3 Commits

Author	SHA1	Message	Date
john	fb3a442e73	Improve MemFuse recall via hybrid ranking and candidate generation. ... Add RRF fusion with English word-level lexical scoring, tiered keyword fetch, and vector margin expansion (0.15/200) to fix pre-rank truncation; wire DashScope embedding bench path and update baseline to 28.8% recall@20. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-09-02 13:41:12 +08:00
john	2c0d903ecf	Improve pgvector keyword recall ranking without changing hybrid sort. ... Rank ILIKE matches by lexical coverage instead of recency and attach lexical scores to keyword rows so MemFuseBench candidate recall rises sharply while Chinese hybrid ordering stays unchanged. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-09-02 10:01:42 +08:00
john	0b6f79ae16	Add MemFuseBench retrieval harness and first baseline report. ... Wire an offline benchmark over the production pgvector ranking path so Memory V2 recall can be measured with candidate/ranking loss split before lifecycle work lands. Co-authored-by: Cursor <cursoragent@cursor.com>	2026-09-02 09:50:57 +08:00