Repository navigation
mnemo 6.2.0
Retrieval fusion rebalance -- rank-level weighted RRF is the new default ranking. Measured +0.23 hit@5 overall, and it wins on BOTH halves of the query distribution.
Retrieval was the weak link the v6.0.0 reliability baseline exposed. The 6-term
weighted sum blended the dense-vector and BM25 rankers at the SCORE level, which
under-weighted the far stronger lexical signal. The fix is rank-level fusion.
-
config.fusion_mode-- the vector<->lexical relevance core is now a swappable
seam:weighted_rrf(new default),weighted_sum(the previous behaviour, kept
as the escape hatch and the comparable baseline),bm25_lead. The CONTEXT terms
(graph / recency / type / project) and the multiplicative finishers (exact-name
boost, isolation penalty) are orthogonal and unchanged in every mode, so a
mode A/B measures one change rather than two. Every mode scores into the same
[0, alpha+zeta]band. An unknown mode fails open to the default. -
Measured on an 80-query set (42 lexical + 38 conceptual) against one pinned
corpus of 18,656 nodes, hit@5:mode lexical conceptual overall weighted_sum(old default)0.500 0.132 0.325 bm25_lead0.690 0.079 0.400 weighted_rrf(new default)0.881 0.184 0.550 rrf_k=60is the literature default, not a value fitted here (a 10..200 sweep is
smooth and unimodal around it), and everyw_bm25in 0.5..1.0 beat the old
default. Tunable viarrf_k/rrf_weight_bm25/rrf_weight_vector. -
Eval set expanded 42 -> 80 and tagged by
query_type(lexical|conceptual);
mnemo evalnow breaks hit@1/hit@5/MRR down per type. This is what makes the
lexical-vs-semantic tradeoff of a ranking change visible instead of hidden inside
one blended average -- and it is what caughtbm25_leadregressing conceptual
recall. The 38 conceptual queries were authored from corpus ground truth and
adversarially verified (a candidate was rejected if a keyword search on the
prompt's own words found the answering file -- that would make it lexical). -
Known limitation:
hit@1is roughly flat (0.175 -> 0.188). RRF reliably gets
the right node INTO the top-5 without sharply ordering the top; mnemo injects
k=5 cited nodes, so hit@5 is the operative metric. -
Known weak spot (next lever, now evidence-backed): conceptual retrieval is
poor in absolute terms (0.184). Prose questions surface CODE nodes because the
corpus is ~17.4k code vs ~1.2k prose nodes and the type-priority weight cannot
overcome those odds. Fusion cannot fix this; prose-vs-code ranking and a harder
project-isolation penalty can.