Coverage

Coverage measures how often a player's actual result lands inside NovaPredict's floor-to-ceiling band, compared against an 80% target. Because the floor is the 10th percentile outcome and the ceiling is the 90th, results should land inside the band about 80% of the time if the bands are honest.

This is the plainest test a range-based projection can face. Anyone can draw a floor and a ceiling; coverage checks whether those lines mean what they claim. NovaPredict locks every band before kickoff and then counts, in public, how often reality stayed inside.

How to Read It

The target is 80%, and both directions of error matter.

• Coverage near 80%: the bands are calibrated. When a player page says his floor is 8 points, that floor is trustworthy enough to plan around.
• Coverage well above 80%: the bands are too wide. They will almost never be "wrong," but they are also too vague to be useful, like a forecast that says tomorrow will be between freezing and sweltering.
• Coverage well below 80%: the bands are too narrow. They project more certainty than the model has earned, and you should treat the published floors and ceilings with more skepticism.

For a fantasy player, healthy coverage is what makes band-based decisions rational: benching a low-floor player in a week you are favored only makes sense if the floors have been meaning something.

What It Is Not

Coverage is not an accuracy percentage and not a win rate. A result landing inside a wide band is not a precise prediction coming true, which is exactly why the 80% target matters: it punishes bands that are too wide just as clearly as bands that are too narrow. It also says nothing about where inside the band results fall; that finer question is graded by CRPS.

For the distribution-wide grade, see CRPS.

Coverage is tracked live, week by week, at /accountability.