Repository navigation
test(bench): external repo benchmark - plain grep wins on a small repo - #110
Merged
Merged
Conversation
B1 lives in our own repository, so B2 runs the same construction on
psf/requests: the repo, its history, its commit messages and the gold all come
from other people. Corpus pinned at e331a288 (2026-03-24), 64 files, 786 chunks.
25 commits (40 asked; no merges, <=20 files, >=1 .py outside tests/docs)
arm hit@1 hit@5 95% CI
grep 6 19 [0.57, 0.89]
search 5 15 [0.41, 0.77]
neighbours 9 13 [0.33, 0.70]
tool 8 10 [0.23, 0.59]
On a 64-file repository a term-count ranking over file text is simply good:
messages name modules and functions, and there are few files to confuse it. The
tool's demotions, gate and merge add nothing here and take something away.
This corrects an earlier claim. "We beat grep 157/160 vs 147/160" was about
symbol lookup on the generated panel; it says nothing about commit localization
and does not generalise to a small external repo.
Caveats, stated rather than buried: n=25 (one task = 4 points, intervals overlap
between search and tool, but grep's lower bound is above the tool's upper); the
grep baseline reads corpus files by path from the repo's ls-files, so files
created after the base are empty for it - that biases grep down and it still
wins; one repository and one base are not a claim about small repos in general.
The harness is `experiments/bench_2026_11/external.py`; the result file records
repo, base, task count and the table, with no commit messages stored.
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars.
 |
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
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.
test(bench): the benchmark on an external repo — plain grep wins
B1 lives in our own repository, so B2 runs the same construction on
psf/requests: the repo, its history, its commit messages and the gold all come
from other people. Corpus pinned at e331a288 (2026-03-24), 64 files, 786 chunks.
On a 64-file repository a term-count ranking over file text is simply good:
messages name modules and functions, and there are few files to confuse it. The
tool's demotions, gate and merge add nothing here and take something away.
This corrects an earlier claim. "We beat grep 157/160 vs 147/160" was about
symbol lookup on the generated panel; it says nothing about commit localization
and does not generalise to a small external repo.
Caveats, stated rather than buried: n=25 (one task = 4 points, intervals overlap
between search and tool, but grep's lower bound is above the tool's upper); the
grep baseline reads corpus files by path from the repo's ls-files, so files
created after the base are empty for it - that biases grep down and it still
wins; one repository and one base are not a claim about small repos in general.
The harness is
experiments/bench_2026_11/external.py; the result file recordsrepo, base, task count and the table, with no commit messages stored.