{
  "schemaVersion": "1.3",
  "slug": "wales-20mph-casualty-attribution",
  "title": "Wales 20 mph Casualty Attribution Audit: A Constructive Non-Identification Result",
  "shortTitle": "Wales 20 mph casualty attribution",
  "url": "https://evidencepress.org/releases/wales-20mph-casualty-attribution/",
  "oneLine": "The first-year casualty decline is descriptively reproducible, but the frozen aggregate evidence identifies neither the point value nor the sign of the policy effect on affected roads.",
  "abstract": "A provisional Welsh Government graphic reported 2,402 casualties on roads recorded as 20 or 30 mph in the four-quarter baseline and 1,725 in the first four complete post-implementation quarters, an exact decline of about 28.18%. This source-bound audit reproduces that historical comparison and preserves later official vintages, then separates the combined posted-speed proxy from the causal estimand on road segments whose speed limit actually changed. The frozen corpus contains neither a versioned historical treatment map nor an observed no-policy outcome for affected roads. A deterministic constructive witness holds the same aggregate observables fixed while hypothetical affected-road counterfactuals imply effects of -50%, 0% and +100%. The values are counterexamples, not estimates or bounds. Within the registered corpus and rival class, neither the causal point nor its sign is identified; the positive 28% attribution remains unsupported. The result does not establish that the policy worked, failed or had no effect.",
  "datePublished": "2026-08-11",
  "dateModified": "2026-08-11",
  "version": "1.0.0",
  "doi": "10.5281/zenodo.21894035",
  "doiUrl": "https://doi.org/10.5281/zenodo.21894035",
  "conceptDoi": "10.5281/zenodo.21894034",
  "pdfUrl": null,
  "altPdfUrl": null,
  "zenodoUrl": "https://zenodo.org/records/21894035",
  "repoUrl": "https://github.com/ipitchford/wales-20mph-policy-analysis",
  "releaseUrl": "https://github.com/ipitchford/wales-20mph-policy-analysis/releases/tag/v1.0.0",
  "markdownUrl": "https://evidencepress.org/releases/wales-20mph-casualty-attribution/index.md",
  "bibtexUrl": "https://evidencepress.org/releases/wales-20mph-casualty-attribution/cite.bib",
  "audioUrl": "https://evidencepress.org/assets/audio/wales-20mph-casualty-attribution.mp3",
  "imageUrl": "https://evidencepress.org/assets/og/wales-20mph-casualty-attribution.png",
  "coverArtUrl": "https://evidencepress.org/assets/art/wales-20mph-casualty-attribution.svg",
  "media": [
    {
      "type": "audio",
      "url": "https://evidencepress.org/assets/audio/wales-20mph-casualty-attribution.mp3",
      "name": "Audio briefing — Wales 20 mph casualty attribution",
      "description": "Plain-English synthetic-voice summary of the descriptive decline, affected-road estimand and constructive non-identification result; a communication aid, not additional policy evidence.",
      "transcriptUrl": "https://evidencepress.org/assets/audio/wales-20mph-casualty-attribution.txt"
    }
  ],
  "authors": [
    "Ian Pitchford",
    "OpenAI Codex agent team"
  ],
  "license": "CC0-1.0",
  "status": "unrefereed-candidate",
  "verification": {
    "peerReviewed": false,
    "independentlyReproduced": false,
    "formallyVerified": false,
    "internallyReplayed": true,
    "detail": "Unrefereed candidate. The claim is limited to corpus-specific non-identification of the affected-road causal point and sign. Producer-side source, citation and exact-calculation checks exist, together with model-assisted adversarial review inside the coordinated workflow. No unaffiliated rerun, independent reimplementation, formal verification, external human specialist review, editorial peer review, policy-effect estimate or real-world impact evaluation has been established. The positive 28% attribution remains unsupported."
  },
  "assurance": [
    {
      "dimension": "availability",
      "label": "Availability and archiving",
      "question": "Is the evidence package publicly retrievable from an archive under a persistent identifier?",
      "state": "passed",
      "evidenceUrl": "https://zenodo.org/records/21894035",
      "note": "The public GitHub v1.0.0 release and published Zenodo record expose the declared archive, validation receipt and checksum manifest. Producer-side readback matched the recorded release hashes; this establishes public byte availability, not correctness or independent reproduction."
    },
    {
      "dimension": "internalReplay",
      "label": "Internal replay",
      "question": "Does the producer’s own pipeline reproduce the stated result from the archived package?",
      "state": "passed",
      "note": "The producer workflow completed a fresh exported-tree replay: 283 ordinary tests, 283 optimized-interpreter collection-smoke tests, the standalone optimized scientific control and the registered case gates passed. This is producer-side replay, not unaffiliated empirical reproduction.",
      "evidenceUrl": "https://github.com/ipitchford/wales-20mph-policy-analysis/releases/tag/v1.0.0"
    },
    {
      "dimension": "independentRerun",
      "label": "Independent rerun",
      "question": "Has someone else run the supplied implementation and obtained the stated result?",
      "state": "not-assessed",
      "note": "No unaffiliated party has reported rerunning the immutable public package."
    },
    {
      "dimension": "independentReimplementation",
      "label": "Independent reimplementation",
      "question": "Has someone else reached the result from an independent implementation?",
      "state": "not-assessed",
      "note": "No unaffiliated implementation has reconstructed the source-vintage ladder or non-identification witness without producer code or intermediates."
    },
    {
      "dimension": "formalVerification",
      "label": "Formal verification",
      "question": "Is a formalised statement machine-checked, and over which trusted base?",
      "state": "not-assessed",
      "note": "The exact arithmetic and schema guards are executable, but no identified theorem has been checked in a proof assistant over a stated trusted base."
    },
    {
      "dimension": "specialistReview",
      "label": "Specialist review",
      "question": "Has a domain specialist assessed the argument?",
      "state": "not-assessed",
      "note": "Read-only external-model adversarial review is retained as producer-workflow evidence; no external human road-safety, official-statistics or causal-inference specialist review has been established."
    },
    {
      "dimension": "editorialPeerReview",
      "label": "Editorial peer review",
      "question": "Has a journal or venue run peer review to a decision?",
      "state": "not-assessed",
      "note": "No journal or other venue has conducted editorial peer review to a decision."
    },
    {
      "dimension": "dataEnvironmentReproducibility",
      "label": "Data and environment reproducibility",
      "question": "Are data and computational environment pinned well enough to rebuild?",
      "state": "partial",
      "evidenceUrl": "https://github.com/ipitchford/wales-20mph-policy-analysis",
      "note": "The workflow records an offline dependency lock, command vector, source hashes, ordinary and optimized paths and a clean exported-tree replay, but no independently recreated container, Nix or Guix environment has been established."
    }
  ],
  "provenance": {
    "aiGenerated": true,
    "aiAssisted": true,
    "generatedBy": [
      "OpenAI Codex agent team",
      "Anthropic Fable read-only adversarial review workflow"
    ],
    "humanRole": "Problem selection, research direction, mediation and publication authorization; Ian Pitchford is the declared human scholarly creator and repository maintainer.",
    "disclosure": "AI systems performed source discovery, structured analysis, implementation, testing, drafting and read-only adversarial review under human direction. Multiple model stages and exact replay are one producer-coordinated workflow, not independent empirical reproduction, external human specialist review or editorial peer review."
  },
  "problem": null,
  "corrections": [],
  "keywords": [
    "Wales",
    "20 mph",
    "road casualties",
    "policy identification",
    "causal inference",
    "non-identification",
    "constructive counterexample",
    "treatment misclassification",
    "official statistics",
    "reproducible policy analysis"
  ],
  "keyResults": [
    "The provisional source statistic reconstructs exactly from 2,402 to 1,725 casualties, or about -28.18%, but it is a before/after comparison on the combined 20/30 mph class rather than the registered affected-road causal estimand.",
    "Later frozen routes are not numerically invariant: the current StatsWales extract gives about -26.10%, while the current DfT collision-row diagnostic gives about -24.18%.",
    "The corpus lacks both a versioned historical map of affected road segments and the affected roads' no-policy post outcome.",
    "Three hypothetical worlds reproduce the same aggregate observables while implying affected-road effects of -50%, 0% and +100%; they establish corpus-specific point-and-sign non-identification but are not estimates or bounds.",
    "The positive 28% causal attribution remains unsupported, and no zero, beneficial, harmful or otherwise signed policy effect is reported."
  ],
  "reviews": [],
  "evidencePackage": "A one-case Policy Identification Observatory package with 25 registered sources and three frozen search records; 125 source bindings; 29 narrow frozen-text anchors; 17 exact calculations; a 45-entry non-circular scientific review target; exact derived tables; registered estimand, rivals, missingness and constructive witness; retained adversarial-review history and transport receipts; ordinary and optimized execution paths, hostile controls, manifests, provenance, licences and a clean-checkout release gate. Final public replay, archive and cross-channel hash evidence must be taken from the authoritative released receipt rather than inferred from this local draft.",
  "openProblems": [
    "Freeze a versioned historical segment-level map of the roads that changed, remained 30 mph exceptions or later reverted, then link collision locations under explicit coding and linkage-error rules.",
    "Develop and independently review a comparison design with multiple pre-period diagnostics, placebo dates and explicit interference, anticipation and concurrent-intervention assumptions.",
    "Measure affected-segment traffic exposure, realised speeds, enforcement, weather and reporting changes where they bear on a risk or displacement estimand.",
    "Register source-backed restrictions prospectively and compute an empirical identified set only if the restrictions support one; preserve sign ambiguity or NOT_BOUNDED otherwise.",
    "Independently reconstruct the official source-vintage ladder and reimplement the observation-map witness without producer code or intermediate files.",
    "Compare this workflow with a matched conventional expert assessment at the same assurance endpoint, measuring elapsed time, human effort, compute, rework and missed valid effects."
  ],
  "relatedWorks": [
    {
      "citation": "Welsh Government (2025). Police recorded road collisions: July to September 2024 (provisional).",
      "url": "https://www.gov.wales/police-recorded-road-collisions-july-september-2024-provisional-html"
    },
    {
      "citation": "Office for Statistics Regulation (2026). 20mph speed limit.",
      "url": "https://osr.statisticsauthority.gov.uk/guidance/2026-what-to-look-out-for-20mph/"
    },
    {
      "citation": "Slower Streets, Safer Streets? The Causal Effect of Wales's 20 mph Default Speed Limit on Road Casualties and Property Values. APEP 0513 v1 (2026).",
      "url": "https://ape.socialcatalystlab.org/papers/apep_0513/v1/paper.pdf"
    },
    {
      "citation": "Does the Default Kill? Speed Limit Reclassification and Road Casualties in Wales. APEP 0734 v1 (2026).",
      "url": "https://ape.socialcatalystlab.org/papers/apep_0734/v1/paper.pdf"
    },
    {
      "citation": "The Speed Penalty: Causal Evidence from Wales's Default 20mph Limit. APEP 0744 v1 (2026).",
      "url": "https://ape.socialcatalystlab.org/papers/apep_0744/v1/paper.pdf"
    },
    {
      "citation": "Edinburgh 20 mph evaluation inspected as a changed-road/stayed-road analogue in the bounded novelty audit.",
      "url": "https://doi.org/10.1136/jech-2023-221612"
    },
    {
      "citation": "Anonymous (2026). Stocks Are Not Flows: An Identification Audit of Investor Pressure and Tenure Change in England. Evidence Press.",
      "url": "https://evidencepress.org/releases/stocks-are-not-flows/"
    }
  ],
  "operatingModel": {
    "version": "1.0",
    "workId": "ep-work:wales-20mph-casualty-attribution",
    "attemptIds": [
      "ep-attempt:wales-20mph-casualty-attribution"
    ],
    "aims": [
      "policy"
    ],
    "artifactRoles": [
      "research-output",
      "evidence-assessment",
      "method-demonstration"
    ],
    "lineageId": null,
    "accelerationPrimitives": [
      "adversarial-controls",
      "productive-failure",
      "identification-gate",
      "counterexample-proxy-first",
      "assurance-vector",
      "agent-readable-research-object"
    ],
    "decisionObject": {
      "type": "counterexample",
      "description": "Three hypothetical affected-road worlds preserve the same aggregate observed baseline and policy-period totals while implying negative, zero and positive causal effects.",
      "scope": "The first-four-complete-quarters affected-road casualty estimand, frozen official aggregate corpus and admitted rival class registered in the Wales 20 mph case. The worlds are counterexamples, not estimates or bounds."
    },
    "bottleneckTargeted": [
      "discovery",
      "assurance"
    ],
    "semanticBridge": {
      "state": "partial",
      "description": "The release explicitly separates the combined 20/30 mph descriptive proxy from the affected-segment causal estimand and machine-checks the constructive observation-map witness. The empirical bridge cannot be completed because the frozen corpus lacks a versioned historical treatment map and affected-road no-policy outcome.",
      "remainingRisks": [
        "The non-identification result is conditional on the registered aggregate corpus, observation map and admitted rival class.",
        "Speed-limit coding corrections, source-vintage changes, under-reporting and reporting-system changes limit empirical interpretation.",
        "No human road-safety, official-statistics or causal-inference specialist has independently assessed the semantic bridge.",
        "Future linked segment data or independently justified restrictions could change the identification state."
      ]
    },
    "humanJudgmentGates": [
      "Assess whether the affected-road estimand and admitted rivals adequately represent the substantive policy question.",
      "Judge the credibility, legality and ethical acceptability of any future linked treatment, collision and exposure data.",
      "Evaluate policy merits, welfare, distributional effects, rights and implementation choices; the identification audit does not decide them.",
      "Authorize authorship, archival publication and any public characterization of external-model review."
    ],
    "parentLinks": [
      {
        "relation": "reuses-method",
        "workId": null,
        "legacyReleaseSlug": "stocks-are-not-flows",
        "externalUrl": null,
        "inheritedClaim": "The project reuses the identification-before-estimation workflow: distinguish the estimand from available proxies, register admitted rivals, test the observation map and publish a justified stopping boundary when point attribution fails.",
        "inheritedAssuranceCeiling": "Method reuse is not independent confirmation of the housing release or of this Wales result. Both remain conditional on their declared observation maps and neither establishes a real-world policy or acceleration effect."
      }
    ],
    "assuranceTarget": {
      "dimensions": [
        "independentRerun",
        "independentReimplementation",
        "specialistReview",
        "semanticValidation",
        "noveltyAssessment",
        "priorityAssessment"
      ],
      "nextAction": "An unaffiliated multidisciplinary team should reconstruct the official source-vintage ladder, independently assess the affected-road estimand and rival class, and reimplement the observation-map witness without producer code or intermediates; a later design should add the missing historical treatment map before estimating an effect.",
      "claimCeiling": "The released aggregate corpus and declared rival class recover neither the point value nor the sign of the affected-road causal effect. This does not establish the policy's true effect, a universal impossibility result, exhaustive novelty, independent reproduction, peer review, policy merit or an acceleration impact."
    },
    "impactClaims": [
      {
        "id": "policy-no-impact",
        "aim": "policy",
        "outcome": "Faster and more decision-adequate assessment of causal policy claims with fewer unsupported point attributions",
        "setting": "Evidence assessment of the Wales default 20 mph first-year casualty attribution claim",
        "status": "NO_IMPACT_EVIDENCE",
        "designClass": "none",
        "comparator": "No matched conventional expert assessment was conducted.",
        "estimand": "No comparative assessment-time, decision-quality or unsupported-attribution effect was estimated.",
        "evidenceRefs": [],
        "registeredDesignRef": null,
        "independentAssessment": null
      }
    ],
    "ibeHypotheses": [
      "policy-assessment-improvement",
      "negative-result-option-value"
    ]
  }
}