Monolith Research
In plain terms
Numerai is a tournament where thousands of data scientists submit stock-ranking models on obfuscated data, scored on how much they add beyond the crowd. This model is built to add exactly that.
How it works
Rather than predict the target directly, it models the residual of the target after removing the crowd’s direction — an ensemble that targets originality, not agreement.
What it’s tested against
Scored independently by Numerai out-of-sample, with a contribution-above-crowd that stays positive across both halves of the validation period; live since June 2026.
0.386 MMC-Sharpe
0.976 Deflated Sharpe
Jun 2026 Live since
Data
Numerai’s obfuscated tournament dataset.
Live — an independently-scored record being written in public.