A player projection is a single number — our best estimate of the most likely outcome, not a prediction of exactly what will happen. Football is high-variance: the same player, in the same matchup, can catch four passes for 30 yards or eight for 120, and both are ordinary. So the useful question is not "was the projection right?" on any single night — it's "how close does it land, on average, over many games?"
How accuracy is measured
The plain-language version is average miss: across many graded games, how far off was the projection, on average? If a receiving-yards projection is off by about 18 yards on a typical night, that is its average miss. Lower is better, but there is a floor no model can beat, because the outcome itself is noisy. (The technical name for average miss is mean absolute error, or MAE.)
A few things follow from that:
- Average miss is a population number, not a promise about one game. "Off by about 18 yards on average" describes how a whole set of projections behaved over many nights. It is not a range around any single projection, and it does not say tonight's number will land within 18 yards — one game can miss by far more, or far less. It tells you how much to trust the projections as a group.
- We publish one number, not a band. The projection is a single figure. We do not dress it up with a high/low range, because a range we have not measured and calibrated would be false precision — the exact overconfidence the average-miss number exists to puncture.
- More stable roles project better. A bell-cow running back who gets 20 carries a week is far more predictable than one in a committee, so his projections tend to carry a smaller average miss. Accuracy is mostly a story about how steady a player's role is, not how clever the model is.
- Accuracy is only believable if it is logged before the game. A projection written down after kickoff proves nothing. The honest test is a projection recorded before the game starts and graded after, with nothing edited, hidden, or reworded once the result is in.
What accuracy does not tell you
A more accurate projection is not automatically a better bet. The betting line already reflects most of what is knowable about a player, so being close to the outcome and being ahead of the market are different things. Treat accuracy as a measure of how much to trust the number — not as a signal to bet.
Common questions
What is a good average miss for a projection?
It depends entirely on the stat and the role. A steady, high-volume player will have a much smaller average miss than a boom-or-bust one, so the only fair comparison is a projection against its own history in the same stat — never one stat against another.
Does a more accurate projection mean a better bet?
No. Accuracy measures how close a projection lands to the real outcome; an edge means beating the price the market offers. The line already prices most public information, so a projection can be accurate and still offer no betting advantage. They are separate questions.
Do you show a range or just a single number?
A single number — the projection is our best estimate of the most likely outcome. Its honest companion is not a range but the average miss: how far our projections land from the result, on average, across many graded games. That is a population-level accuracy figure, not a high/low band around any one player, and we do not publish a band we have not measured.