Skip to content

Add wiki-bench v0 fixture-only retrieval benchmark - #845

Open
tobi wants to merge 3 commits into
mainfrom
wiki-bench-v0-fixture
Open

Add wiki-bench v0 fixture-only retrieval benchmark#845
tobi wants to merge 3 commits into
mainfrom
wiki-bench-v0-fixture

Conversation

@tobi

@tobi tobi commented Aug 13, 2026

Copy link
Copy Markdown
Owner

Summary

Fixture-only wiki-bench v0 for retrieval quality over the public-appearances wiki. This deliberately does not vendor any private wiki page bodies (unlike closed PR #844, which was deleted for that reason).

  • Ships src/bench/fixtures/wiki-v0.json (44 queries; expected_files are wiki-relative paths only)
  • Always-on schema test: test/wiki-bench-fixture.test.ts
  • Optional BM25 quality floors: test/wiki-bench-bm25.test.tsskips unless QMD_WIKI_BENCH_DOCS or QMD_WIKI_PATH points at a local corpus directory
  • Docs: docs/WIKI_BENCH.md

No test/wiki-bench-docs/**, no sync script that writes corpus into the tree, no wiki .md bodies anywhere in this PR.

Privacy

Included Not included
Fixture JSON (queries + expected paths/ids) Private tobi/wiki page bodies
Schema + env-gated BM25 tests test/wiki-bench-docs/**
docs/WIKI_BENCH.md Any sync that vendors corpus into the repo

Default CI (no corpus env) stays green.

How to validate

# Always (no corpus needed)
npx vitest run test/wiki-bench-fixture.test.ts

# Optional local BM25 floors
export QMD_WIKI_BENCH_DOCS=/path/to/wiki/wiki   # or QMD_WIKI_PATH
npx vitest run test/wiki-bench-bm25.test.ts

# Full multi-backend harness after indexing a local checkout
qmd collection add /path/to/wiki/wiki --name wiki-bench
qmd bench src/bench/fixtures/wiki-v0.json -c wiki-bench

See docs/WIKI_BENCH.md for details.

Test plan

  • npx vitest run test/wiki-bench-fixture.test.ts (green, no wiki docs)
  • npx vitest run test/wiki-bench-bm25.test.ts (4 skipped + gate pass when env unset)
  • Confirmed commit has only: fixture JSON, two test files, docs/WIKI_BENCH.md
  • CI on this PR

Ship queries + expected paths only (no private wiki page bodies). Includes
schema validation that always runs in CI, and optional BM25 floors that skip
unless QMD_WIKI_BENCH_DOCS or QMD_WIKI_PATH points at a local corpus.
@PowderAddicts

PowderAddicts commented Aug 13, 2026 via email

Copy link
Copy Markdown

tobi and others added 2 commits August 13, 2026 00:38
Indexes a temporary wiki-bench collection outside the git tree from
QMD_WIKI_PATH/wiki or QMD_WIKI_BENCH_DOCS, runs BM25 via runBenchmark,
and fails if exact/alias quality floors are not met. No wiki page bodies
are committed.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants