Reproduce “Same Model, Different People: Numerical Execution Contracts and the Reproducibility of Fairness Audits” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Erzheng Zhang, Agnes Chu

The authors test how arithmetic and deployment configurations affect thresholded classification decisions and fairness audits. Tier 1 holds serialized weights fixed to isolate arithmetic; Tier 2 includes reduced-precision configurations that may also change weights or software. Tier 1 flipped decisions but changed no prespecified unweighted audit verdicts; a 24B model’s FP8 pass was conditional and did not show greater fairness.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.