513,801 collision records
Great Britain, 2021–2025, with vehicle and casualty joins. Fresh 2025 files checked against the archived vintage.
Retrieving and verifying the requested analytical release. Result data loads before optional map boundaries and source checks.
Your filters and release stay in the address. Navigation remains available.
Archived phase 1 snapshot. Open the phase 2 pilot and updated reconciliation →
20MPH EVIDENCE LAB · PHASE 1 · 3 NOVEMBER 2026
A Wales-first investigation of injury collisions, actual speeds and journey times. The source audit and study design are now available; the effect of the policy has not been estimated.
PHASE 1 DECISION
The record imports pass the join checks. Two source discrepancies remain visible below. No road section yet meets the study’s requirement for both a verified legal date and versioned geometry.
Great Britain, 2021–2025, with vehicle and casualty joins. Fresh 2025 files checked against the archived vintage.
English areas with complete, unflagged pre-period records in the initial screen. Local 20mph policies still need checking; none is an approved control.
Historical network length and the percentage covered remain unknown. This is a data-coverage result, not a count of roads changed.
Recommended next approval: Phase 2, an area-level descriptive pilot. First extend the historical imports and check stable geography. Hold road-level causal estimates until dated geometry and credible comparisons are established.
CHECK THE NUMERATOR
Wales, 2025. Country is assigned from the recorded ONS district of the collision. The refreshed DfT files are compared with the Welsh Government publication of 4 June 2026.
| Measure | Imported | Published | Difference | Check |
|---|---|---|---|---|
| Injury collisions · all roads | 3,004 | 3,004 | 0 | Matched |
| All casualties · all roads | 4,078 | 4,078 | 0 | Matched |
| Fatalities · all roads | 91 | 91 | 0 | Matched |
| Recorded serious casualties · all roads | 1,116 | 1,116 | 0 | Matched |
| Fatal collisions · all roads | 83 | 83 | 0 | Matched |
| Injury collisions · recorded 20/30mph | 1,488 | 1,469 | +19 | Unresolved |
The 19-collision difference remains unresolved. Release vintage or speed classification may explain it, but that has not been established. The source values are retained. A recorded 20/30mph category is also not a verified fixed road network.
SPEED DISTRIBUTION · CHECK FAILED
For summer 2025, TfW Table 1 reports 51.5% at or below 24mph. The matching bins in Table 4 sum to 54%. Table 4 is excluded from interpretation pending clarification.
Official speed tables →JOURNEY DURATIONS · CHECK PASSED
60 journey-time and 60 variability rows reproduce the published differences after duration normalization. Source cells contain date offsets that must be removed before reading them as seconds.
Official journey tables →DEFINE WHAT CHANGED
The register distinguishes a legal commencement date from a consultation, an observed limit and evidence that a change had happened by a certain date. Named roads are not treated as verified geometries.
| Evidence | What the date establishes | Remaining limitation |
|---|---|---|
| Wales restricted roads subject to default legislation; existing local orders/exceptions must be checked | Default restricted-road speed limit becomes 20mph exact legal commencement · 2023-09-17 | National commencement is not the start date of pilot schemes or every 20mph road. |
| Wrexham proposed 20-to-30mph reversals | Consultation open; final decisions still pending proposal stage only · 2024-12-13 | Published TRO proposal and consultation date do not establish legal commencement. |
| B5605 Wrexham Rd/High Street, Johnstown; A525 Bryn-Y-Grog Road | Two named roads reported returned to 30mph effective by publication upper bound · by 2025-05-21 | Exact segment extents, sealing date and commencement date not established by release. Do not map by road name alone. |
| 52 Wrexham highway sections approved for reversal | Council reports reversals complete and signage in place effective by publication upper bound · by 2025-09-03 | Completion statement is an upper bound, not 52 individual commencement dates. May forecast of end-June completion must not be used as fact. |
| Five unnamed Wrexham TfW monitoring sites, excluded from revised 38-site series | 20mph in July 2025 monitoring; 30mph in September 2025 monitoring interval censored monitoring evidence · July / September 2025 | Precise monitoring days, site identities and individual legal commencement dates are not provided here. Do not invent 1 July or 1 September start dates. |
The current DataMapWales services were checked. The 30mph layer is a buffered, changing snapshot. The national 20mph sample contains geometry quality problems, and a verified complete historical download was not obtained. Stage 3 describes objection and order-sealing procedures; it does not establish that a limit is operative.
Required for each section: stable geometry identifier and checksum, authority, old/new limits, exact effective and end dates, legal order, treatment class, traffic calming, enforcement changes and exposure quality. A complete inventory must include roads with zero collisions.
Current 30mph layer · Section schema · Empty verified-section register
VERSION 1 · FROZEN 3 NOVEMBER 2026
This protocol follows known published results and earlier site work. It is a versioned commitment for the next analysis, not a claim that outcomes were unseen.
Primary outcomes: injury collision counts and total casualties. Keep fatalities separate; use adjusted serious injuries only after checking force/year coverage. Per-mile rates require matching road and time coverage.
Use pre-policy information to match English roads or weight comparison areas. Check local policies, trends, recording changes and spillovers. Welsh 30mph exceptions are a sensitivity comparison because their selection differs.
The initial area screen uses all-road counts in 2016–2019 and 2022, indexed to 2016. It finds 136 complete, unflagged English candidate areas among 153 screened. Similar trends alone do not make an area a valid counterfactual; none has passed the local-policy check.
The downloadable power table varies control volume and variance inflation using 2022 baseline counts. It assumes 80 pre and 27 post months and an 80% target power at a two-sided 5% level. It is an optimistic count-based sensitivity, not established design power. Block simulations and placebo analyses are still required.
Each source has its own coverage and access state. Archived inputs are identified by SHA-256; a checked catalogue is not counted as an imported dataset.
| Source | Coverage | Access / readiness |
|---|---|---|
| STATS19 collision records, latest five final years | 2021–2025 | Imported; archive hash verified by auditProvenance and limitsOriginal bytes archived losslessly as gzip; sha256 applies to decompressed bytes. Initial reproducible extraction, no human source review asserted. Open Government Licence v3.0 06a03c0e5504175e23ed98e3e0b4767ac61d396c073405a1bddab9339c18f948 |
| STATS19 vehicle records, latest five final years | 2021–2025 | Imported; archive hash verified by auditProvenance and limitsOriginal bytes archived losslessly as gzip; sha256 applies to decompressed bytes. Initial reproducible extraction, no human source review asserted. Open Government Licence v3.0 05453dca2651ad8a27b4ac0e6a0450a852c506e4d495111ffca398f3d311e26a |
| STATS19 casualty records, latest five final years | 2021–2025 | Imported; archive hash verified by auditProvenance and limitsOriginal bytes archived losslessly as gzip; sha256 applies to decompressed bytes. Initial reproducible extraction, no human source review asserted. Open Government Licence v3.0 870be5a934ae10baf5c52163d59eccba0ab2f40327ffa6b2bd354d2e7ee6b108 |
| STATS19 codebook and historical mappings | 2021–2025 | Imported; archive hash verified by auditProvenance and limitsOriginal bytes archived losslessly as gzip; sha256 applies to decompressed bytes. Initial reproducible extraction, no human source review asserted. Open Government Licence v3.0 51b08606ef3dad1ae5d4c8726241b022fc8e4bbdc910ef0e5cc2fb92b054fc12 |
| STATS19 collision records: 2025 refresh | 2025 final | Imported; fresh bytes override archived 2025 rowsProvenance and limitsKept separate from the existing site analytical release. Reconciliation includes speed-stratum differences. Exact fresh bytes can be reconstructed from the pinned five-year input plus a small order/delta file; all reconstructed SHA-256 checks pass. Open Government Licence v3.0 e699c19bdb82e47e449b6f3381ac87631865f6a31cef118c9955ba80ca870e2b |
| STATS19 casualty records: 2025 refresh | 2025 final | Imported; fresh bytes override archived 2025 rowsProvenance and limitsKept separate from the existing site analytical release. Reconciliation includes speed-stratum differences. Exact fresh bytes can be reconstructed from the pinned five-year input plus a small order/delta file; all reconstructed SHA-256 checks pass. Open Government Licence v3.0 b343039991f63f526bbcac0a5e6102e3a6e4c0b65d077e3ae37dbd53d1a7319a |
| STATS19 vehicle records: 2025 refresh | 2025 final | Imported; fresh bytes override archived 2025 rowsProvenance and limitsKept separate from the existing site analytical release. Reconciliation includes speed-stratum differences. Exact fresh bytes can be reconstructed from the pinned five-year input plus a small order/delta file; all reconstructed SHA-256 checks pass. Open Government Licence v3.0 dbeadb9745a8821ab8132f0dde59c246159e23059ab5db2c9a4381d44141a430 |
| Existing authority-year geography extract | 2016–2025; pre-policy years only used for comparison screening | Reused with extract hash; underlying source provenance embeddedProvenance and limitsNo 2015 coverage in this extract. Area totals cannot identify individual treated roads. DfT inputs: Open Government Licence v3.0 fb5498178b3045194f74e771b5a0cb6a50e4c99b0d892c11a480243462596f0f |
| StatsWales quarterly road casualties | 2015–Q2 2026 | Catalogue verified; not importedProvenance and limitsFinal and provisional vintages must remain separate; shared STATS19 origin is not independent corroboration. See publisher; no raw redistribution for this catalogue-only entry |
| DataMapWales road limits and exceptions | Current map | Partial: map services checked; dated network unavailableProvenance and limitsMap omits some authorities and part-time limits. Proposal stages are not legal commencement. See road-history manifest. See publisher; no raw redistribution for this catalogue-only entry |
| DfT traffic counts and annual flows | 2000–2025 | Existing area mileage available; road-level match pendingProvenance and limitsCounted/estimated flags must be retained. Annual AADF does not supply measured monthly mileage. See publisher; no raw redistribution for this catalogue-only entry |
| OS Open Roads geometry | Versioned road network | Catalogue verified; not imported for this studyProvenance and limitsGeometry contains no verified legal speed-limit history. See publisher; no raw redistribution for this catalogue-only entry |
| Welsh annual collision and casualty publication | 2025; published 4 June 2026 | Reconciliation controls transcribed with sourceProvenance and limitsNo adjustment or silent overwrite when imported speed totals differ. See publisher; no raw redistribution for this catalogue-only entry |
| Vehicle speed monitoring data 2025 | July2023–September2025 | Imported and parsed; source consistency checks recordedProvenance and limitsLimited monitored sites/routes. See monitoring audit for source inconsistencies and duration parsing. Dataset-specific reuse licence unverified; original files linked at publisher, not mirrored publicly. 32d07beea649ab3316abaeb7602d4f5dffd7385a4801b133c32d911d06e0c08c |
| Vehicle journey times Feb23 to Nov24 | 2023–2024 published monitoring | Imported and parsed; source consistency checks recordedProvenance and limitsLimited monitored sites/routes. See monitoring audit for source inconsistencies and duration parsing. Dataset-specific reuse licence unverified; original files linked at publisher, not mirrored publicly. c73a995c62c69e7d4c35c6d235805d5e51453f891d61d1aeea9f8696a04708d4 |
| OS Speed data | 2023–2024 published monitoring | Imported and parsed; source consistency checks recordedProvenance and limitsLimited monitored sites/routes. See monitoring audit for source inconsistencies and duration parsing. Dataset-specific reuse licence unverified; original files linked at publisher, not mirrored publicly. 3528901d80158e0dd0e4fabf066066f5f1258bb64d34de24af83e36717830014 |
The audit runs offline from pinned inputs using Python’s standard library. The reproduction script downloads those inputs, checks hashes and reruns both audits. A changed source fails its checksum instead of silently changing this release.