Filters are evaluated once into an id set used by both search branches.
Adds tests for age/genre/mechanic filters with a search term, rank 2 vs 3
and null p_limit. approxNormalize uses NFKD. Perf script checks confirm_local
value and unix socket and re-analyzes after rollback. Adds
docs/specs/3-katalog-suche-umsetzung.md (measurements, log check incl. GET
finding), release checklist section 2a and the CSV-only note for alt titles.
Refs #3
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Rank 4 and the genre/mechanic filters ran a correlated subplan per game
(13 s at 10 000 games). Now: one set-based word coverage, filters as one hashed
IN set, 'like all' patterns, similarity only for hits. Adds pgTAP cases found
by mutation testing (paging by relevance, alt title choice, duplicate words,
mechanic duplicates) and a Deno case for mark characters.
Refs #3
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Replace tests that could not fail (manual_fields stability, no-op update), select runbook games by id, add grant matrix and forged-admin tests, assert vocabulary seeding and Deno checksum in CI, clarify the set local convention and API notes.
Refs #2
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Sources, publishers, hidden/merged columns, manual_fields triggers, catalog_games view and column grants on games (KAT-05, PR 2a). Fixtures for the tests are generated from supabase/catalog/testdata.
Refs #2
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
