Conversation
…scoring, overhead metric - Fix base64 classification gap: add FORCE_T0_PATTERNS entry for 40+ char base64 strings, add to HARD_T0_REASONS. Hex/UUID probes already pass, base64 probe now passes too. - Sentence adjacency scoring: boost sentences sharing entities with neighbors (+2 one side, +3 both) to improve topical coherence in summaries. Uses exported extractMessageEntities from importance.ts. - Compression overhead ratio metric: computeOverheadRatio() measures compress() wall-clock time vs estimated LLM inference time. Displayed as OvhdR column in quality bench output. - High-entropy quality bench scenario with hex dump, UUID array, base64 blob, and mixed entropy+prose messages (4/4 probes pass). - Unit tests for base64/hex/UUID classification and compress preservation, adjacency scoring behavior, and documented known limitations (UUID gap, camelCase false-positive on base64 pattern).
v1.4.0 baseline with high-entropy content scenario. Zero regressions vs v1.3.0 across all 13 shared scenarios. New scenario: High-entropy content at 1.35x ratio, 100% entity retention, 4/4 probes.
Settings bar: - depth dropdown (gentle/moderate/aggressive/auto) - relevance toggle + threshold input - flow, importance, contradiction, coreference, clustering toggles - budget strategy dropdown (binary-search/tiered, visible when budget on) - visual divider between v1 and v2 controls Stats bar: - quality_score, entity_retention, structural_integrity chips - messages_relevance_dropped, importance_preserved, contradicted chips - color-coded: green >=90%, amber >=70%, red <70% Examples: - "Q&A + corrections" — demonstrates flow + contradiction detection - "Topic-scattered" — 3 interleaved topics for clustering demo Help panel: V2 Features section with all new options explained
A/B testing showed adjacency scoring (+2/+3 boost for entity-linked neighbor sentences) produces identical results across all quality bench scenarios. The summarizer budget is wide enough that sentence selection pressure never triggers the tiebreaker. Removing to avoid dead complexity.
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
base64_contentpattern toFORCE_T0_PATTERNSandHARD_T0_REASONS— these are now preserved verbatim.computeOverheadRatio()measures compress() wall-clock time vs estimated LLM inference cost. Displayed asOvhdRcolumn in quality bench output.Verification
npm run lint && npm run format:checkcleannpm run bench:quality --checkpassesTest plan
npm test— all 671 tests passnpm run bench:quality— high-entropy probes 4/4npm run bench:quality -- --check— no regressions vs v1.3.0npm run lint && npm run format:check— clean