Show HN: ReadoutBench – forecast 653 clinical trials before they report

Category: ai-ml

Tags: benchmark, clinical-trials, forecasting, ai-ml, reproducibility

Score: 5.8/10 (Innovation: 7, Technical: 5, Documentation: 6, Utility: 5)

ReadoutBench is a forecasting benchmark that asks participants to predict the outcomes of 653 real clinical trials before they report results. It uses proper scoring (Brier score) and includes reproducibility artifacts like fingerprints and rebuild scripts. It's interesting as a live, leakage-resistant benchmark for AI models on a high-stakes domain.

Target audience: AI researchers, data scientists, clinical trial analysts

Repository: https://readoutbench.com · Python · MIT

View on Hacker News