How OVR Works
Preseason team & player ratings, explained
Team Ratings
Every team's preseason OVR blends four standardized inputs into a single 2026 team-strength number, weighted:
- 85%: 2026 SP+, as of Sep 9, 2026. CFBD's own current-season power rating, frozen to a dated snapshot taken after Week 1 and refreshed deliberately, not nightly, so one day's poll swing can't quietly re-rank every team. Real 2026 results are meant to be folded back in by our own in-season update once there is enough of a season to trust. Results are being counted as of week 12 (590 games in). This is the team's actual current form, not last year's.
- 7%: overall roster talent score. A composite of the roster's recruiting pedigree across all classes, not just this year's.
- 6%: 2026 recruiting class strength. How this year's incoming class grades out nationally.
- 2%: returning production percent. How much of last year's snaps and production is walking back through the door. Deliberately the lightest input, since on its own it can't tell a team that's rebuilding apart from one whose stars simply graduated to the NFL.
These blend into one 2026 team rating (Overall, Offense, and Defense). SP+ now carries most of the weight because it already accounts for talent, recruiting, and returning production on its own, so the other three inputs are a light touch, not the main driver. If CFBD's SP+ data ever stops updating properly, the model falls back to last season's final SP+ blended more heavily with the other three inputs, so a rating is never built on broken data.
Real results, folded in as the season goes. Once there is enough of a season played for the numbers to be trustworthy, real 2026 scoring margins get blended into the rating too, leaning heavily on the preseason view in the first couple of live weeks and trusting results more as a team plays more games. A team's preseason projection always keeps a small share of its published number, so no team is ever ranked purely on a handful of games — a fast start is rewarded, but it does not erase the whole picture, and a bad one does not either. Measured against two full past seasons, this update needs a handful of games before it clearly beats not updating at all. The earliest live weeks are a real, disclosed tradeoff (a touch noisier than the plain preseason number), accepted so one early result can't swing a rating harder than the evidence supports. A convincing win moves a team's rating meaningfully; a one-score escape barely moves it. The model rewards how a team has actually won, not just that it won.
Rank vs. Rating — Why They Can Differ
The OVR / Offense / Defense numbers measure raw team strength in points — how good a team's offense and defense actually are. Offense and Defense, which every spread, win probability, and player rating is built from, are never touched by results-based credit (below). The Overall rating does include this year's results, and they can move it a lot.
The rank — the # you see in the Top 25 list — mostly follows that Overall rating (one exception, head-to-head, is below). The results part, called strength of record, asks one question: how would a top-5 team have done against this exact schedule?
- Wins — every win earns credit for how hard it would have been for a top-5 team. Beating a strong team, especially on the road, earns a lot. Beating a weak team earns very little, because a top-5 team would almost always win that game too.
- Losses — every loss costs real points. The weaker the team you lost to, the more it costs: losing to a team a top-5 side would almost always beat costs close to a full win.
- How much — once results count fully, each win more than a top-5 team would be expected to have adds about 4 points to the Overall, and each win fewer takes about 4 away.
- Every game counts — games against FCS teams count too, as an opponent weaker than any FBS team. Only games already played count, so a tough schedule still to come earns nothing yet.
- No unbeaten bonus — a clean record is rewarded game by game, with no extra bonus on top.
- Unbeaten teams come first — an unbeaten SEC, Big Ten, Big 12 or ACC team with 3 or more wins is ranked above any team with 2 or more losses (head-to-head aside), unless that team is far stronger on paper: 15 or more points higher in combined offense and defense, a gap where the stronger team won more than 80% of the time in our tests.
- Head-to-head — if two closely-rated teams already played this season, the team that won the game never ranks below the team it beat.
The results part starts at zero (a brand-new season ranks exactly by the raw rating) and grows with each game a team plays, counting fully from its 6th game on. It can move a team a long way: a team with several losses can drop well down the list even with a strong raw rating. It never changes the Offense and Defense numbers that predictions are built from.
During the season the very top is compressed: above 90, each extra point is harder to reach, and only one team at a time can read 99.
Player Ratings
A player's OVR starts by anchoring to their team's offense or defense rating, then moves up or down from there:
- Production percentile — for players with meaningful stats (yards, tackles, sacks, etc.), how they graded out relative to peers at their position.
- Recruiting percentile — used instead of production when there isn't enough of a track record to grade — true freshmen, or positions like offensive line where individual stats don't really exist.
- Depth-chart penalty — backups are docked relative to the starter ahead of them.
Limitations — Being Honest
- Offensive linemen don't have an individual stat to grade — their ratings lean almost entirely on recruiting pedigree.
- True freshmen and transfers are projections, not track records. A transfer's rating leans on how they produced at their old school, which doesn't always translate directly.
- Preseason numbers are inherently uncertain — every team and player here is a projection until games are actually played. Treat this as a well-reasoned starting point, not a guarantee.
We also track whether these predictions actually hold up — see the calibration record for the honest answer to "when we say 70%, does it happen 70% of the time?"