{
  "id": "social-lsoa-exposure-2026-10-10-v1",
  "geography": "England LSOA21; a new scale, separate from authority coefficients",
  "status": "Engineering protocol; no LSOA outcome model fitted",
  "crs": "EPSG:27700 metres; full-resolution compatible boundaries required for allocation. Display centroids/generalised maps cannot allocate road mileage.",
  "tolerances": {
    "motorway|urban": {
      "maximumAbsoluteBias": 0.1,
      "maximumWeightedAbsoluteError": 0.2,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.05
    },
    "motorway|rural": {
      "maximumAbsoluteBias": 0.1,
      "maximumWeightedAbsoluteError": 0.2,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.05
    },
    "A|urban": {
      "maximumAbsoluteBias": 0.1,
      "maximumWeightedAbsoluteError": 0.2,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.05
    },
    "A|rural": {
      "maximumAbsoluteBias": 0.1,
      "maximumWeightedAbsoluteError": 0.2,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.05
    },
    "B|urban": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    },
    "B|rural": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    },
    "C|urban": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    },
    "C|rural": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    },
    "unclassified|urban": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    },
    "unclassified|rural": {
      "maximumAbsoluteBias": 0.15,
      "maximumWeightedAbsoluteError": 0.3,
      "minimumIndependentTrafficObservations": 30,
      "minimumHeldOutRegions": 3,
      "minimum95IntervalCoverage": 0.9,
      "maximumAuthorityReconciliationDifference": 0.1
    }
  },
  "thresholdRationale": "Prospective research acceptance tolerances, not DfT standards. Stratum-specific transport checks are frozen before LSOA outcome analysis. No outcome-informed relaxation. Domain review can propose a new version before outcome access.",
  "networkGate": "Every intended physical segment/year must appear exactly once and be fully assigned, with known measured/estimated AADT and uncertainty bounds. Missing and zero traffic are different. No arbitrary aggregate coverage threshold.",
  "mileage": "Two-way all-motor AADT \u00d7 segment length in metres / 1609.344 \u00d7 actual calendar days. Input adapter must consolidate directional counts; no extra factor of two.",
  "collisionGate": "One event ID/year, unique polygon assignment. Record border/near-border, positional and unmatched events. Ambiguous events are unassigned and block release. A major-road study additionally needs a validated unique event-to-covered-link match.",
  "trafficModelPredictorsAllowed": [
    "count_history",
    "junction_density",
    "lanes",
    "network_connectivity",
    "road_class",
    "road_function",
    "road_length",
    "speed_limit",
    "vehicle_mix"
  ],
  "prohibited": "Social features, deciles, deprivation, education, crime and collision outcomes must not enter the primary mileage model.",
  "independence": "Validation sites and entire holdout regions must be disjoint from traffic-model training observations and regions; evaluate weighted absolute error, aggregate bias and interval coverage by road stratum and year.",
  "reconciliation": "Compare original segment sums to official authority \u00d7 year \u00d7 road-class \u00d7 rurality totals. Optional documented stratum factor preserves original estimates and intervals. Agreement after scaling is not independent validation.",
  "exposureUncertainty": "Carry lower/upper AADT into mileage intervals. Require >=95% of reconciled mileage within relative half-width 25%; then repeat separate LSOA collision models across exposure scenarios. No predictive release until that sensitivity and Phases 3\u20135 pass.",
  "toleranceMetres": {
    "allocationAbsolute": 0.01,
    "nearBorder": 10
  },
  "humanReviewRecorded": false,
  "registeredAt": "2026-10-10T17:47:34.265711+00:00"
}