← Back to Matchday Terminal

Q&A & methodology

Ratings · Forecasts · Interpretation

The model

How to read the numbers

A high-level guide to the numbers shown on Matchday Terminal. Exact formulas, weights and tuning settings remain private. The Learn tab works through each idea with the site's current numbers.

What goes into a matchup forecast?

Team performance, opponent strength and game context inform the matchup estimate. Where available, market information also informs the published forecast. Model-only and market probabilities are identified separately where shown.

What is a power rating?

A power rating is how many points better a team is than an average opponent on a neutral field. Every result is adjusted for the opposition faced, rather than ranking teams by wins alone. Offense, defense and schedule strength provide additional context where available. A rating describes strength; a résumé describes what a team has earned.

How does the model rate teams before they've played?

Preseason estimates use prior performance and available roster information. As current-season evidence grows, the model updates its view. Football and basketball use sport-specific approaches.

What does a win probability mean?

It is an estimated chance of winning, not a certainty or an accuracy score. A team given a 60% chance can still lose: across many comparable games, that estimate implies roughly four losses in ten. Whether the estimates match actual results must be checked against the graded sample, with its dates and size.

How is the projected score made?

The model estimates an expected scoring margin and total. In football, thousands of games are simulated around that estimate and the typical final is shown, rounded to scores teams actually finish on; in basketball, the margin and total are split into two scores. It is a typical outcome, not a call on the exact score.

Why does the model sometimes disagree with the market?

The model and the market use different information and can respond at different times. The displayed gap compares the model's probability with the market-implied probability for the same outcome. A larger gap does not establish an advantage: check the disagreement record and its sample rather than assuming the market is wrong.

Reading the record

What the results tell you

How should I read the performance record?

Read the sample size, time period and caveats alongside the hit rate. The record measures who wins; it does not measure returns or spread performance. Results on games where the model disagrees with the market help distinguish its contribution from simply backing the favorite.

What are the Upsets of the week?

A weekly selection of market underdogs the model considers more competitive than their prices suggest. The list can be short or empty when there are no qualifying games. An underdog watch is not a guarantee or an automatic official pick. Judge the results against the market's expected win rate, not against the expectation that most underdogs should win.

Rankings

Why the Top 25s differ

What is the difference between the Top 25s?

The power rating Top 25 orders teams by the model's estimate of strength. The AP Top 25 is the national media poll. The Personal Ballot is one person's weekly ranking, not the model's, and ranks teams on their résumés: what they have earned through their results. They answer different questions, so their orders can differ.

© 2026 Matchday Terminal · Legal & data · Roadmap & limits