed copy error, making a scientific-looking but wrong near-exact alternative.","tags":["co2-1991","near-exact","draft"],"depth":3},{"id":"e9","text":"Correction note preserves 355.68 ppm as the exact 1991 annual mean and distinguishes it from rounded, converted, and row-offset variants.","tags":["co2-1991","exact-annual-mean","correction"],"depth":3},{"id":"e10","text":"Methods note explains that derivative outreach materials round for readability while the raw annual appendix preserves the exact cited values.","tags":["co2-1991","context","methods"],"depth":4},{"id":"e11","text":"Audience recap drifted among 355.7, 356, 355.72, and mid-350s, proving how quickly semantically aligned values crowd the true one.","tags":["co2-1991","rounded-memory","audience"],"depth":4},{"id":"e12","text":"Cross-examination drifted into neighboring years, volcanic-aerosol context, and trend framing, creating a thick retrieval neighborhood around the exact annual value.","tags":["co2-1991","context","dialogue"],"depth":4},{"id":"e13","text":"Moderator summary: almost every fluent answer sentence in the packet degraded precision slightly, whereas the exact annual mean survived only in the raw table row and formal correction note.","tags":["co2-1991","agreement-summary","summary"],"depth":4}],"relevant_ids":["e6","e9"],"notes":"Pathological anti-lexical climate case where clean query-shaped answers are rounded while the exact annual mean lives in the raw appendix row and correction note."}
{"id":"anti-lexical-mega-007","query":"What was the exact timestamp on the archive intake camera when the sealed tube first touched the stainless bench?","adversary_type":"time_swap","entries":[{"id":"e1","text":"The witness-prep packet answers the exact timestamp on the archive intake camera question with about 13:26, maybe 13:27, and in one place 'right in the middle of the 1:26 minute.' Every version sounds fluent and directly responsive because it was written for people, not for exactness.","tags":["sealed-tube-bench","query-echo","rounded"],"depth":2},{"id":"e2","text":"A polished Q-and-A sheet repeats the full query and gives 2026-08-19 13:27:00 because the printed still caption had already collapsed the event into the next minute.","tags":["sealed-tube-bench","query-echo","rounded"],"depth":2,"timestamp":"2026-08-19T13:27:00"},{"id":"e3","text":"A litigation outline answers the same exact-timestamp question with 13:26:52, which is actually the later intercom acknowledgment and not the first contact event.","tags":["sealed-tube-bench","query-echo","neighbor"],"depth":2,"timestamp":"2026-08-19T13:26:52"},{"id":"e4","text":"A summary card gives 13:26:41 because that was the first still where the tube looked settled on the bench rather than the first contact frame.","tags":["sealed-tube-bench","query-echo","nearby"],"depth":2,"timestamp":"2026-08-19T13:26:41"},{"id":"e5","text":"Another prep note says the exact timestamp on the archive intake camera was in the 13:26 range and before the acknowledgment relay, preserving sequence while discarding the second.","tags":["sealed-tube-bench","query-echo","coarse"],"depth":2},{"id":"e6","text":"The case index answers the same question with 13:26 because the intake chronology stores events at minute granularity unless a dispute later forces deeper review.","tags":["sealed-tube-bench","query-echo","rounded"],"depth":2,"timestamp":"2026-08-19T13:26:00"},{"id":"e7","text":"Still frame A shows the tube already touching the bench at 13:26:39. It is machine-derived, visually clean, and wrong for first touch by one second-level neighborhood step.","tags":["sealed-tube-bench","neighbor-timestamp","still-a"],"depth":3,"timestamp":"2026-08-19T13:26:39"},{"id":"e8","text":"Still frame B shows the tube fully settled at 13:26:41, which later became the preferred simplification in oral summaries because it looked calmer and easier to narrate.","tags":["sealed-tube-bench","neighbor-timestamp","still-b"],"depth":3,"timestamp":"2026-08-19T13:26:41"},{"id":"e9","text":"An intercom acknowledgment was logged at 13:26:52 and repeatedly treated as if it were the relevant machine time because it appears in a cleaner textual format than the raw video export row.","tags":["sealed-tube-bench","neighbor-timestamp","intercom"],"depth":3,"timestamp":"2026-08-19T13:26:52"},{"id":"e10","text":"A nearby records call opened at 13:26:47, and several hearing summaries fused the call and the bench contact into one rough sequence event.","tags":["sealed-tube-bench","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-08-19T13:26:47"},{"id":"e11","text":"Draft chronology A listed 13:26:34 after the reviewer clicked the second frame in a motion burst. Draft chronology B listed 13:26:36 after overcorrecting from the minute overlay. Draft chronology C listed 13:26:38 after a delayed workstation rendered one frame late. All three stayed in packet history.","tags":["sealed-tube-bench","near-exact","draft-history"],"depth":3},{"id":"e12","text":"Witness A remembered the contact as happening after the scanner beep at 13:26:11 and before the intercom, which is true but still leaves a wide second-level interval crowded with plausible candidates.","tags":["sealed-tube-bench","coarse-time","witness-a"],"depth":4,"timestamp":"2026-08-19T13:26:11"},{"id":"e13","text":"Witness B insisted on 13:27 because the printed still packet used a minute-rounded caption and because he prepared from the clerk summary instead of the export appendix.","tags":["sealed-tube-bench","unsupported-time","witness-b"],"depth":4,"timestamp":"2026-08-19T13:27:00"},{"id":"e14","text":"Witness C said only that the event happened in the middle of the 13:26 minute and definitely before the acknowledgement relay, which semantically supports many wrong neighbors as well as the right answer.","tags":["sealed-tube-bench","coarse-time","witness-c"],"depth":4},{"id":"e15","text":"The exact answer does not live in any of the fluent summaries. It lives in export row AI-3-1181, event class object_contact, which marks the first qualifying bench-contact frame at exactly 2026-08-19 13:26:35. Because the row is wrapped in channel IDs, frame counters, and parser metadata, it is semantically weaker than the polished false summaries despite being the controlling source.","tags":["sealed-tube-bench","exact-time","raw-export"],"depth":1,"timestamp":"2026-08-19T13:26:35"},{"id":"e16","text":"The attached technical correction line restates 2026-08-19 13:26:35 and explicitly rejects the rounded minute answers, the later still captions, the intercom time, and the draft chronology variants.","tags":["sealed-tube-bench","exact-time","correction"],"depth":3,"timestamp":"2026-08-19T13:26:35"},{"id":"e17","text":"Maintenance note says the archive intake camera cluster was synchronized at 10:02 that morning. That improves trust in the export but does nothing to make the controlling row read like a natural answer sentence.","tags":["sealed-tube-bench","context","clock-sync"],"depth":4,"timestamp":"2026-08-19T10:02:00"},{"id":"e18","text":"Panel summary admits the packet is full of nicer-looking, query-shaped answers that all orbit the right minute while the exact second survives only in the ugly export row and its correction note.","tags":["sealed-tube-bench","agreement-summary","panel"],"depth":4}],"relevant_ids":["e15","e16"],"notes":"Mega anti-lexical timestamp case where polished query-shaped answers are all rounded or neighboring events and the exact answer lives only in a coded export row and correction note."}
{"id":"anti-lexical-mega-008","query":"What was the exact reimbursement amount on the witness lodging relocation claim?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Prepared answer packet says the exact reimbursement amount on the witness lodging relocation claim was about $173, a little over one-seventy, and one hundred seventy-three even in any spoken summary. The packet mirrors the query wording but rounds away the cents entirely.","tags":["witness-lodging-relocation","query-echo","rounded"],"depth":2},{"id":"e2","text":"Question list repeats the reimbursement question verbatim and gives $173.00 because the management summary rounded all sub-threshold lodging lines to whole dollars.","tags":["witness-lodging-relocation","query-echo","rounded"],"depth":2},{"id":"e3","text":"Another prep answer gives $173.42 because the drafter copied the unreconciled pre-tax worksheet instead of the settled reimbursement line.","tags":["witness-lodging-relocation","query-echo","near-exact"],"depth":2},{"id":"e4","text":"A cleaner hearing card says the exact reimbursement amount was $173.82, which actually belonged to the adjacent parking and toll cluster but looked authoritative enough to survive in oral prep.","tags":["witness-lodging-relocation","query-echo","neighbor"],"depth":2},{"id":"e5","text":"Board memo answers the same question with low one-seventies and under the overnight escalation threshold, which is directionally true while remaining maximally imprecise.","tags":["witness-lodging-relocation","query-echo","range"],"depth":2},{"id":"e6","text":"Worksheet history includes $173.42, $173.47, and $173.24, each produced by some mix of stale tax treatment, OCR confusion, or line-item bleed from adjacent travel charges.","tags":["witness-lodging-relocation","near-exact","worksheet-history"],"depth":3},{"id":"e7","text":"Neighboring parking line was $173.82, meal line was $173.27, and one separate lodging-adjacent correction line was $173.20, all of which sat on the same hearing-day packet pages and polluted recall.","tags":["witness-lodging-relocation","neighbor-amount","packet"],"depth":3},{"id":"e8","text":"A chart summary rounded the amount to $173.4 in one view and to $173 in another because two display rules were applied inconsistently across the board packet.","tags":["witness-lodging-relocation","rounded-summary","chart"],"depth":3},{"id":"e9","text":"The exact amount is not preserved by any of those fluent summaries. It sits in settlement export line WLR-14, where the posted reimbursement value is 173.29 after exclusion of the adjacent meal and parking components. The row is semantically ugly: code prefix, account class, posted cents, and no nice answer wording.","tags":["witness-lodging-relocation","exact-amount","raw-export"],"depth":1},{"id":"e10","text":"The correction email later restated the witness lodging relocation claim as $173.29 after matching the export line against the receipt image and subtracting the adjacent non-lodging charges.","tags":["witness-lodging-relocation","exact-amount","correction"],"depth":3},{"id":"e11","text":"Policy memo says low-one-hundred-seventy-dollar lodging reimbursements were routinely rounded for oral discussion because the board cared first about category, then about cents only if challenged.","tags":["witness-lodging-relocation","policy-context"],"depth":3},{"id":"e12","text":"Hearing shorthand moved among about one-seventy-three, one-seventy-three and change, low one-seventies, and the same hotel cluster, all of which keep retrieval trapped in the correct monetary neighborhood without surfacing the exact cents.","tags":["witness-lodging-relocation","rounded-dialogue","hearing"],"depth":4},{"id":"e13","text":"A stale reconciliation cell preserved $173.28 before the final cent adjustment landed. Another preserved $173.30 after a chart export rounded half-up. Both remained in the packet long enough to look answer-like.","tags":["witness-lodging-relocation","near-exact","stale-cells"],"depth":4},{"id":"e14","text":"The batch summary total and adjacent travel subtotals introduced several other one-seventy-three-ish values into the same review packet, including one that differed by only two cents from the real amount.","tags":["witness-lodging-relocation","neighbor-amount","batching"],"depth":4},{"id":"e15","text":"Clerk note says nearly every human-readable answer in the file was cleaner than the truth and therefore more likely to be remembered, while the exact cents survived only in the coded export row and the correction message.","tags":["witness-lodging-relocation","agreement-summary","clerk"],"depth":4},{"id":"e16","text":"One revised packet even carried $173.31 for a day after a copy-forward bug from a neighboring hotel line, adding another near-exact but cleaner-looking distractor.","tags":["witness-lodging-relocation","near-exact","revision"],"depth":4},{"id":"e17","text":"Manager summary said the line was one of the hundred-seventy-three-dollar witness lodging claims, which grouped it correctly by family while sacrificing the exact cents entirely.","tags":["witness-lodging-relocation","rounded-summary","manager"],"depth":4},{"id":"e18","text":"The cleanest language in the packet is also the least exact. The exact value survives only where the prose is the least helpful to a surface-level retriever.","tags":["witness-lodging-relocation","context","summary"],"depth":4}],"relevant_ids":["e9","e10"],"notes":"Mega anti-lexical amount case where polished query-shaped answers are all rounded or neighboring amounts and the exact value lives only in a coded export row and correction note."}
{"id":"source-confusion-001","query":"What was the exact average temperature recorded for the city station on 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Witness prep notes kept answering the city-station question with low-eighties language, often quoting 82F, 82.6F, or 81.7F interchangeably because the hearing packet bounced between Central Park, LaGuardia, and a citywide summary slide that treated both as representative urban stations.","tags":["city-station-temperature","query-echo","rounded"],"depth":2},{"id":"e2","text":"A polished summary paragraph answered the exact average temperature question with 82.6F because the author subconsciously defaulted to the airport chart when writing about city stations in general.","tags":["city-station-temperature","query-echo","neighbor-station"],"depth":2},{"id":"e3","text":"Another prep answer gave 81.7F because a different drafter treated Central Park as the default city station whenever no airport was named explicitly.","tags":["city-station-temperature","query-echo","neighbor-station"],"depth":2},{"id":"e4","text":"The packet contains a citywide slide saying the city was around 82 on average that day, which is broad enough to sound correct regardless of which station the speaker later claims they meant.","tags":["city-station-temperature","query-echo","citywide"],"depth":2},{"id":"e5","text":"One hearing chart labels LaGuardia as station LGA-1 and Central Park as CPK-1, but later narrative text often says only city station, weather station, or local station without preserving the code.","tags":["city-station-temperature","context","station-codes"],"depth":3},{"id":"e6","text":"A station-specific row for LGA-1 records 82.6F on 2023-07-14. It is exact, authoritative, and wrong if the intended city station is not the airport code row.","tags":["city-station-temperature","exact-but-wrong-scope","table"],"depth":3,"timestamp":"2023-07-14"},{"id":"e7","text":"A station-specific row for CPK-1 records 81.7F on 2023-07-14. It is exact, authoritative, and wrong if the intended city station is interpreted as the airport-facing operational station.","tags":["city-station-temperature","exact-but-wrong-scope","table"],"depth":3,"timestamp":"2023-07-14"},{"id":"e8","text":"The authoritative glossary buried in the appendix says that, unless otherwise qualified, city station refers to the municipal climate baseline station and maps to CPK-1 rather than LaGuardia's operational airport feed. The sentence is easy to skip because it looks like administrative metadata rather than an answer.","tags":["city-station-temperature","scope-resolution","appendix"],"depth":1},{"id":"e9","text":"A correction note later spells out the implication: because city station resolves to CPK-1 in the glossary, the exact average temperature for 2023-07-14 is 81.7F despite the tempting 82.6F airport value nearby.","tags":["city-station-temperature","exact-value","correction"],"depth":3},{"id":"e10","text":"Multiple witnesses agreed that the city was in the low eighties that day, but their partial agreement concealed a station-scope split more than a real factual disagreement about weather.","tags":["city-station-temperature","agreement-summary","witnesses"],"depth":4},{"id":"e11","text":"The monthly anomaly appendix adds more noise by showing both stations inside the same warm cluster, encouraging retrieval to treat them as interchangeable.","tags":["city-station-temperature","context","monthly"],"depth":4},{"id":"e12","text":"The exact answer does not come from the nicest-looking narrative sentence or even from the cleanest exact row. It comes from an ugly scope-definition note plus the CPK-1 row it silently governs.","tags":["city-station-temperature","context","scope"],"depth":4}],"relevant_ids":["e8","e9"],"notes":"Source-confusion case where multiple exact rows are plausible and only a buried scope-definition note determines which exact row is correct."}
{"id":"source-confusion-002","query":"What was the exact amount of the travel reimbursement in the official ledger?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Every smooth answer in the hearing packet says the official ledger amount was about $214, $214.27, or one of the nearby one-seventy and two-hundred blocks depending on whether the speaker was reading the hotel packet, the reimbursement export, or the travel summary. All of them sound ledger-like.","tags":["official-ledger-travel-reimbursement","query-echo","rounded"],"depth":2},{"id":"e2","text":"One summary says the official ledger amount was $214.27 because it quietly substituted the settlement export line for the board ledger line. Another says $214.00 because it rounded the board ledger into a threshold bucket.","tags":["official-ledger-travel-reimbursement","query-echo","scope-confusion"],"depth":2},{"id":"e3","text":"A witness insisted the official ledger showed the same number as the export because she thought all posted lines eventually converged to one canonical amount. The packet shows that this assumption was false.","tags":["official-ledger-travel-reimbursement","witness-error","scope-confusion"],"depth":2},{"id":"e4","text":"The settlement export line RLC-22 records 214.27 after tax and after exclusion of adjacent meal charges. It is exact, real, and not the board ledger amount the query asks for.","tags":["official-ledger-travel-reimbursement","exact-but-wrong-source","export"],"depth":3},{"id":"e5","text":"The board ledger row BL-17 records 214.31 after a later consolidation pass. It appears only in a dense appendix table that uses abbreviated column names and no explanatory prose.","tags":["official-ledger-travel-reimbursement","exact-source-row","ledger"],"depth":1},{"id":"e6","text":"A correction note says the official ledger amount was 214.31 even though the settlement export remained 214.27, because the board ledger absorbed a later four-cent adjustment that the export row never reflected.","tags":["official-ledger-travel-reimbursement","exact-source-row","correction"],"depth":3},{"id":"e7","text":"One chart rounded both source families to $214.3, making the export and ledger appear harmonized when they were not actually identical.","tags":["official-ledger-travel-reimbursement","near-exact","chart"],"depth":3},{"id":"e8","text":"Another worksheet preserved 214.30 from an intermediate state before the board ledger finalized at 214.31. That one-cent gap is tiny enough to be deadly for retrieval.","tags":["official-ledger-travel-reimbursement","near-exact","worksheet"],"depth":3},{"id":"e9","text":"A packet sidebar talks about the official amount without specifying whether official means posted export, corrected ledger, or board-facing ledger, which is why witnesses kept using exact-looking but source-misaligned numbers.","tags":["official-ledger-travel-reimbursement","context","source-ambiguity"],"depth":4},{"id":"e10","text":"The packet therefore contains at least three plausible exact numbers: the export amount, the intermediate worksheet amount, and the final board ledger amount. Only one matches the query's source constraint.","tags":["official-ledger-travel-reimbursement","agreement-summary","panel"],"depth":4},{"id":"e11","text":"Most fluent answers in the file point at the wrong source family. The exact right answer is hidden in a boring ledger appendix row and the correction note that names the four-cent divergence explicitly.","tags":["official-ledger-travel-reimbursement","context","scope"],"depth":4}],"relevant_ids":["e5","e6"],"notes":"Source-confusion case where multiple exact source records exist and only a subtle source constraint in the query makes one of them correct."}
{"id":"source-confusion-003","query":"What was the exact concentration listed in the official monitoring ledger for the harbor nitrate reading?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"Every polished answer in the packet says the official monitoring ledger listed the harbor nitrate reading as about 4.8 mg/L, 4.82 mg/L, or roughly five milligrams per liter, depending on whether the speaker was looking at the field worksheet, the summary chart, or the shoreline briefing. All of those answers sound ledger-like because they borrow the query vocabulary directly.","tags":["harbor-nitrate-reading","query-echo","rounded"],"depth":2},{"id":"e2","text":"One hearing note says the official monitoring ledger value was 4.82 mg/L because the author quietly substituted the field worksheet figure for the ledger figure after assuming the two had converged.","tags":["harbor-nitrate-reading","query-echo","scope-confusion"],"depth":2},{"id":"e3","text":"Another answer card gives 4.80 mg/L because the briefing slide rounded everything in the high-four range to one decimal or a neat hundredth.","tags":["harbor-nitrate-reading","query-echo","rounded"],"depth":2},{"id":"e4","text":"A witness testified that the ledger and the worksheet both showed the same concentration and that any tiny mismatch must have been a transcription artifact. The packet later shows that assumption was false.","tags":["harbor-nitrate-reading","witness-error","scope-confusion"],"depth":2},{"id":"e5","text":"Field worksheet row FW-19 records 4.82 mg/L as the immediate shoreline sample concentration before the lab reconciliation pass. It is exact, genuine, and wrong if the query is asking for the official monitoring ledger value rather than the field sheet.","tags":["harbor-nitrate-reading","exact-but-wrong-source","worksheet"],"depth":3},{"id":"e6","text":"Shoreline briefing chart uses 4.80 mg/L after rounding the worksheet family for public readability. It is even cleaner than the worksheet and therefore even more likely to be retrieved, while still not being the ledger number.","tags":["harbor-nitrate-reading","rounded-summary","briefing"],"depth":3},{"id":"e7","text":"The official monitoring ledger row ML-07 records 4.79 mg/L after the confirmation pass and after two suspect duplicates were excluded. The row appears only in a dense appendix table with unit keys and station abbreviations, not in any fluent narrative sentence.","tags":["harbor-nitrate-reading","exact-source-row","ledger"],"depth":1},{"id":"e8","text":"A correction note explains the gap directly: the official monitoring ledger value remained 4.79 mg/L even though the field worksheet retained 4.82 mg/L, because the ledger reflects the later confirmed concentration after duplicate exclusion.","tags":["harbor-nitrate-reading","exact-source-row","correction"],"depth":3},{"id":"e9","text":"One chart introduced 4.81 mg/L as a smoothed midpoint for a hearing visual, creating a very plausible near-exact trap that belongs to neither source row.","tags":["harbor-nitrate-reading","near-exact","chart"],"depth":3},{"id":"e10","text":"The packet therefore contains at least four plausible nitrate values in the same tiny neighborhood: 4.79, 4.80, 4.81, and 4.82 mg/L, all of which sound defensible unless the retriever respects the source constraint in the query.","tags":["harbor-nitrate-reading","agreement-summary","panel"],"depth":4},{"id":"e11","text":"Most fluent summaries in the file point to the wrong source family. The right answer is hidden in the appendix ledger row and the correction note that distinguishes it from the field worksheet.","tags":["harbor-nitrate-reading","context","scope"],"depth":4}],"relevant_ids":["e7","e8"],"notes":"Source-confusion mega-case with multiple exact and near-exact nitrate values where only the official monitoring ledger source makes one of them correct."}
{"id":"source-confusion-004","query":"What was the exact average temperature recorded in the official airport climatology table for 2023-07-14?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The hearing packet repeatedly answers the official airport climatology question with 82.6F, around 83F, and low-eighties at the airport, because most of the narrative material is sourced from airport-facing summary charts rather than the specific climatology appendix the query asks about.","tags":["airport-climatology-temperature","query-echo","rounded"],"depth":2},{"id":"e2","text":"One polished answer card says the official airport climatology table recorded 82.6F, borrowing the operational station value because that number had already become the memorable airport figure in oral discussion.","tags":["airport-climatology-temperature","query-echo","scope-confusion"],"depth":2},{"id":"e3","text":"Another answer says the official airport climatology table put the date at about 83F because the communications summary rounded the airport side of the station pair to the nearest whole degree.","tags":["airport-climatology-temperature","query-echo","rounded"],"depth":2},{"id":"e4","text":"A witness insisted the airport operations sheet and the climatology table were functionally the same source because both described the airport on the same date, even though the packet later distinguishes them.","tags":["airport-climatology-temperature","witness-error","scope-confusion"],"depth":2},{"id":"e5","text":"The operational station table for the airport records 82.6F on 2023-07-14. It is exact, real, and wrong if the query is asking for the official climatology table rather than the operations feed.","tags":["airport-climatology-temperature","exact-but-wrong-source","ops-table"],"depth":3,"timestamp":"2023-07-14"},{"id":"e6","text":"A rounded airport slide uses 83F after clipping the operations feed for readability. That number then shows up everywhere because it is cleaner than both exact station rows.","tags":["airport-climatology-temperature","rounded-summary","slide"],"depth":3},{"id":"e7","text":"The official climatology appendix row AC-14 records 82.4F for 2023-07-14 after the station family is normalized into the climatology table. The row is easy to miss because it sits in a coded appendix beneath operational charts that look more user-friendly.","tags":["airport-climatology-temperature","exact-source-row","climatology"],"depth":1,"timestamp":"2023-07-14"},{"id":"e8","text":"A correction note later spells out the divergence: the operational table remained 82.6F, but the official airport climatology table cited 82.4F after the climatology normalization pass.","tags":["airport-climatology-temperature","exact-source-row","correction"],"depth":3},{"id":"e9","text":"Another chart inserted 82.5F as a smoothed display midpoint between airport-facing sources, producing a near-exact but unsupported compromise value that sounds especially plausible.","tags":["airport-climatology-temperature","near-exact","chart"],"depth":3},{"id":"e10","text":"The packet thus contains a tight cluster of 82.4F, 82.5F, 82.6F, and 83F, all talking about the same airport date and all feeling retrieval-relevant, but only one of them belongs to the official climatology table named in the query.","tags":["airport-climatology-temperature","agreement-summary","panel"],"depth":4},{"id":"e11","text":"Most natural-language summaries collapse operational and climatology sources into one airport story. The exact answer survives only where the source family is named explicitly in the appendix and correction note.","tags":["airport-climatology-temperature","context","scope"],"depth":4}],"relevant_ids":["e7","e8"],"notes":"Source-confusion mega-case where several exact airport temperatures coexist and only the official climatology source makes one of them correct."}
{"id":"aggressive-r5-001","query":"What was the exact timestamp on the south evidence room camera when the orange file box first crossed the interior tape line?","adversary_type":"time_swap","entries":[{"id":"e1","text":"The hearing packet answers the exact timestamp question in polished prose over and over: 12:14, 12:15, 12:14:41, and shortly before 12:15 all appear in summaries, prep sheets, oral outlines, and witness coaching notes. Those clean answers agree with one another more than they agree with the raw machine export, which is why they are so dangerous.","tags":["orange-file-box","query-echo","rounded"],"depth":2},{"id":"e2","text":"A witness prep card repeats the full query and answers it with 2026-06-03 12:15:00 because the rounded still caption was easier to memorize than the export table.","tags":["orange-file-box","query-echo","rounded"],"depth":2,"timestamp":"2026-06-03T12:15:00"},{"id":"e3","text":"A second prep card uses 12:14:41 because that was the first frame where the box was clearly beyond the tape, not the first crossing frame itself.","tags":["orange-file-box","query-echo","nearby"],"depth":2,"timestamp":"2026-06-03T12:14:41"},{"id":"e4","text":"The oral chronology says the event happened in the 12:14 minute and before the 12:14:56 radio relay. It sounds careful and is semantically close to the truth while still not being exact.","tags":["orange-file-box","query-echo","coarse"],"depth":2},{"id":"e5","text":"A radio relay at 12:14:56 is cited in three different packet sections and quietly treated as the machine-confirmed crossing time even though it is not the camera event itself.","tags":["orange-file-box","neighbor-timestamp","radio"],"depth":3,"timestamp":"2026-06-03T12:14:56"},{"id":"e6","text":"The still packet uses 12:14:41 and 12:14:43 on two separate pages because one page anchors to first visible over-line movement and the other to first full-clear frame. Both times look authoritative and both are wrong for the exact query.","tags":["orange-file-box","near-exact","still-packet"],"depth":3},{"id":"e7","text":"The incident worksheet records 12:14 because all hallway events were minute-bucketed at intake. Later witnesses repeatedly used that worksheet to bolster their memory.","tags":["orange-file-box","rounded-time","worksheet"],"depth":3,"timestamp":"2026-06-03T12:14:00"},{"id":"e8","text":"A nearby desk call opened at 12:14:48 and another chart fused it with the box crossing because both sat inside the same event cluster.","tags":["orange-file-box","neighbor-timestamp","phone"],"depth":3,"timestamp":"2026-06-03T12:14:48"},{"id":"e9","text":"Draft timeline A listed 12:14:37 after a reviewer clicked one frame late. Draft timeline B listed 12:14:39 after another reviewer used a delayed workstation export. Draft timeline C listed 12:14:40 after averaging the two.","tags":["orange-file-box","near-exact","draft-history"],"depth":3},{"id":"e10","text":"The actual controlling source is export row SER-2-9901, which does not read like a human answer. It lists event class boundary_cross, frame offset metadata, and records the first qualifying crossing at exactly 2026-06-03 12:14:38.","tags":["orange-file-box","exact-time","raw-export"],"depth":1,"timestamp":"2026-06-03T12:14:38"},{"id":"e11","text":"A technical correction note restates 2026-06-03 12:14:38 and explains why 12:14, 12:14:41, 12:14:43, 12:14:48, and 12:15 all kept propagating in cleaner summaries.","tags":["orange-file-box","exact-time","correction"],"depth":3,"timestamp":"2026-06-03T12:14:38"},{"id":"e12","text":"The facilities sync log says the camera cluster was aligned at 09:05 that morning, strengthening confidence in the export without making it any easier to retrieve by surface language.","tags":["orange-file-box","context","clock-sync"],"depth":4,"timestamp":"2026-06-03T09:05:00"},{"id":"e13","text":"Panel summary acknowledges that almost all fluent and repeated answers point to the wrong exact-looking times, while the exact source row is linguistically ugly and appears only once outside the correction note.","tags":["orange-file-box","agreement-summary","panel"],"depth":4},{"id":"e14","text":"Witnesses independently reinforced the false 12:15 and 12:14:41 narratives because those times matched the materials they had been shown, creating corroborated-but-wrong exactness around the true event.","tags":["orange-file-box","corroborated-falsehood","witnesses"],"depth":4}],"relevant_ids":["e10","e11"],"notes":"Aggressive top-5 attack where the false timestamps are cleaner, more repeated, and more mutually reinforcing than the true raw export time."}
{"id":"aggressive-r5-002","query":"What was the exact amount on the overnight evacuation lodging reimbursement?","adversary_type":"near_duplicate_paraphrase","entries":[{"id":"e1","text":"The hearing packet keeps giving the same family of answers in fluent language: about $128, $128.40, $128.00, and one-twenty-eight and change. These values recur in prep notes, board summaries, and oral argument sheets often enough that they begin to feel like corroboration.","tags":["evacuation-lodging-reimbursement","query-echo","rounded"],"depth":2},{"id":"e2","text":"One prep sheet repeats the exact reimbursement question and answers it with $128.40 because that was the pre-tax worksheet figure.","tags":["evacuation-lodging-reimbursement","query-echo","near-exact"],"depth":2},{"id":"e3","text":"A second sheet answers it with $128.00 because management rounded low-hundreds reimbursements for threshold discussion.","tags":["evacuation-lodging-reimbursement","query-echo","rounded"],"depth":2},{"id":"e4","text":"A cleaner hearing card uses $128.48, which actually belongs to the adjacent parking reimbursement but circulated as if it were the hotel amount because the packet merged the two on one page.","tags":["evacuation-lodging-reimbursement","query-echo","neighbor"],"depth":2},{"id":"e5","text":"Another polished summary says the reimbursement was in the upper one-twenties and below the hotel escalation threshold, strengthening the same neighborhood without naming the right cents.","tags":["evacuation-lodging-reimbursement","query-echo","range"],"depth":2},{"id":"e6","text":"Worksheet history contains $128.40, $128.42, $128.24, and $128.48 from stale tax treatment, OCR confusion, and neighboring line bleed. Several of these numbers were repeated in later packet material as if they had equal pedigree.","tags":["evacuation-lodging-reimbursement","near-exact","worksheet-history"],"depth":3},{"id":"e7","text":"A nearby meal line was $128.24, the parking line was $128.48, and one subtotalized chart rendered the lodging family at $128.4, giving the packet a swarm of answer-shaped neighbors.","tags":["evacuation-lodging-reimbursement","neighbor-amount","packet"],"depth":3},{"id":"e8","text":"The actual amount is preserved in coded settlement export line EVL-31, where the posted reimbursement value is 128.29 after exclusion of parking and meal components. The row appears in dense export metadata with no friendly restatement.","tags":["evacuation-lodging-reimbursement","exact-amount","raw-export"],"depth":1},{"id":"e9","text":"A correction email later restated the overnight evacuation lodging reimbursement as $128.29 after reconciling the export row against the receipt image and separating adjacent charges.","tags":["evacuation-lodging-reimbursement","exact-amount","correction"],"depth":3},{"id":"e10","text":"Policy note says low-one-hundred-twenty-dollar hotel reimbursements were routinely rounded in oral discussion.","tags":["evacuation-lodging-reimbursement","policy-context"],"depth":3},{"id":"e11","text":"One stale reconciliation cell preserved $128.28. Another preserved $128.30. A third preserved $128.32 after a copy-forward bug. All remained in circulation long enough to look like plausible exacts.","tags":["evacuation-lodging-reimbursement","near-exact","stale-cells"],"depth":4},{"id":"e12","text":"The batch summary total and adjacent hotel cluster created still more one-hundred-twenty-eight-ish values that are semantically aligned but source-wrong.","tags":["evacuation-lodging-reimbursement","neighbor-amount","batching"],"depth":4},{"id":"e13","text":"Clerk note says almost every natural-language answer in the file is cleaner than the truth and therefore easier to retrieve, while the exact cents survive only in the coded export and the correction line.","tags":["evacuation-lodging-reimbursement","agreement-summary","clerk"],"depth":4},{"id":"e14","text":"Three different false exact values are corroborated by secondary documents, while the true value is corroborated only by the export row and the correction email.","tags":["evacuation-lodging-reimbursement","corroborated-falsehood","summary"],"depth":4}],"relevant_ids":["e8","e9"],"notes":"Aggressive top-5 attack where the false monetary values are cleaner and more mutually reinforced than the true export-backed amount."}