fix regex bug in browsecomp#67
Open
tengyaolong2000 wants to merge 1 commit into
Open
Conversation
match.group(0) captures "correct: yes/no" when we just want "yes" or "no" (without "correct:") , match.group(1) fixes this
|
Great to see a PR to fix this issue |
JoyboyBrian
added a commit
to Osmosis-AI/simple-evals
that referenced
this pull request
Jul 14, 2026
grade_sample returned match.group(0) ("correct: yes"/"correct: no"),
which never equals the bare "yes"/"no" the aggregation compares
against, so is_correct and is_incorrect were always False and reported
accuracy was always 0. Return the captured group instead.
Same one-line fix as unmerged upstream PRs openai#67 and
openai#93.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
JoyboyBrian
added a commit
to Osmosis-AI/simple-evals
that referenced
this pull request
Jul 14, 2026
* Add BrowseComp web-search wrapper for the Responses API
Upstream simple-evals cannot produce meaningful BrowseComp results:
1. grade_sample returns match.group(0) ("correct: yes"), while
aggregation compares against the bare string "yes", so reported
accuracy is always 0 (upstream PRs openai#67 / openai#93, still unmerged).
2. No sampler supports web search, so the official entrypoint runs
this browsing benchmark closed-book; questions were explicitly
filtered at construction time to be unanswerable without browsing
(GPT-4o closed-book: 0.6%).
This wrapper adds a Responses API sampler with the hosted web_search
tool, accepts both grader output forms when aggregating, and provides
a deterministic oracle mode to validate the grading pipeline before
paying for search runs. Reference answers are never written in
plaintext to result files.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* Fix BrowseComp grader regex group extraction
grade_sample returned match.group(0) ("correct: yes"/"correct: no"),
which never equals the bare "yes"/"no" the aggregation compares
against, so is_correct and is_incorrect were always False and reported
accuracy was always 0. Return the captured group instead.
Same one-line fix as unmerged upstream PRs openai#67 and
openai#93.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
8 tasks
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
2 participants
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.
match.group(0) captures "correct: yes/no" when we just want "yes" or "no" (without "correct:") , match.group(1) fixes this