Beat the model

Two real NFL players, one real week. Pick who scored more. Then see what a machine-learning model predicted — and what actually happened.

The model was trained only on seasons before the one it is predicting, so it had never seen these games. It gets the call right about 68% of the time. Most people are confident they can do better.

Predictions come from gradient-boosted quantile models over 17,307 player-weeks, 2022–2024, using recent usage, target share, and betting-market game totals. Everything on this page runs locally in your browser. No signup, no tracking.

You 0 Model 0 Streak 0

The model's edge over the dumb baseline is about two percentage points. That is real but small — roughly 90% of week-to-week fantasy scoring is irreducible luck, so the realistic ceiling for anyone, human or machine, is near 70%.

Browse any week