How a win probability becomes a point spread
The model prices every game as a percentage first and a spread second. The conversion is one division — and understanding it tells you when a number disagrees with the market.
Every game on this site carries two numbers: a win probability and a spread. They are not separate estimates. The probability comes first, and the spread is a translation of it — which means if you understand the translation, you can tell precisely where the model and a sportsbook disagree.
Step one: add up the rating edge
Start with the two ratings. Add 25 points to the home team. Add up to 25 more for a rest advantage, at 5 points per extra day off. The result is a single number: the adjusted rating gap.
Step two: run it through the logistic curve
The gap goes into the standard Elo expectation formula — one divided by one plus ten raised to the negative gap over 400. That curve is the whole prediction. It is steep in the middle and flat at the edges, which is exactly right for football: the difference between a 0-point and a 100-point edge is huge, and the difference between a 400-point and a 500-point edge is almost nothing.
| Adjusted rating edge | Win probability | Model spread |
|---|---|---|
| 0 | 50.0% | Pick'em |
| 25 | 53.6% | -1.0 |
| 50 | 57.1% | -2.0 |
| 100 | 64.0% | -4.0 |
| 150 | 70.3% | -6.0 |
| 200 | 76.0% | -8.0 |
| 300 | 84.9% | -12.0 |
Step three: divide by 25
The spread is the adjusted gap divided by 25. That divisor is not arbitrary — it was fitted so that the model's spreads line up with actual scoring margins across a full season. A 150-point edge becomes a 6-point favourite, which the curve independently prices at 70.3%.
Using it to spot a disagreement
Because the chain is fixed, any gap between the model's line and the market's line reduces to one question: which input do you trust less? If the model has a team at -6 and the book has them at -3, the model is claiming roughly 75 rating points of edge that the market does not see. That is a specific, checkable claim, and usually it comes down to information the model does not have — a signing, an injury, a coaching change — rather than a mispricing.
A model that cannot tell you why it disagrees is not giving you an edge. It is giving you a number.
That is what the factor lab on each matchup page is for. Drag the rating, the home field constant or the rest edge and watch both the probability and the spread respond. If moving one slider a plausible distance closes the gap with the market, the disagreement was never very deep.