Reproduce “VerifiedLab: Harnessing Knowns for Unknowns” 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

Yu Yao, Jiayi Dong, Yang Yang, Shuiwang Ji, Ju Li

VerifiedLab builds records from perovskite solar-cell papers using a generator, a separate rule-based verifier, and category-level reliability. Its database contains 10,258 champion-device records. Verification rules beat the largest unverified model in three of four families. For 858 qualifying 2025 devices, T5-base predicts reverse-scan PCE with 2.20 ± 0.06 percentage-point MAE, versus 2.87 ± 0.01 for LightGBM; this is one-year-ahead prediction, not causal discovery.

Summary by alphaXiv.

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