20mph Evidence Lab — Phase 5, 4 November 2026

This release compares published journey observations and supplies a conditional
car-travel break-even calculator. It does not estimate the policy effect, national
net benefit, NHS cash saving or insurer premium effect. Phase3 identification
checks failed; no casualty coefficient is monetised.

REPRODUCE OFFLINE
Unzip reproduce-20mph-v5.zip into a directory. Python 3 standard library only:
  python reproduce.py --input-dir inputs --output-dir reproduced
Input bytes are checked against input-pins.json before any calculation. Compare
reproduced/chart-checksums.json with chart-checksums.json. Outputs results.json,
journeys.csv and scenarios.csv reproduce byte for byte. No network is needed.
The same individual inputs and code are available under this release URL.

JOURNEYS
Frozen phase2 extract:15 routes x2 directions x2 peak periods =60 combinations;
180 observations for spring2023, spring2024 and autumn2024. Original workbook
SHA256 c73a995c62c69e7d4c35c6d235805d5e51453f891d61d1aeea9f8696a04708d4.
Original source: https://tfw.wales/sites/default/files/2025-07/Vehicle_journey_times_Feb23_to_Nov24.ods
The existing phase2 reproduction validates all source means, 360 differences
and180 percentile ranges. V5 preserves every phase2 value and adds percentage
changes, descriptive spread counts and an explicit same-season primary view.
Spring samples: Tuesdays–Thursdays, February–April, Easter public holidays avoided.
Autumn changes season; unweighted summaries are not traveller averages. Percentile
ranges are neither confidence intervals nor standard deviations. No regression,
reliability valuation or independence claim is made for these overlapping routes.
No journey-specific roadworks history or matched traffic weighting was verified.

SCENARIOS AND UNITS
Every traffic, occupancy, delay, cost and severity input is hypothetical. Defaults:
100000 car trips/day,250days/year,1.5people/car,30seconds assumed policy delay,
all commuting,10years. These are a unit-cohort illustration, not Wales estimates.
Unentered costs are excluded, not proven zero. No casualty effect is selected by
default: benefit and component-balance fields are null, including annual rows.
Each break-even threshold is an alternative annual expected casualty count.
Do not add the fatal/serious/slight equivalents. No at-risk baseline is available
for a percentage-reduction threshold. Negative entered prevention means extra harm.

All money: GBP2023 market prices. TAG May2026v2.03 values uprated to2026 using
current macroeconomic series (including forecast2026 GDP). PV reference2026,
upfront costs at t0, annual flows at year-end2027 onward, horizon1–30years.
Human casualty component: frozen real value, discount1.5%.
Nonhuman casualty and time: real growth1.5%, discount3.5%.
Entered annual resource costs: constant real value, discount3.5%.
The usual DfT reporting PV base is2023; our labelled2026 reference is an explicit
presentation choice. This is not a complete TAG appraisal. Do not paste nominal
budget cash into2023 market-price inputs. Missing future/residual/other effects
are not zero effects. Existing implementation spending is sunk for prospective
decisions; retrospective appraisal must place it on its historical payment dates.

Full formula, source-cell, price and discount audit: inputs/valuation.json.
Original TAG workbook and extraction code retained in the source repository under
research-archive/20mph/v5/valuations; concatenate the two ordered source parts and
check workbook-parts.json, then run extract-valuation-inputs.py (openpyxl required).
No workbook macros are executed. Independent checks start from1997 source values.

COSTS AND OPTIONAL AIR QUALITY
Government disclosures mix forecast, paid/awarded, provisional spent-to-date and
allocations. Do not sum them. grant-rows.json retains the17 explicitly named20mph
FY2025/26 awards behind the derived total. Missing authority entries are not zero.
The optional air-quality extension does not proceed: existing pilot has no pre-
implementation baseline; national dated treated/control pairs and hourly usable
capture have not been verified. No emissions or carbon benefit is included.

PROVENANCE AND RIGHTS
source-ledger.json gives URLs, retrieval dates, original hashes and scope.
checksums.json hashes the release files, including the compact ZIP (itself not
included inside the ZIP). TAG/Crown data underOGL where stated. TfW/Jacobs/OS
originals retain source rights and are not publicly remirrored here. Earlier public
phase1–4 evidence files remain unchanged. Raw audit originals are retained in the
source repository with an inventory; split large files verify to original hashes.

PUBLICATION SCOPE
Phase5 trade-offs complete as conditional analysis. Phase6 article is not part of
this release. No automation, national policy recommendation or benefit claim.
