fix: store scripted assistant content in eval + session-affinity recall boost#431
Merged
Conversation
…ll boost Two fixes for eval recall quality: 1. Store scripted assistant turns in temporal during eval replay. The gateway stores its own API response (which differs from the scripted content), so the scenario's actual content was never searchable via recall. This fixes m4 (5000ms timeout error) going from 1.2 → 4.0. 2. Session-affinity RRF boost in recall: when scope=all and sessionID is known, add extra RRF lists for same-session temporal/distillation results. This boosts current-session content over cross-session LTM entries. Eval score: 3.69 → 4.38 at 400K inflation (13 of 15 questions >= 4.0).
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
Two fixes for eval recall quality that improve the inflated CM-1 score from 3.69 → 4.38.
Changes
1. Store scripted assistant turns in eval replay (
harness.ts)The eval replays scripted conversation turns through the gateway, but the gateway stores its own API response (which differs from the scripted content) as the assistant temporal message. The scenario's actual content (e.g., "staleness check vs flock", "5000ms timeout error") was never stored in temporal and therefore never searchable via recall.
Fix: After each scripted assistant turn is added to history, store it directly in temporal with the replay session ID. This makes the scenario's ground-truth content available for recall search.
Impact: m4 (exact error message) went from 1.2 → 4.0.
2. Session-affinity RRF boost (
recall.ts)When
scope=allandsessionIDis known, add extra RRF lists for same-session temporal and distillation results. This gives current-session content a ranking boost over cross-session LTM entries that may match keywords but lack session-specific context.Eval Results (CM-1, 400K inflation)
Remaining failures: m3 (2.2, LTM entry outranks session content), h4 (1.7, initial wrong hypothesis not found).
Tests