A guide to the percentages

Favorability is the model’s estimate of how often each team would win this matchup under the current assumptions. A 60% team is favored, but the 40% underdog still has a very real path to winning.

These are not betting odds or a prediction of the final score. Week 1 estimates are intentionally pulled closer together, and division matchups stay more competitive because those games are often unusually volatile.

Power rankings answer a simple question: how strong does the model think each team is right now? Playoff projections answer a different one: how likely is that team to survive its actual schedule and conference race?

A highly rated team can face a difficult path, while a slightly weaker team can benefit from its schedule or division. That is why ranking position, projected record, and playoff chance will not always line up.

Completed games replace predictions with actual results. One shared rating engine updates both rankings and future game probabilities from completed results, opponent strength, and scoring margin. Only results observed before a forecast are used. Preseason EPA and scoring inputs remain the starting prior; they are not refreshed play by play.

Early forecasts are saved within 30 hours of kickoff. A separate final forecast is saved in the last hour, when the scheduled sync runs. The scorecard uses that final forecast when available and preserves the early forecast. The previous model also runs in parallel for a fair comparison on the same games.

Pretzel Quest does not copy betting markets, expert power rankings, or fantasy projections. It also avoids grading every draft pick or routine roster move, where confident-looking adjustments can quickly become guesswork.

The model cannot fully capture locker-room chemistry, scheme fit, weather surprises, or every injury. Rare manual adjustments are reserved for clearly important situations and are labeled on the Picks page.