# Benchmark fixtures are deterministic output of mappers running against pinned
# upstream sources (HF datasets, GitHub artifacts, Latitude API in the future).
# We regenerate them via `pnpm --filter @tools/ai-benchmarks benchmark:fetch`
# rather than committing the generated JSONL.
#
# The .meta.json sidecars record the mapper source hash so benchmark:run can
# detect stale fixtures — regenerated whenever benchmark:fetch runs.
#
# See prd/llm-flagger-evaluation-system.md — Decisions #1 and #15.
*.jsonl
*.meta.json
