docs: restructure card with numbered sections (Overview / Evaluation / Future Work / Meta Info)

docs: replace prose tildes with U+2248 to avoid GFM strikethrough rendering

docs: add LLM-assisted quality audit (78/100), duplicate-query finding, qrels-enrichment plan; translate to English

docs: clarify per-pair length variance vs systematic bias + data-side vs serving-side BM25 hybrid

docs: rename Roadmap section to Future Work

docs: add surface-signal measurements to Limitations + Roadmap section

Team re-name

License -> cc-by-4.0 + section rename (Licensing & Limitations -> Limitations)

Highlight test split as Golden set (TL;DR + Splits section + Quick Start)

v3 - source_chunk_id parent_doc split + relevance labels (Phase 1+2)

v3 - source_chunk_id parent_doc split + relevance labels (Phase 1+2)

v3 - source_chunk_id parent_doc split + relevance labels (Phase 1+2)

v3 - source_chunk_id parent_doc split + relevance labels (Phase 1+2)

v2 — 45,394 triplets (K=10 LLM judge, fallback 0.46%)

initial commit