{"version":"ew-v4","status":"Ready for national outcome/exposure assembly; model fit and insurer validation not performed","phase3":{"key":["lsoa","year"],"geography":"Join to all 35,672 master areas before applying documented model eligibility. Crosswalk counts and denominators, never average deciles.","outcome":"Distinct police-reported injury collisions with at least one in-scope motor vehicle, with log(all-motor vehicle miles / 100000) offset. Vehicle and collision records must not be counted interchangeably.","exposure":"Matched in-scope motor-vehicle miles, period, source, counted/estimated status, uncertainty and allocation features. Keep zero-collision positive-mile rows; zero/missing miles do not imply zero risk.","rate":"100000 × sum(distinct in-scope injury collisions) / sum(matched motor-vehicle miles); never average area rates.","offset":"log(motor_vehicle_miles / 100000)","zeroOrMissingExposure":"Retain master row, flag model-ineligible, publish counts and excluded collisions/miles by country, region and rurality.","sourceTime":"All current features are retrospective mixed-vintage context. A historical forecast needs verified source editions AND bands available at its origin."},"phases4to6":{"actions":["Pre-register baseline, feature families, continuous/smooth/decile alternatives and multiplicity policy before fitting.","Compare models on the same rows and offsets; report missingness sensitivity.","Audit which features enter exposure allocation; test independently measured traffic subsets and alternative mileage allocations.","Report adjusted effects, uncertainty, annual stability and geographic transfer. Do not select using final test outcomes."],"priorOutcomeAccess":"Existing social, parking and frost studies inspected 2023–2025 results. No 2025 claim of untouched confirmation is permitted.","reserve":"No fresh outcome reserve has been certified. Before prospective confirmation, audit prior access and register a still-uninspected period (candidate: final 2026), without altering this feature registry."},"phases7to10":{"actions":["Fit supported components jointly; freeze index definitions and weights on development data only.","Treat crime as a separate insurer-peril module; STATS19 injury collisions cannot validate theft claims.","For insurer trials specify collision location versus policyholder/garaging geography, vehicle mix, coverage/peril, claim count, earned miles or policy time, observation dates and source vintages.","Agree claims holdouts and metrics before access; retain calibration and out-of-time/geographic tests. No insurer performance claims before the trial."]},"claimsValidation":false,"predictiveValidation":false}