feat: lower echo threshold + clustering for broader behavioral pattern detection#377
Merged
Conversation
…n detection Lower ECHO_THRESHOLD from 0.78 to 0.65 (CANDIDATE_THRESHOLD) to cast a wider net for pattern candidates. Add greedy clustering that groups candidates by mutual similarity (CLUSTER_SIMILARITY >= 0.72). A cluster spanning 3+ distinct sessions triggers pattern extraction. This catches patterns where individual pairs don't reach 0.78 but a cluster of related observations across sessions reveals the behavior. Eval PR-2: 4.58 → 4.63 (tail-window: ~4.85) - Tests alongside implementation: 3.6 → 5.0 - ORM stance: 4.6 → 5.0
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Lowers the pattern echo similarity threshold and adds greedy clustering to detect behavioral patterns that individual pair comparisons miss.
Eval Results
PR-2 (implicit preferences): 4.58 → 4.63 (tail-window: ~4.85)
How it works
Before: Single threshold at 0.78 — requires each candidate to be highly similar to the current segment. Misses patterns where surface text varies across sessions.
After: Two-stage approach:
This catches 'tests alongside implementation' — individual test-related observations are only 0.65-0.75 similar to each other (below the old 0.78 threshold), but they form a coherent cluster across sessions.
Files Changed
packages/core/src/pattern-echo.ts— clustering logic, lower threshold, new constants