Skip to content

Change diff scoring to only penalize outlier-large diffs#101

Merged
that-github-user merged 1 commit into
mainfrom
issue-97-scoring-outliers
Mar 28, 2026
Merged

Change diff scoring to only penalize outlier-large diffs#101
that-github-user merged 1 commit into
mainfrom
issue-97-scoring-outliers

Conversation

@that-github-user

Copy link
Copy Markdown
Owner

Summary

  • Replace "smaller is better" with outlier penalty: only diffs >2x median are penalized
  • Normal and thorough diffs get equal diff scores (no bias toward minimal changes)
  • Updated docs/architecture.md scoring section
  • 1 new test for outlier detection behavior

Generated by thinktank Opus — 5 agents, 3/5 pass tests, Agent #1 recommended (130.6).

Change type

  • New feature

Related issue

Closes #97

How to test

npm test  # 93 tests pass

Breaking changes

  • Scoring formula changed — recommendations may differ from previous behavior

🤖 Generated with thinktank (Opus)

Replace "smaller is better" bias with outlier detection: agents with
diffs >2x the median are penalized proportionally. Normal and thorough
diffs get equal scores. Updates scoring formula and architecture docs.

Generated by thinktank Opus (5 agents, 3 pass, Agent #1 recommended).

Closes #97

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
@that-github-user

Copy link
Copy Markdown
Owner Author

Self-review: Outlier threshold at 2x median is reasonable — catches runaway agents without penalizing thoroughness. Docs updated. 93 tests pass, CI green.

@that-github-user that-github-user merged commit 0ff64bd into main Mar 28, 2026
4 checks passed
@that-github-user that-github-user deleted the issue-97-scoring-outliers branch March 28, 2026 20:50
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Reconsider diff size scoring: smaller isn't always better

1 participant