Search Relevance Laboratory
Index a corpus, judge relevance, and compare lexical ranking experiments.
Application guideVerifying your session…
Using Search Relevance Laboratory
Measure whether a lexical retrieval configuration improves rankings on judged queries.
Inverted index · BM25 ranking · Graded relevance
- Inputs and working data
- tooldrobe.search-experiment version 1 with a corpus, unique document IDs, per-query graded judgments, evaluation depth and candidate BM25 parameters.
- Working in the editor
- Corpus builder, per-document relevance grid, parameter controls and side-by-side rankings for every query.
- What can I produce?
- Two ranked result lists per query with comparable metrics, editable experiment JSON and reproducible evaluation report.
- What are the limits?
- 1–1,000 documents, two million corpus characters, up to 100 judged queries, depths 1–100, grades 0–3. No stemming or remote crawling.
- Example
- Two-document corpus: removing frequency saturation at k1=0 produces an alphabetical tie and changes nDCG from 1 to 1/log2(3).