Tarot-Draw
Three binary forecasting models trained on the same 81,870 questions. A controlled study of what model size changes, and what it does not.
Research
Releases
Models, evaluations, and the evidence needed to inspect their claims.
2026
Standard
Claims should travel with evidence
Public releases link their model artifacts, evaluation inputs, metric code, limitations, and reproduction instructions. Where a result is not statistically resolved, the release says so.