{
  "schemaVersion": "data-lab-evaluation/1",
  "released": "2026-10-21",
  "status": "Released evidence summaries; no new validation or refit",
  "records": [
    {
      "id": "london",
      "study": "London regional baseline and qualifications",
      "href": "/data-lab/london-qualifications#validation",
      "modelRelease": "london-qualifications-reproducible-v1",
      "population": "33 London authorities; 133 authorities supply the national anchor",
      "unit": "Authority × calendar year",
      "outcome": "Police-reported personal-injury collisions",
      "exposure": "DfT all-motor miles; actual assessment-year exposure",
      "training": "Expanding training from 2018; 2020–2021 outcomes excluded; training-only transforms and nested baseline selection",
      "evaluation": "2023–2025 temporal assessment, 99 rows; 2025 borough holdout, 33 rows; central-pair exclusion, 93 rows",
      "priorUse": "All assessment outcomes inspected before this release. V2 protocol amended after V1 results. None of these assessments is fresh confirmation.",
      "overlap": "Temporal training windows overlap. Borough holdout excludes held-out authority outcomes but retains its available lagged predictor history. Geographic dependence remains.",
      "selection": "Original seven baseline candidates; six social representations. No phase-two candidate selection or threshold tuning.",
      "metrics": [
        "Poisson deviance (lower)",
        "Deviance gain against named comparator (higher)",
        "Observed / expected (target 1)",
        "Exposure-weighted absolute rate error per 100,000 miles (lower)"
      ],
      "uncertainty": "399 paired bootstrap draws; 395 complete and 4 failed, unreplaced. Marginal intervals; no multiplicity or neighbouring-borough adjustment.",
      "sensitivities": "Five refitted mileage scenarios; ±20% uniform changes and total-preserving Inner/Outer reallocations. Diagnostic assumptions, not estimated exposure errors.",
      "decision": "Exploratory regional baseline candidate; no social block passes every original gate. Pricing approval absent.",
      "nextEvidence": "A prospectively frozen specification and genuinely uninspected period or external geography, followed by human actuarial validation.",
      "source": "/research/london-qualifications/v1/study-specification.json",
      "checks": "/research/london-qualifications/v1/validation.json",
      "referenceRows": 99,
      "independentCurrentVehicles": null,
      "sources": [
        {
          "path": "lib/london-flagship.json",
          "sha256": "dc35a3e1c1e04a9500b4b96bc272500f7c52f6d8cd6db344d3106fe5da97230f"
        },
        {
          "path": "/research/london-qualifications/v1/predictions.csv",
          "sha256": "d83a656f8c3d06e4b4227447714efcc31c6c85a579e7507b6f4178d329371cca"
        }
      ],
      "legacyId": "london-qualifications-assessment-v2"
    },
    {
      "id": "valuation",
      "study": "Original-price car valuation",
      "href": "/data-lab/car-valuation",
      "modelRelease": "price-anchor-shape-v1",
      "population": "UK petrol/diesel research vehicles; current tool accepts ages 1–15",
      "unit": "Historical listing record; public feedback report is not a verified unique vehicle",
      "outcome": "Historical asking price; feedback distinguishes asking, sale and professional value",
      "exposure": "Not applicable; age, mileage and original price are model inputs",
      "training": "66,081 historical petrol/diesel asking-price records; assumed 2020 observation cohort; no verified historical original-price anchors",
      "evaluation": "Historical family-transfer experiment: 2,761 older cars from 10 families; near-new asking prices proxy original prices. Current independent vehicle evaluation set: 0.",
      "priorUse": "Historical experiment previously inspected; no winning variant selected from that test. Public reports default to prior-use unknown.",
      "overlap": "Exact advert dates and verified unique-vehicle identity unavailable. Public submissions can duplicate or self-select; analytical deduplication does not establish independence.",
      "selection": "Grouped, pooled and fixed 15% annual comparator reported separately on the historical experiment. No automatic recalibration from feedback.",
      "metrics": [
        "Historical median absolute percentage error: grouped 10.11%, pooled 9.37%, fixed comparator 16.07%",
        "Feedback: median signed and absolute percentage difference, separately by evidence type and cohort"
      ],
      "uncertainty": "Original-price input sensitivity only; no calibrated individual prediction interval or current coverage claim.",
      "sensitivities": "Original-price bounds; age and mileage through frozen curves. Market index stays at the 19 October 2026 snapshot.",
      "decision": "Research test only. Zero independently validated current vehicles in the released evidence; quote mode remains blocked.",
      "nextEvidence": "Reviewed, matched original-price and current-price evidence with sampling, date alignment and development-use separation. Keep collection exports for later independent validation.",
      "source": "/valuation/price-anchor/SERVICE.txt",
      "checks": "/valuation/price-anchor/model.json",
      "referenceRows": 2761,
      "independentCurrentVehicles": 0,
      "sources": [
        {
          "path": "/valuation/price-anchor/model.json",
          "sha256": "c40b225cf5164176395a51bb3e8f500ff9d388cce2e6233a6cce6e1c7a2ba8b3"
        },
        {
          "path": "/valuation/price-anchor/market-index.json",
          "sha256": "621747ae06cfb04c4dacb5e369b3a56d47b3f77be9b7002e9788a84a5809374f"
        }
      ],
      "legacyId": "price-anchor-evaluation-v2"
    },
    {
      "outcome": "Police-reported personal-injury collisions at road location; not insurance claims or resident-driver risk.",
      "priorUse": "Earlier outcomes and model results were inspected. This release adds documentation, not a fresh independent test.",
      "nextEvidence": "Matched annual weather, independent LSOA exposure and a genuinely prospective assessment. No 2025 LSOA panel exists in this release.",
      "metrics": [
        "Held-out Poisson deviance and gain against a same-cohort comparator",
        "Observed / expected",
        "Exposure-weighted absolute rate error"
      ],
      "id": "frost",
      "study": "Frost and collision frequency",
      "href": "/data-lab/frost?version=v1",
      "modelRelease": "frost-phase8-v1",
      "population": "Frozen English authority cohort; 130 authorities in the 2025 assessment. Separate conditional LSOA pilot ends in 2024.",
      "unit": "Authority × year; LSOA supplement reported separately.",
      "exposure": "DfT all-motor miles at authority level. LSOA mileage is allocated and remains unvalidated; climate normals are not annual weather.",
      "training": "Primary frozen models trained on 2021–2023; phase-seven selection and phase-eight assessment checkpoint retained.",
      "evaluation": "2024 and 2025 comparisons, geographic folds, annual tail bands and conditional mileage supplement.",
      "overlap": "2025 descriptive outcomes had already been inspected. Geographic blocks and area histories are dependent; the LSOA supplement is not independent confirmation.",
      "selection": "Frozen phase-seven models; no post-assessment reselection.",
      "uncertainty": "Conditional paired geographic-block resampling; 1,000 draws for the primary 2025 comparison. Does not cover model selection or exposure measurement error.",
      "sensitivities": "Alternative geographic folds, annual bands and tails; LSOA benchmark and ±10% / ±20% exposure assumptions and common counted-network comparison.",
      "decision": "Authority screening gain persists, but calibration and spatial gates fail. Do not promote frost bands to a validated rating factor.",
      "source": "/research/frost/phase8-v1/design.json",
      "checks": "/research/frost/phase8-v1/audit.json",
      "reproduce": "/research/frost/phase8-v1/REPRODUCE.txt",
      "sources": [
        {
          "path": "/research/frost/phase8-v1/design.json",
          "role": "method",
          "sha256": "963b87b76c7d228a8be50d875b3823fcfe32a544626dd0822fca81a33e3f86ad"
        },
        {
          "path": "/research/frost/phase8-v1/source-register.json",
          "role": "inputs",
          "sha256": "94d113cff2e2587c52c4c6187afc0ec54b60ecfaecdd501687d8548bf0c906f3"
        },
        {
          "path": "/research/frost/phase8-v1/evidence.json",
          "role": "results",
          "sha256": "2c5d96dddbff27dc37e3e286fdf1ac25a42f09795fabd365f1b7626fa9290c1e"
        },
        {
          "path": "/research/frost/phase8-v1/audit.json",
          "role": "results",
          "sha256": "ff235d29778371f1f5fd8c270a9543652f1a516dc06b2672cfbbaa4a2329af46"
        }
      ]
    },
    {
      "outcome": "Police-reported personal-injury collisions at road location; not insurance claims or resident-driver risk.",
      "priorUse": "Earlier outcomes and model results were inspected. This release adds documentation, not a fresh independent test.",
      "nextEvidence": "A prospectively frozen specification, uninspected outcomes and independent exposure validation before any pricing decision.",
      "metrics": [
        "Held-out Poisson deviance and gain against a same-cohort comparator",
        "Observed / expected",
        "Exposure-weighted absolute rate error"
      ],
      "id": "parking",
      "study": "Parking and garaging",
      "href": "/data-lab/parking?version=v4",
      "modelRelease": "parking-v4",
      "population": "137 complete English authorities, 2021–2025; identical cohort across stages and variants.",
      "unit": "Authority × year",
      "exposure": "All-motor miles. Parking is derived from housing/car ownership and assumed provision/use, not observed garaging behaviour.",
      "training": "NB2 models with raw, urban, expanded, interaction and network specifications; training-only scaling in spatial folds.",
      "evaluation": "Leave one of nine English regions out, keeping all years for an authority together.",
      "overlap": "2021 household and 2024 provision data applied retrospectively. The 2026 network snapshot is a retrospective sensitivity. Spatial transfer is not year-ahead prediction.",
      "selection": "All 27 parking variants and both off-street/inside-garage bases retained; no best scenario selected.",
      "uncertainty": "Authority-cluster coefficient intervals; 2,000 resamples of nine regional loss blocks. Descriptive instability, not independent confirmatory inference.",
      "sensitivities": "Capacity and housing/car association assumptions, non-trunk subcohort, unresolved parked involvement and network controls.",
      "decision": "Exploratory incremental spatial prediction using an unvalidated proxy. No causal garage-use, driver-level or future-year validity claim.",
      "source": "/research/parking/v4/design.json",
      "checks": "/research/parking/v4/control-tests.json",
      "reproduce": "/research/parking/v4/REPRODUCE.txt",
      "sources": [
        {
          "path": "/research/parking/v4/design.json",
          "role": "method",
          "sha256": "3e8711a0fa9938b3965bd3e628473dda6bffcbc0a1c3551cf59e0fac2d7c0c27"
        },
        {
          "path": "/research/parking/v4/model-parking-phase4.py",
          "role": "method",
          "sha256": "2347bd8fc21103156b4b8fe7b325294a85368f247b5c1072dfc25abad206626d"
        },
        {
          "path": "/research/parking/v4/input-manifest.json",
          "role": "inputs",
          "sha256": "af03d8ee47fbe2c505660215ca3b8804680946f7f4a54af9c249181bb37b46bb"
        },
        {
          "path": "/research/parking/v4/source-register.json",
          "role": "inputs",
          "sha256": "e9b5202dab367c4d3032331058cfcf1c57fd9b47114bd565af7c55ef4d110aa5"
        },
        {
          "path": "/research/parking/v4/analysis.json",
          "role": "results",
          "sha256": "5b644ba1fd09e2005c2be730fdc6b4ffae7766bf7929b6b45cf06592c395c198"
        },
        {
          "path": "/research/parking/v4/control-tests.json",
          "role": "results",
          "sha256": "8c58a9a14d3729beb1adf8be4af7cbaa98d2d2b3a78e7745dc27430fb1d843c0"
        }
      ]
    },
    {
      "outcome": "Police-reported personal-injury collisions at road location; not insurance claims or resident-driver risk.",
      "priorUse": "Earlier outcomes and model results were inspected. This release adds documentation, not a fresh independent test.",
      "nextEvidence": "A prospectively frozen specification, uninspected outcomes and independent exposure validation before any pricing decision.",
      "metrics": [
        "Held-out Poisson deviance and gain against a same-cohort comparator",
        "Observed / expected",
        "Exposure-weighted absolute rate error"
      ],
      "id": "social",
      "study": "Social factors and learned adjacent bands",
      "href": "/data-lab/social-factors/deciles?version=v3",
      "modelRelease": "social-deciles-fusion-v3",
      "population": "133 exact-code English authorities, 2018–2025.",
      "unit": "Authority × year",
      "exposure": "Realised all-motor miles; historical deprivation vintages and 2011 no-qualifications share.",
      "training": "Expanding temporal training and nested geographic tuning; relearn cutpoints, penalty and fusion inside each fit.",
      "evaluation": "399 temporal assessment rows in 2023–2025 and 133 geographic assessment rows; nine leave-region-out folds.",
      "overlap": "Overlapping training windows and only nine regional blocks. Earlier outcomes and the original numerical attempt were inspected before the amendment.",
      "selection": "Strongest eligible penalty within 0.5% of minimum inner deviance; six comparator forms. Exact constrained ratio-one endpoint documented in numerical amendment.",
      "uncertainty": "199 complete paired region-bootstrap refits. Cutpoints, tuning and fusion relearned; limited tail precision and no post-selection guarantee.",
      "sensitivities": "Continuous, spline, ten-decile and fixed-pair comparators; annual/regional scores and boundary stability.",
      "decision": "No incremental gain: selected fusion returns the baseline. Candidate and replacement gates fail; no pricing promotion.",
      "source": "/research/social-factors/deciles-fusion-v3/protocol.json",
      "checks": "/research/social-factors/deciles-fusion-v3/findings.json",
      "reproduce": "/research/social-factors/deciles-fusion-v3/reproduce-v3.zip",
      "sources": [
        {
          "path": "/research/social-factors/deciles-fusion-v3/protocol.json",
          "role": "method",
          "sha256": "51c30452bb05ad248cdaf785fba1afef3250fc362b7b3c3acd095b9ae7220ff6"
        },
        {
          "path": "/research/social-factors/deciles-fusion-v3/social-deciles-fusion-v3.py",
          "role": "method",
          "sha256": "ce6bb924b875d19a7b27072da2ab5f482bbb35f6e417258d053419ede79a243d"
        },
        {
          "path": "/research/social-factors/deciles-fusion-v3/numerical-amendment.json",
          "role": "method",
          "sha256": "77f02a7fdff0d8aaa40f03969a4e7db7e54201842ae91bdd0986dc8b6424c64e"
        },
        {
          "path": "/research/social-factors/deciles-fusion-v3/frozen-design.json",
          "role": "inputs",
          "sha256": "b3f2e0035472e0ff3574e54c82cecd291cad02048868e2f4dddd5d81e4acfc75"
        },
        {
          "path": "/research/social-factors/deciles-fusion-v3/findings.json",
          "role": "results",
          "sha256": "ac99df90274e5d1dc4129d8c08038902191875a532093a08be13a6fce5842434"
        },
        {
          "path": "/research/social-factors/deciles-fusion-v3/checkpoint-audit.json",
          "role": "results",
          "sha256": "34058b66b0483b1d4d56c8ca6f693e1ba0daefea1f47aa22fc47cf982bd22405"
        }
      ]
    },
    {
      "outcome": "Police-reported personal-injury collisions at road location; not insurance claims or resident-driver risk.",
      "priorUse": "Earlier outcomes and model results were inspected. This release adds documentation, not a fresh independent test.",
      "nextEvidence": "A prospectively frozen specification, uninspected outcomes and independent exposure validation before any pricing decision.",
      "metrics": [
        "Held-out Poisson deviance and gain against a same-cohort comparator",
        "Observed / expected",
        "Exposure-weighted absolute rate error"
      ],
      "id": "national",
      "study": "England and Wales joint GLM review",
      "href": "/data-lab/ew-deciles/glm-review",
      "modelRelease": "ew-glm-review-v1",
      "population": "32,789 development LSOAs in England and Wales; complete supported cohorts vary for diagnostic extensions.",
      "unit": "LSOA × year",
      "exposure": "Allocated vehicle miles; road-length, benchmark and matched-counted-route alternatives are sensitivities, not independent exposure validation.",
      "training": "2022–2023 development, 40 nationally banded features jointly fitted in a Poisson frequency GLM; ten separate numeric diagnostic extensions.",
      "evaluation": "2022 fit → 2023 test; ten leave-region-out folds. Previously inspected 2024–2025 diagnostics.",
      "overlap": "Shared source-area/authority exposure allocations; unsupported categorical levels withheld from both comparator and candidate scoring.",
      "selection": "No automatic feature dropping, band merging or shrinkage. Human Keep/Drop/Simplify proposals require a complete refit.",
      "uncertainty": "162 source-area/authority clusters; 1,000 paired fixed-prediction block draws. No refit, selection or exposure uncertainty.",
      "sensitivities": "Benchmark and road-length offsets, matched-counted-route fit and numeric linear/spline diagnostics.",
      "decision": "User-led exploratory GLM review. No independent or insurer-claims validation; not completion of the earlier penalised/nested-selection protocol.",
      "source": "/research/ew-deciles/glm-review-v1/specification.json",
      "checks": "/research/ew-deciles/glm-review-v1/release-verification.json",
      "reproduce": "/research/ew-deciles/glm-review-v1/reproduce-ew-glm-review-v1.zip",
      "sources": [
        {
          "path": "/research/ew-deciles/glm-review-v1/specification.json",
          "role": "method",
          "sha256": "7e1de56fb406b25b72d5c51571cdbd8e9f991ac5713ce4d9f01318a81c301147"
        },
        {
          "path": "/research/ew-deciles/glm-review-v1/build-ew-glm-review.py",
          "role": "method",
          "sha256": "4bb1835f4b6affd251f14790ceaa94e3085ffd507dcb6d042ffa0b300bcf850b"
        },
        {
          "path": "/research/ew-deciles/glm-review-v1/input-checksums.json",
          "role": "inputs",
          "sha256": "542352cf91b7b96ba2af9526be7cadaebd26d11c161d46ae41fb2f574dacebfa"
        },
        {
          "path": "/research/ew-deciles/glm-review-v1/summary.json",
          "role": "results",
          "sha256": "d792d51a6778d59d594279f6d1c4dbafdf9738f355c2865a6baa187ff11ab023"
        },
        {
          "path": "/research/ew-deciles/glm-review-v1/release-verification.json",
          "role": "results",
          "sha256": "f8cf84a1b8bd8917e8b329daae1df172d78a0f03b4c80129a713d84d12a21bab"
        }
      ]
    }
  ]
}