Why this exists
About Sporacle
An NFL decision service and a calibration research project.
How the numbers are made
The game model predicts each team's score as a distribution. Win likelihood, projected score, margin, total, and their published ranges come from that distribution.
The player model separates expected usage from per opportunity production, then publishes fantasy points in each supported scoring format with p20 and p80 uncertainty. Pages display those published cells without recalculating them in the browser.
Honesty commitments
- Published predictions remain visible after kickoff.
- Corrections are dated annotations beside the original call.
- Misses use the same presentation as hits on the track record.
- Public numbers come from readable, pinned pipeline artifacts.
- Likelihoods always carry the calibration research disclaimer.
Sources and attribution
Public inputs are attributed on the data and methodology page. In each position's rankings table, a row is marked imputed when one of its figures was filled in, or marked mixed when its floor, projection and ceiling disagree without any of them being imputed; every other row shares one source across all three. Full source detail for every published figure is on that player's own page.
This published build
- Run
- 2026-03-reprice-b65256aa
- Season and week
- 2026, week 3
- Generated
Source: the pinned published run.json artifact.