← ClaudeAtlas

elasticsearch-relevance-testslisted

Use before changing analyzers, synonyms, boosts or query templates on Elasticsearch, OpenSearch or Solr search, so a relevance baseline exists. Builds judgment lists and _rank_eval tests with precision, recall, MRR and DCG, plus OpenSearch and Solr specifics in its references.
testland/qa-skills · ★ 0 · Testing & QA · score 72
Install: claude install-skill testland/qa-skills
# elasticsearch-relevance-tests Per the [Elasticsearch Rank Eval API], the `_rank_eval` endpoint "evaluates search result quality across typical queries using relevance metrics." ## Engine routing This skill is the single home for term-based engine relevance testing. The core `_rank_eval` workflow below is authored against Elasticsearch; the per-engine deltas live in references: | Engine | Where | Delta | |---|---|---| | Elasticsearch | this SKILL.md | Built-in `_rank_eval`; the canonical workflow | | OpenSearch | [references/opensearch.md](references/opensearch.md) | `_rank_eval`-compatible fork; Search Relevance Workbench, neural query DSL, hybrid BM25 + neural pipelines, ES-to-OS migration parity | | Apache Solr | [references/solr.md](references/solr.md) | No `_rank_eval`; `debugQuery` explain, LTR feature/model store REST, eDisMax tuning, external nDCG harness | ## When to use - Search-driven product (e-commerce, docs site, internal portal) where relevance regression directly affects business outcomes. - Pre-deploy gate before changing analyzers, synonyms, boosts, query templates. - A/B baseline: capture today's NDCG/MRR before tuning so you can prove improvement (or detect regression). ## Step 1 - Build the judgment list A judgment is `(query, doc_id, rating)`. Ratings: 0 = irrelevant, 1 = somewhat, 2 = relevant, 3 = highly relevant (4-point scale). Build judgments via: | Source | Method | |---|---| | Query logs + click data | Click model (clicked = ≥1, mu