What the numbers actually claim. Graded on a season the model never saw: the projections beat last year's totals by about 11% on average miss, and the low-to-high ranges covered 83.3% of real outcomes against a promised 80%. The second number is the one that matters: it means the ranges are honest, not decorative. Availability is the largest remaining error and is published as a range rather than pretended away. Why “about”. Run the same grading again and that figure lands anywhere between 10.6% and 11.2%, because the model draws a random subset of the data for every tree. Nothing else changes. Quoting one decimal place would claim a precision it has not earned, so we hold our own headline to the standard we hold every projection to.

Measured on the standard model, not the MoneyPuck-free one these pages ship. The two differ by about 0.03 points of average miss. When the MoneyPuck-free variant is graded, these numbers are replaced by its own.