VERITAS

University of Chicago, University of IndonesiaBuilt by: research, non-profit or individualStatus: Not statedYear: 2026

Overview

VERITAS is a general-purpose replication tool for scientific research built around command-line coding agents, from Haokun Liu and Chenhao Tan of the University of Chicago and Filbert Aurelian Tjiaranata of the University of Indonesia, who described it in a July 2026 preprint. It takes a paper, its code or both, extracts the paper’s claims, runs the methods and fixes problems as they arise, and judges each claim against the results, producing a Replication Score weighted by the importance of each claim, a log of every fix with its severity, and the corrected code.

The authors tested it on 65 papers in computer science, social science, medicine and astrophysics from two benchmarks, CORE-Bench and ReplicationBench.

Benefits and outcomes

  • The authors report that VERITAS led on every metric on both benchmarks, against two Claude Code baselines using the same model and environment, for example passing 88.9% of CORE-Bench Hard tasks under the official scorer, compared with 80.0% and 84.4% for the baselines.

Sources

This entry was written with the help of AI, from the sources listed. Please check the sources before relying on it, and let us know if anything needs correcting.