How RankSlate works
Two components, blended. A results value that rewards wins by opponent quality and site, with road wins worth more than home wins and bad home losses punished hardest. And an adjusted efficiency: scoring margin per possession, adjusted for opponent strength and site. Both lean on opponent strength, so the rating is solved iteratively: rate the teams, feed those ratings back in as strength of schedule, and repeat until it settles. Same games in, same ranking out, every time.
The number. The rating you see is a z-score: how many standard deviations above an average Division I team a side rates, with both halves of the blend put on that common scale before they're mixed. Zero is an average team, 1.0 is a good one, and the best teams in the country sit around 2. Every team in a sport is scored in one pool, so an FCS number and an FBS number mean the same thing. The rank is just that number sorted, and the gaps matter as much as the order: two teams a tenth of a point apart can swap on one result, two teams half a point apart cannot. The bar beside each rating is drawn from the 25th-ranked team up to the best, so a cushion, or the lack of one, is visible at a glance.
The cold start. A results rating means nothing in week one, so each season opens on a prior: last season's final RankSlate, regressed a third toward the average to respect roster turnover, mixed half and half with the national preseason polls for the teams they cover, because poll voters watch rosters all offseason. The prior starts as the whole number, and every real result replaces a piece of it, until it's gone completely: by a team's 25th game in basketball, and by the 12th in football. From there the number is pure results.
Tuned to the committee, not to us. For men's basketball, the selection committee sorts teams with its NET rating, whose exact weights are not public. RankSlate mirrors the NET's published methodology and is re-fit each season against the most recent final NET, because the committee re-tunes its own formula yearly.
Football is tuned differently, on purpose. The playoff committee publishes rankings but no formula, so there is nothing to reproduce. RankSlate football is the same engine on football's terms. Games are scored against expectation: a win counts for the chance you had of losing it, and a loss costs the chance you had of winning, so losing to a better team costs little and losing to a worse one costs a lot. A game can only land 17 points from what the rating expected, in either direction, because a 50-point rout says little more than a 17-point cover, and that cap no longer punishes a strong team for a weak schedule. FBS and FCS are rated in one pool, so games between them count.
Football's preseason start listens to the polls, once. In today's football, more than half of a team's production walks out the door every year through the transfer portal. Last season's results can't see who left. The preseason polls can, because the voters watch rosters all offseason. So football's starting point is a 50/50 mix: half last season's rating, half the preseason polls.
The national polls only cover the top 40 or so teams. For everyone else, each conference runs its own preseason poll, and we use it to give a small nudge to the order inside that league.
We tested all of this against the real 2025 season before shipping it. The poll mix called early games better than last season's rating alone, and the conference nudge raised our win rate on early games between teams the national polls skip from 59 to 63 percent. We also tried measuring roster turnover straight from player stats. It added nothing the polls didn't already know.
And the polls don't overstay their welcome. At kickoff the starting point is the whole number. Every real game replaces a piece of it: after four games it's about half the number, after eight about a quarter, and by a team's 12th game it's gone. From then on, only results count. We tested that schedule week by week against 2025 too, and it gave up nothing at any point in the season.
Women's basketball runs the same engine. It uses team box scores from every game (the possession-based efficiency path, like the men's), with the preseason prior and poll fade working the same way. The women's committee sorts with the women's NET, but that number is not public data we can pull, so the sanity check is the final AP Top 25 instead, the same reference football uses.
What it is not. RankSlate is not the NET and never claims to be; it is our own rating, built so you can see the machinery. It is also not a poll and not a prediction market: nothing here is voted on and nothing is placed by hand. The only human input is the preseason polls each sport opens on, and it is disclosed above and faded out by midseason.
Where it goes next. The same engine runs on any sport with results and schedules. Baseball is next, tuned against the selection committee's RPI, with its own backtest before it ships.
See the rating drive a bracket at bracketprojections.com, where the projection is audited against the committee's published bracketing principles every night.