Football Rating & Prediction Method

OSR Power Rating-style team strength

The frozen research architecture combines recursive opponent and network strength (80%), nonlinear result/margin information (10%), and opponent-adjusted offense/defense (10%). Components are standardized within the population. Recency, independent-opponent evidence, schedule connectivity and reliability limit the influence of sparse records.

Margins use sport-specific units: soccer goals capped at 7; football points divided by 7 and capped at 6 units; volleyball sets divided by the observed winning total, capped at 1. These are not shared soccer margin constants.

A prior uses only the same program and playing-format population in the immediately preceding season: half its reliability-adjusted final strength. Current strength uses n/(n+3) times reliability-adjusted current power plus 3/(n+3) times the prior. New programs have a zero prior. All same-day results are excluded.

Frozen probabilities

Let x = 4.07572108 × (home strength − away strength) + 0.19046754 for non-neutral games. Neutral games omit the home term. Home and away weights are exp(x/2) and exp(−x/2). The draw probability is zero. Normalize these weights to sum to one.

Calibration is frozen from 10,022 chronological training games through 2025-11-29. No 2026 result tunes it. The standalone model was selected over the tested hybrid; the hybrid is not deployed.

Soccer and 11-player football retain the existing simple conditional score estimator: an observed earlier same-format score near the conditional median, requiring at least ten examples and agreement with the most likely outcome. It is less matchup-specific than the outcome probability. Volleyball remains outcome-only when future match format is not verified.

Power Rating presentation and history

Displayed Power = 100 + 15 × effective strength. This is presentation only; probability calibration uses the unscaled strength. At least three current-season games are required to rank. A similar numeric display does not make different sports comparable. Boys Soccer mathematics and its Combined V2 forecast remain unchanged.

Earlier snapshots and genuine pregame predictions retain their original model versions and scales. Historical charts separate model eras. Current predictions are saved before game locking; completed predictions are not rewritten.

Smaller football formats

Verified 9-player and 8-player teams now have rating-only rankings from the same opponent-network architecture, calculated separately for each format. Verified 6-player teams use the simpler independently studied football rating reference. Their displayed Power values use a 100 + 15 × strength presentation, but they are not comparable with 11-player Power or one another. At least three same-format games are required.

No calibrated game probabilities or score forecasts are published for these three formats. Seasons with unverified playing-format evidence remain unranked rather than borrowing another format’s results.