Datasets:

anon-ops
/

ops-lite-review-sample

license: apache-2.0
task_categories:
  - other
  - graph-ml
language:
  - en
tags:
  - root-cause-analysis
  - microservices
  - observability
  - chaos-engineering
  - benchmark
size_categories:
  - n<1K

ops-lite-review-sample

This repository is a reviewer-facing sample of the full ops-lite benchmark.

Full dataset: https://huggingface.co/datasets/anon-ops/ops-lite
Full dataset release used for this sample: the same public ops-lite release associated with this submission
Sample dataset URL: https://huggingface.co/datasets/anon-ops/ops-lite-review-sample

What is included

10 complete RCA cases under cases/
a subset manifest.jsonl containing metadata for those same 10 cases
the same per-case file structure as the full dataset (injection.json, env.json, result.json, causal_graph.json, abnormal/normal parquet tables, etc.)

How the sample was created

This sample is a manually curated reviewer subset, not a random sample. It was created from the 500-case full benchmark using the following rules:

Include both naming patterns that appear in the full release: human-readable case names and opaque batch-* identifiers.
Preserve coverage across all three systems (ts, hs, otel-demo).
Preserve coverage across several common fault families in the benchmark (Network*, Pod*, *Stress, HTTP*, and a small number of hybrid cases).
Preserve some path-depth variation by including cases with longest_path from 2 to 5.
Keep the sample small enough for quick manual inspection by reviewers.

Because the sample is intentionally curated for inspection quality, it should not be used for quantitative evaluation or for estimating corpus-level statistics. Use the full dataset for benchmarking.

Sample composition

case	naming style	system	family	primary kind	longest_path	root services
`ts1-ts-route-plan-service-stress-pvnmb5`	named	ts (Train-Ticket)	`*Stress`	`JVMMemoryStress`	4	`ts-route-plan-service`
`ts0-ts-preserve-service-response-abort-7gp2mq`	named	ts (Train-Ticket)	`HTTP*`	`HTTPResponseAbort`	2	`ts-preserve-service`
`ts1-ts-station-service-delay-hwcd55`	named	ts (Train-Ticket)	`Network*`	`NetworkDelay`	5	`ts-station-service`
`batch-01KQJB58WRCZZKW2Z0SYFET4DR`	batch-*	ts (Train-Ticket)	`hybrid_clean`	`hybrid`	3	`ts-food-service, ts-train-food-service`
`hs1-geo-cpu-exhaustion-q8nln6`	named	hs (Hotel Reservation / DeathStarBench)	`*Stress`	`CPUStress`	3	`geo`
`hs1-search-delay-bdcn5c`	named	hs (Hotel Reservation / DeathStarBench)	`Network*`	`NetworkDelay`	3	`search`
`batch-01KQHDBB5Y8K69Z13G7VSWG9FS`	batch-*	hs (Hotel Reservation / DeathStarBench)	`Pod*`	`PodFailure`	2	`geo, profile`
`otel-demo3-shipping-delay-m6fhpx`	named	otel-demo (OpenTelemetry Demo)	`Network*`	`NetworkDelay`	5	`shipping`
`otel-demo1-product-catalog-pod-kill-dk4528`	named	otel-demo (OpenTelemetry Demo)	`Pod*`	`PodKill`	4	`product-catalog`
`batch-01KQJS94VKJXFZZ6K1RK19XQCB`	batch-*	otel-demo (OpenTelemetry Demo)	`hybrid_clean`	`hybrid`	4	`payment, shipping`

Coverage summary

Systems: 4 ts, 3 hs, 3 otel-demo
Naming styles: 6 named cases, 4 batch-* cases
Fault families represented: *Stress, HTTP*, Network*, Pod*, hybrid_clean

Notes for reviewers

The sample is hosted in the same way as the full dataset: a public Hugging Face dataset repository.
All files in this sample are copied directly from the full release; only the case count is reduced.
If you need aggregate statistics or broader fault coverage, inspect the full dataset instead of this sample.