fix(bench): scrub the real run id from evaluator workspace inputs - #7
Merged
Conversation
The eval workspace directory is named by a blind handle, but the copied inputs/verification.json still carried the real runId, so evaluator agents could learn which model produced the artifact. Rewrite every occurrence of the run id in the verification record to the blind handle before writing it into the workspace, and sweep all harness-written workspace files in the pipeline test to assert the run id never leaks (artifact/ exempt: sealed bytes pinned by the digest check). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
pnpm bench scorebuilds a blind-handled evaluator workspace, but the copiedinputs/verification.jsonstill carried the realrunId(e.g.kimi-k3), defeating blinding — evaluator agents read that file and learned which model produced the artifact.createEvalWorkspacenow rewrites every string occurrence of the real run id (values and keys) in the verification record to the blind handle before writing it intoinputs/, covering check output and embedded paths, not just therunIdfield.inputs/(rubric, filtered TASK.md, projected specs, variants, manual/reference image) plusEVAL.mdand the assessment template — none carry run identity.Test plan
artifact/exempt: sealed bytes pinned by the digest check).pnpm test(36/36),pnpm typecheck,pnpm lintclean on Node 24.claude-fable-5stage 1: scrubbedrunIdreadsrun-b0aa66eba48b; recursive grep outsideartifact/finds no occurrence of the run id.🤖 Generated with Claude Code