results: real IPT-vs-judges result on the Obfuscation Atlas MBPP-Honeypot - #7
Merged
Merged
Conversation
…ypot Second non-synthetic artifact, public data only (taufeeque/mbpp-hardcode). IPT (judge-free): recall 0.994/0.998 on blatant/obfuscated hacks, 0% FP, 0 model calls, $0. A GPT-4o-mini code-judge matches recall but flags 40% of genuine code (unusable as an in-the-loop verifier). Honest scope: Helff et al. = mechanism; ours = coding extension; the win is operational, not a recall win. results/atlas_honeypot_ipt.json + reproducibility-notes section + README link. Outside evidence/, validator unaffected (green). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Second non-synthetic artifact (public data only). IPT judge-free: recall 0.994/0.998, 0% FP, 0 model calls, $0; GPT-4o-mini judge matches recall but 40% FP on genuine (unusable as an in-the-loop verifier). Honest scope: Helff et al. = mechanism, ours = coding extension; operational win, not a recall win. Outside
evidence/, validator green.🤖 Generated with Claude Code