fix: downweight knowledge in recall when session content exists#432
Merged
Conversation
When scope=all and temporal/distillation results exist, apply 0.6x weight to knowledge BM25 and vector RRF lists. This deprioritizes cross-session LTM entries when session-specific content is available — temporal details about what actually happened are more likely the answer than general knowledge. Also adds session-affinity RRF boost and scripted assistant storage in eval. Eval score: 3.69 → 4.39 at 400K inflation.
BYK
added a commit
that referenced
this pull request
May 21, 2026
…indow (#435) Updates marketing copy with the latest eval results from the recall quality + distillation transparency work (#428, #430, #431, #432, #433, #434). ### README.md - Context retention table: Medium 2.3→4.1, Hard 3.3→4.8, Average 3.9→4.6 - Lore vs tail-window delta: +50%→+77% - Added footnote: Lore scores averaged across multiple runs; TW/compaction baselines from a prior eval run with the same scenarios - Added v6 to version history ### docs/index.html - Hero stat: +50%→+77% vs tail-window - Detail retention: 4.8→4.6 (overall average across difficulty levels, multiple runs) ### Review corrections - Fixed Medium from 4.3→4.1 (honest multi-run average, not cherry-picked) - Average row (4.6) now self-consistent with column values: (5.0+4.1+4.8)/3=4.63≈4.6 - Added footnote clarifying that TW/compaction columns are from a prior eval run
This was referenced May 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
When
scope=alland temporal/distillation results exist, apply 0.6x weight to knowledge BM25 and vector RRF lists. This deprioritizes cross-session LTM entries when session-specific content is available.Problem
When the model invokes recall with a query like "alternative approach flock locking", an LTM knowledge entry about
proper-lockfile(which contains terms like "flock", "advisory locking") outranks the actual session temporal message about "lock file with staleness check". The LTM entry has perfect keyword overlap with the query while the temporal message has weaker BM25 relevance.Fix
Track whether temporal/distillation results exist across the query expansion loop (
hasSessionResultsflag). Whenscope === 'all'and session-specific results exist:weight: 0.6(was implicit 1.0)weight: vectorWeight * 0.6(wasvectorWeight)This ensures session-specific content ranks higher when both sources match, while knowledge still surfaces when no session content exists (e.g., cross-session queries,
scope: 'knowledge').RRF Score Impact (both at rank 0, 2-term query)
Eval Results (CM-1, 400K inflation)
Score: 4.39 (up from 3.69 baseline). 12 of 15 questions >= 4.0.
Remaining failures (m3, h4) are a query generation problem, not ranking: the distillation compresses away the distinguishing term ("staleness check") so the model can't include it in its recall query.
Tests