Skip to content

Add combined GPT-or-Claude recall of injected errors#97

Open
dangng2004 wants to merge 1 commit into
mainfrom
feat/combined-recall
Open

Add combined GPT-or-Claude recall of injected errors#97
dangng2004 wants to merge 1 commit into
mainfrom
feat/combined-recall

Conversation

@dangng2004

@dangng2004 dangng2004 commented Jun 5, 2026

Copy link
Copy Markdown
Contributor

Pulled out of #91, where it was out of scope (recall numbers, not overlap analysis).

Computes recall on the 24-paper frontier subset where a perturbation counts as detected if either GPT-5.5 or Claude Opus 4.7 detected it under the progressive (OpenAIReview) method. Perturbation IDs are namespaced by (results_dir, paper, type) to avoid cross-paper collisions. Feeds tab:recall-overall in perturbation.tex.

Stdlib-only; reads local result JSONs (gitignored).

A perturbation counts as detected by the combined system if either
GPT-5.5 or Claude Opus 4.7 detected it under the progressive
(OpenAIReview) method. Feeds tab:recall-overall in perturbation.tex.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
@dangng2004 dangng2004 changed the title Add combined GPT-or-Claude recall for the frontier subset Add combined GPT-or-Claude recall of injected errors Jun 5, 2026
@dangng2004 dangng2004 marked this pull request as draft June 5, 2026 22:27
@dangng2004 dangng2004 marked this pull request as ready for review June 6, 2026 02:31
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