hussEyquation
Explore rankings

Public test receipt · current-ability-game-log-v1-2027

How the live Box Forecast test will be scored

These rules were frozen before the 2026–27 season. Weekly predictions will be saved, never overwritten, and scored after each player's next 20 and 40 regular-season appearances.

Download the machine-readable protocol →
StatusRegistered

No live snapshot has been graded.

CadenceWeekly

Every edition gets an as-of date.

Horizons20 + 40

Player appearances, not team games.

RetuningNot allowed

The live candidate stays fixed.

The target

What gets predicted

The model predicts context-neutral box production and average minutes for each player's next 20 and 40 appearances. Scoring starts with his first appearance after the snapshot date.

A player needs at least 10 prior games to enter a snapshot. No player-specific overrides are allowed, and lineup evidence is excluded from this candidate.

The gates

What counts as a pass

250scored predictions minimum
3scored weekly snapshots minimum
5%lower error than season-to-date
2%lower error than trailing 20
0.50maximum absolute bias
Both20- and 40-appearance gates required

Rank agreement may trail the season-to-date baseline by no more than 0.01 Spearman. At least 20 players must be ranked in each scored snapshot.

The limit

What a pass would prove

A pass would support this specific claim: for the registered 2026–27 cohorts, the frozen Box Forecast predicted future box production better than the two named baselines under these scoring rules.

It would not prove that Box Forecast measures total basketball impact, predicts injuries, or should replace Impact Rate. Historical results shown elsewhere remain exploratory until the live cohorts mature.

Return to the forecast ranking →