README: replace arXiv-ID-forthcoming placeholders with canonical arXiv:2605.02063

Croissant: correct tier_1_5_2d_slcd record count 3,062 -> 1,563 (matches deployed tier_1_5_2d_slcd/ JSONL shards)

Add tier_1_5_2d_slcd shard: tier_1_5_2d_slcd_calibration.jsonl (Tier-1.5 2D SLCD sensitivity sweep, recovered from instance a25ca083c0a1 done-tarball)

Add tier_1_5_2d_slcd shard: tier_1_5_2d_slcd_MASAC.jsonl (Tier-1.5 2D SLCD sensitivity sweep, recovered from instance a25ca083c0a1 done-tarball)

Add tier_1_5_2d_slcd shard: tier_1_5_2d_slcd_MADDPG.jsonl (Tier-1.5 2D SLCD sensitivity sweep, recovered from instance a25ca083c0a1 done-tarball)

Add tier_1_5_2d_slcd shard: tier_1_5_2d_slcd_ISAC.jsonl (Tier-1.5 2D SLCD sensitivity sweep, recovered from instance a25ca083c0a1 done-tarball)

Add tier_1_5_2d_slcd shard: tier_1_5_2d_slcd_IPPO.jsonl (Tier-1.5 2D SLCD sensitivity sweep, recovered from instance a25ca083c0a1 done-tarball)

Croissant: restore sha256 on FileObject (schema-required; removing it broke schema validation)

Croissant: route record-generation test to git-clone path with .git URL (validator records test is non-blocking per NeurIPS CFP)

Croissant: fix FileObject contentUrl to use git-clonable .git suffix (was the /tree/main view URL which is not git-clonable)

Croissant: add NeurIPS-required RAI fields (dataSocialImpact, hasSyntheticData) and FileObject containedIn references

Add retry-cycle deduplication explanation (27,613 raw -> 17,930 unique records; latest-timestamp policy)

Update Croissant: align author name to Eric S. K. Yu (matches OpenReview profile and AI-CP-1 author block)

README: update record count to 17,930 (post-dedup)

Croissant: update record count to 17,930 (post-dedup)

Deduplicate retry runs: keep latest-timestamp per (algo, env, seed, tr_mode); 27613 -> 17930 records

Filter progress-log records from training_runs_0559.jsonl

Filter progress-log records from training_runs_0546.jsonl

Filter progress-log records from training_runs_0545.jsonl

Filter progress-log records from training_runs_0544.jsonl

Filter progress-log records from training_runs_0538.jsonl

Filter progress-log records from training_runs_0536.jsonl

Filter progress-log records from training_runs_0503.jsonl

Filter progress-log records from training_runs_0502.jsonl

Filter progress-log records from training_runs_0501.jsonl

Filter progress-log records from training_runs_0495.jsonl

Filter progress-log records from training_runs_0494.jsonl

Filter progress-log records from training_runs_0492.jsonl

Croissant: revert parent FileObject (caused Download error); restore FileSet-only structure

Croissant: add parent FileObject (repo) + containedIn refs from FileSets to enable mlcroissant streaming

Fix Croissant training_runs_records field extracts: column -> jsonPath, nested metrics -> $.metrics.X

Refresh Croissant manifest with canonical hardware/wording (RTX 5090 only; replace campaign with study)

Correct shard count: 949 (not 950) JSONL shards in training_runs

Patch README with correct subdirectory names (training_runs, behavioral_audit) and JSONL-shard read instructions

Update README for canonical naming, Zenodo DOI, and v1.0.0 release

Replace em-dash in LICENSE.md heading with colon to comply with authorial style guide

V1.1.0 release — top-level metadata (README, LICENSE, Croissant)

V1.1.0 release — training_runs batch 12/12 (70 files, 303.1 MB)

V1.1.0 release — training_runs batch 11/12 (80 files, 360.8 MB)

V1.1.0 release — training_runs batch 10/12 (80 files, 384.3 MB)

V1.1.0 release — training_runs batch 9/12 (80 files, 357.5 MB)

V1.1.0 release — training_runs batch 8/12 (80 files, 346.1 MB)

V1.1.0 release — training_runs batch 7/12 (80 files, 408.5 MB)

V1.1.0 release — training_runs batch 6/12 (80 files, 349.7 MB)

V1.1.0 release — training_runs batch 5/12 (80 files, 378.6 MB)

V1.1.0 release — training_runs batch 4/12 (80 files, 392.9 MB)

V1.1.0 release — training_runs batch 3/12 (80 files, 402.6 MB)

V1.1.0 release — training_runs batch 2/12 (80 files, 387.1 MB)

V1.1.0 release — training_runs batch 1/12 (80 files, 386.7 MB)

V1.1.0 release — behavioral_audit batch 1/1 (3 files, 7.7 MB)