Completion percentage is the oldest accuracy stat in football, and one of the most misleading. A quarterback who throws a lot of screens and checkdowns can complete 70% of his passes without being especially accurate. One who attacks downfield can be more accurate and still finish in the low 60s.
Expected completion
The fix is a standard idea in football analytics: estimate how likely each throw was to be completed, given how hard it was, then compare the result with the estimate. A 3-yard flat route might be caught 80% of the time by an average passer. A 35-yard throw down the sideline might be caught 35% of the time.
CPOE, completion percentage over expected, is a passer's completion percentage minus his expected completion percentage on the same throws. +5 means he completed five more passes per hundred than an average passer would have, on throws of the same difficulty.
How ACE works
ACE (Accuracy Compared to Expectation) is QuarterbackStats.com's rating built on that idea. For every pass attempt we estimate the completion chance from how far the ball travelled in the air, which third of the field it went to, the down, the distance, field position, whether the quarterback was hit as he threw, whether the game was indoors, and cold or wind. The model is fit on every NFL attempt over the last five seasons, so each era is judged against itself.
ACE is completions minus those expected completions, added up over the season:
- Complete a throw with a 35% chance and you earn +0.65.
- Miss a throw with an 80% chance and you lose −0.80.
We show ACE as a season total (it rewards volume, the way a counting stat does) and as ACE per 100 attempts, which compares a 600-attempt starter with a 300-attempt backup. ACE per 100 is the same scale as CPOE.
Does it agree with other models?
Closely. For the top 20 quarterbacks in 2025, our ACE per 100 and nflverse's own CPOE agree with a correlation of 0.92, and the average gap is under a third of a point. Drake Maye led both; Next Gen Stats, which uses tracking data we do not have, put him at +9.1, and our model has him at +9.3.
What it cannot see
The public play-by-play does not mark throwaways or drops. A deliberate throwaway counts as a missed throw, and a drop counts against the passer. Those even out over a season more than they do in one game, which is why we would not judge a quarterback's accuracy from a single week. We also don't know where each receiver was standing or how tightly he was covered, so a throw into a tight window and a throw to an open man at the same depth look the same.
Read ACE with EPA per dropback: EPA says how much a quarterback helped his team score, and ACE says how much of that came from putting the ball where it could be caught.