case.schema.yaml — case descriptor (language, sinks, deterministic build/test, environment, optional inline ground truth summary).
entrypoints.schema.yaml — declared entrypoints grouped by type (http, cli, scheduled, etc.).
truth.schema.json — ground-truth labels + evidence per sink (reachable/unreachable/unknown).
coverage.schema.json — coverage artifact emitted by oracle tests (lines covered or function-level coverage).
trace.schema.json — dynamic trace artifact (structured path or event log) proving reachability/unreachability.
submission.schema.json — submission format (tool/run metadata, per-sink predictions, explanations).
examples/ — minimal valid examples for each schema.
benchmark/schemas/benchmark-manifest.schema.json — kit manifest with hashed artifacts, resource limits, and optional DSSE signatures.

Validate quickly

# install minimal deps (offline-friendly, pinned)
python -m pip install -r ../tools/requirements.txt

# validate individual files
python ../tools/validate.py case examples/case.sample.yaml
python ../tools/validate.py entrypoints examples/entrypoints.sample.yaml
python ../tools/validate.py truth examples/truth.sample.json
python ../tools/validate.py coverage ../cases/js/unsafe-eval/outputs/coverage.json
python ../tools/validate.py trace ../cases/js/unsafe-eval/outputs/traces/traces.json
python ../tools/validate.py submission examples/submission.sample.json

# or validate everything in one shot
python ../tools/validate.py all examples