Closure-pressure model card
Choose Detailed Stats to compare and sort all records in this report.
The closure-pressure model turns historical public-record features into probabilistic research leads. A model forecast is not an official closure determination.
License delinquency and ownership changes are separate signals. Neither alone proves that a restaurant has closed.
Evidence taxonomy
| Official record | A direct DBPR or Sunbiz status, inspection, filing, or action. |
| RiskyEats calculation | A deterministic calculation from cited official records, with scope, period, numerator, and denominator. |
| Model forecast | A probabilistic output with an as-of date, model version, uncertainty, and known limitations. |
| Editorial finding | A reviewed interpretation supported by linked evidence and subject to the correction policy. |
Metric dictionary
| Emergency-order recommendation | inspection disposition count | stated report window | An official DBPR inspection outcome; it is not the same as a permanent closure. |
| DBPR 45/46 status | license count or share | as-of date shown | A delinquent or inactive regulatory status, not automatic proof that the storefront closed. |
| Inspection-pattern anomaly | peer-relative analytical band | stated observation window | A record outside its comparison range and a lead for review, not a violation finding. |
| Closure-pressure forecast | calibrated probability | 90-day model horizon | A model estimate, not an official status or guaranteed outcome. |
Closure-pressure model card
| Objective | Prioritize aggregate closure-pressure research and reviewed case leads. |
| Excluded use | Do not treat a forecast as an official closure, food-safety determination, credit decision, or unreviewed accusation. |
| Outcome horizon | 90 days. |
| Training window | 2014-01-01 through 2024-12-31 snapshots by default; 2025 is reserved for drift review, not nightly fitting |
| Holdout strategy | stratified random holdout within training cohort |
| Training rows | 15,444 training rows; 1,188 positive (7.7% base rate); 3,861 holdout (297 positive) |
| AUC | 0.704871385006065 |
| Brier score | 0.06656307095328352 |
| Calibration error | 0.016698468910737455 |
| Top-decile precision | 28.7% |
| False-positive rate at 0.50 | 0.0% |
| Missing data | No-inspection and unresolved-identity records are excluded from named publication. |
| Publication status | Named-business forecasts withheld pending completed governance review. |
| Corrections | Email corrections@riskyeats.org with the license number and the discrepancy; see Contact & corrections on /disclosure.html. |
Data sources: ORSM. Built from DBPR public inspection records, Florida Sunbiz business filings, and Foursquare/OSS location data. All counts are derived from public records — no estimates or projections unless labeled as forecast.