Skip to content

Plain-language: trust the whole-question ranking; fix skeleton panic; baseline comparison script - #128

Merged
ParsaVictor merged 1 commit into
mainfrom
phase-k/plain-language-evidence
Oct 1, 2026
Merged

ParsaVictor merged 1 commit into
mainfrom
phase-k/plain-language-evidence

Conversation

@ParsaVictor

Copy link
Copy Markdown
Owner

Concept precision 0.392→0.404, ripgrep holdout 0.152→0.156, recall and the nine sets unchanged. Fixes an index-out-of-bounds panic in skeleton.rs (span past EOF). Adds scripts/compare_baselines.py (BM25 + Aider RepoMap).

🤖 Generated with Claude Code

…o longer panics on a span past EOF; baseline comparison script

- a question with no file path and no identifier but prose words drops
  every word-by-word guess seed outside the ranker's picks:
  concept precision 0.392 -> 0.404, ripgrep holdout 0.152 -> 0.156,
  recall unchanged, the nine sets unchanged
- skeleton: a span whose start line is past the end of the file is
  skipped (index out of bounds panicked the whole query); docstring_block
  clamps its end
- scripts/compare_baselines.py: plain BM25 and Aider's RepoMap ranking on
  any gold set, recall/precision at k = 1, 3, 5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@ParsaVictor
ParsaVictor merged commit d7d1120 into main Oct 1, 2026
3 checks passed
@ParsaVictor
ParsaVictor deleted the phase-k/plain-language-evidence branch October 1, 2026 09:42
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.

1 participant