We tried to beat the market.
The market won.
Sooth tested prediction against the closing market and published the loss. No version reached the 52.38% ATS needed to break even at standard -110 odds. That result changed the product: Sooth now focuses on the measurable part bettors can control, the price a sportsbook offers. Every figure below regenerates from published evidence.
The market is better calibrated than our model.
Expected calibration error is the gap between what we predict and what actually happens. Lower is better. The de-vigged market beats our best model on both samples. We publish this rather than the one framing where we'd look good, because a number nobody can check is a claim, not a measurement.
Per probability band: what we predicted vs how often it happened. A positive gap means we were overconfident.
How far the weaker line source (nflverse) sits from the real consensus close.
Regenerate every figure: python scripts/published_figures.py.
Raw data: figures.json. Full spec on the
methodology page; every graded slate is in the
ledger.
Analysis and entertainment only. Not a sportsbook. We do not accept wagers or hold funds, and we never sell winning picks — the pick engine sells timing and instrumentation on a slate this page shows losing. Past results do not indicate future results. Must be 21 or older where lawful. Problem gambling help: 1-800-522-4700, available 24 hours.