MBModelBall

Tool · Match prediction

Calibrate a prediction

Five AI models, the same prompt, four layers of bias correction derived from the published behavioural fingerprinting study. Submit a match — we'll run the full pipeline and return the corrected probability plus the bias dimensions that were active for this matchup.

The pipeline queries five AI models in parallel and applies the four-layer calibration. Typical wall-clock: 20–45 seconds.

Methodology and validation in the two preprints on /research. Predictions are made before kickoff and time-stamped in our database; we do not adjust them retroactively.