Better than the simple alternatives?
The share of winners picked right in boys games, compared with easy rules anyone could use. Last season’s ratings alone is the fairest comparison: it shows what updating ratings every week adds.
| Method | Winners right |
|---|
Do our percentages mean what they say?
We group games by the win chance we gave the favorite. If the percentages are honest, the favorite should win about that often. Toss-ups are the weak spot: when we say 55%, the real number is closer to a coin flip.
Close games vs lopsided ones
Grouped by how big a margin we projected. Close games are close to coin flips. The average miss stays about the same no matter how big the projected margin is.
| Projected margin | Games | Winners right | Avg miss |
|---|
Week by week
The model should hold up as the season goes on. Late weeks have fewer games, so they swing more.
| Week of | Games | Winners right | Avg miss |
|---|
Season win totals
From February 9 we projected every team's final regular-season win total by adding up the win chance of each game left on its schedule. This compares those projections with what happened. Only a handful of games were left, so this is an easy test. It gets harder, and more useful, early in a season.
| Method | Average miss (wins) |
|---|
Season-long projections and the 90% range
The harder test: project every team's final win total on December 1 using only last season's ratings, then compare with what happened. This is the situation the site is in right now.
| Full-season projection from the start | Result |
|---|
The first weeks of a season
Until teams play, every rating comes from last season. To see how much weaker that makes the picks, we ran the same test on the first three weeks of 2025-26 using only 2024-25 ratings, with no 2025-26 games at all. This is what the opening weeks of 2026-27 look like.
| First three weeks of the season | Winners right | Avg spread miss |
|---|
How to read this
Bias is the average of real margin minus projected margin, from the home team’s side. Calibration is whether a 70% chance really wins about 70% of the time.
Limits: the sample is one season, and it counts all varsity games with a result, tournaments included. Ratings were refit each week using only earlier games, but the settings that turn ratings into percentages were fit on January 1 to February 8, before the games shown. Boys and girls numbers should not be compared directly, because girls games have bigger gaps between teams, which makes winners easier to pick.