TasteFit

Prediction ledger

Each batch is a SHA-256 fingerprint of the concept, the area, the shortlist verdicts and the UTC time. The file is committed in the repository. A live memo shows its own fingerprint and does not write this page. No Qloo payload and no personal data are stored here.

How the check is scored

Brier score (Brier 1950): the mean of (p - o) squared over the shortlisted concepts. p is not a Qloo fit score and is not stored in this file. When every verdict is GO, ADJUST or NO-GO, p is the fixed map 0.66, 0.50 or 0.33. Any other verdict, including unscored, leaves p undefined, and the Brier score is not defined for that batch. The map is not a calibrated probability of a new licence.

On 2026-11-17, o is 1 when the City of Chicago business-licence file shows a new food licence whose description matches the keyword frozen in this batch, with a start date on or after the registration timestamp and before 2026-11-17, and an address within 1.2 km of the registered point. Otherwise o is 0. The keyword and the point are not revised after publication.

Skill is the Brier score of one constant, the citywide rate of new food licences over the same window, minus TasteFit's Brier score. The constant is computed on the check date from the same public file and is not invented here. The 2023-2024 Chicago backtest found no measurable gain, so a negative skill score is a possible result and will be reported.