How accurate is the rating?

The test. For each past season and each week, the rating was fitted on that season's games up to that week, plus what it had learned from earlier seasons, never later ones, and asked to pick every game the following week. Picks start in week 3 (two games of evidence per team; regular-season games only, so the count is lower than the number of games played). The rating's settings and its win-chance uncertainty setting were chosen on 2023 and 2024, so 2025 is the one season the test never saw while tuning. The chances here use the same uncertainty setting as the game pages. "Crabtree pick" is the official seeding number used the same way: take the team with the higher Crabtree Index. Generated 2026-10-06T14:22 by version 0.8.24, model v2.

11-man

Season	Games picked	niftisports correct	Crabtree pick correct	Avg. margin error
2023	164	76.8%	73.5%	11.7 pts
2024	151	70.2%	67.1%	12.5 pts
2025	177	86.4%	78.0%	10.7 pts
All	492	78.3%	73.3%	11.6 pts

By week, 2025: week 3 77.3% · week 4 88.5% · week 5 92.3% · week 6 84.6% · week 7 92.0% · week 8 88.0% · week 9 69.2% · week 10 92.9%. Early weeks are hardest: the carryover from last season is what keeps week 3 respectable.

Calibration

When the rating gives the picked team a 70–79% chance, how often does that team actually win? Good calibration means the two columns match.

Stated chance	Games	Picked team won
50–59%	92	54%
60–69%	74	72%
70–79%	98	69%
80–89%	93	88%
90–99%	135	98%

8-man

Season	Games picked	niftisports correct	Crabtree pick correct	Avg. margin error
2023	80	77.5%	71.2%	12.1 pts
2024	62	82.3%	83.6%	12.1 pts
2025	59	79.7%	77.6%	11.8 pts
All	201	79.6%	76.9%	12.0 pts

What the numbers mean, and what they don't

Picking the higher Crabtree team is a fair yardstick because it is the number the playoffs run on. The rating beats it by 4–8 points a season, mostly by using scores (bent so blowouts don't pile on), opponents' opponents, class, and last season. Margin error of 11–13 points is typical for high school football anywhere; a team picked by 7 points losing is normal, not a failure. Parameters were chosen on 2023 and 2024 and then checked on 2025 untouched, so the 2025 row is the honest one. Past-season games have no home/away in our sources, so those tests ran with no home-field term; this season has one. This page regenerates whenever the model or the data changes. Playoffs: over the 53 playoff games of 2023–2025 where one team had the higher Crabtree index, the higher-rated team won 74% and the higher Crabtree seed 77% (they agreed in 45 of the 53): the higher seed also gets home field, which our neutral-site history can't see, so we make no claim to out-seed the MPA. This season: every pick is frozen on the Thursday its week begins (Week 3: the Tuesday evening the page went live) and graded on the Picks page. Nothing here is advice.