Last updated:Reading time: 9 minInspections documented:1,700+
Open methodology
The research behind the braking simulator: 974 real tests, every source named
The simulator predicts stopping distances from tyre grade, tread depth, age, surface
and weather. This page shows what those predictions are checked against: 974 published braking
tests, the physics inside the engine, the runs it failed along the way, and how to verify any
number on it yourself.
The simulator is scored against 974 published braking tests
from ADAC, Auto Bild, TÜV SÜD, Tekniikan Maailma, Vi Bilägare, Car and Driver and 212 other named sources.
For each test the engine is given the conditions and asked to predict the measured stopping distance.
848 of the 871 testable benchmarks land within tolerance. That is 97.4%, with a mean error of
6.5%. Every test carries its source, so any row can be checked.
Real tests, not simulations: every row is a measurement from an instrumented test.
A real car, real tyres, a real surface, a real measured distance.
It failed first: the first saved validation run passed 9 of 101 tests. Months of
iteration against the corpus got it to 97.4%. That history is below.
Exclusions are written down: 103 entries are marked untestable, each with a reason
recorded in the corpus. The 23 tests that still fail stay in the score.
You can rerun it: the corpus and the scoring script are available on request.
How to ask.
★ 5.0 Google (350+ reviews)✓ 100% Trade Me (1,250+ feedback)🚚 Free North Island delivery🛡 Full manufacturer warranty
On this page
On this page 7 sections
1 · The evidenceThe test corpus: 974 measured braking tests
One row of the corpus is one published braking test: the source, the vehicle, the tyre (brand, model,
size, EU grade where labelled), the surface and water depth, the start and end speeds, and the measured
distance. The rows come from the published programmes of independent testing organisations across Europe,
North America, Scandinavia and Australasia.
415 dry
Dry tarmac stops, mostly instrumented 100 to 0 km/h tests.
364 wet
Wet braking, typically 80 to 20 km/h on a 1 mm water film.
109 snow
Packed-snow programmes from Nordic winter testing.
67 ice
Polished-ice rink tests, plus a 146-run public ice and snow catalogue.
The largest contributors: Tekniikan Maailma (42 tests), Vi Bilägare (64),
Car and Driver instrumented tests (27), Auto Bild (49+ across summer,
winter and all-season programmes), ADAC (63+), TyreReviews,
TÜV SÜD, Continental's Contidrom tread programme, and NZ skid-resistance
work in NZTA Research Reports 293, 244 and 382. 218 named sources in all.
Why published tests? Because anyone can look them up.
We did not run these tests. ADAC, Auto Bild and the rest did, with instrumented cars on controlled surfaces.
The work here is collecting them into one standardised corpus and building an engine that has to answer to
all of them at once.
2 · The buildHow it was built, including the failures
1 · Collect
Gather every published braking test that could be verified,
plus the research base: 71+ papers and standards on friction, hydroplaning, ABS and tread mechanics.
2 · Score
Build a test suite that feeds each test's conditions to the
engine without the answer and compares the prediction to the measured distance.
3 · Iterate
The first saved run passed 9 of 101. Hundreds of runs
followed. Each failure was traced to a physics rule, fixed at the rule, then rescored against the
whole suite so a fix in one place could not quietly break another.
4 · Expand
The corpus grew through eight documented expansions, 109 rows
to 974, adding non-German sources, vans, heavy vehicles, off-road tyres, ice and snow.
5 · Audit
A data-quality pass excluded 103 untestable entries (missing
conditions, unverifiable figures), each with a written reason, leaving 871 testable benchmarks.
The real test: categories added late, vans,
3,000 to 4,000 kg vehicles, off-road tyres on tarmac, arrived after the engine was calibrated.
The pass rate held without any physics changes. Predicting data it was never tuned on is what separates
a model from a curve fit, and that is why the corpus kept growing.
3 · The modelThe physics inside
The engine is classical braking physics with a composite friction model. It is not fitted to individual
tests. Four pieces do most of the work:
Stopping distance
d = v² / (2·g·(μ·cosθ + sinθ))
Braking distance from speed, effective friction and gradient. Reaction distance
(v × t) is reported separately, never hidden inside the braking figure.
Surface friction (25 surfaces, peak against locked-slide) scaled by tyre and condition factors,
each with a named source: EU 2020/740 grades, TÜV and ADAC tread data, SAE J2246 load sensitivity.
Tread depth
η(d) = (d / 8)^0.25
A one-parameter power law, smooth from new to bald, checked against Continental's Contidrom
programme and Discount Tire's seven-point tread test.
Hydroplaning
V = 10.35 × √P
NASA's Horne and Dreher formula (TN D-2056, 1963), applied only at standing-water depths,
the regime it was derived for.
When a table value has to move, the change is logged: what moved, the evidence, the physical
justification, and the result on the full suite. The current engine is v3.7.1; its one logged calibration
raised packed-snow grip 5% after the 93 snow tests showed the old value predicting distances 4.7% too long.
The corrected figure sits mid-range of published packed-snow numbers for current winter tyres. Nothing in
the model is tuned to a single test.
4 · The claimAccuracy in full
One claim, scored one way. For each testable benchmark the engine predicts the stopping distance from
the test's conditions alone. A prediction passes if it lands within tolerance of the
measured figure: 15% for most sources, 12% for the highest-quality programmes, wider only where the source
itself published a range.
Surface
Testable benchmarks
Within tolerance
Pass rate
Mean abs. error
Dry tarmac
378
374
99%
5.8%
Wet tarmac
318
307
97%
6.4%
Snow
93
88
95%
8.1%
Ice
59
56
95%
8.4%
Wet concrete, gravel and other
23
23
100%
6.0%
All testable benchmarks
871
848
97.4%
6.5%
Engine v3.7.1, validation run of 1 September 2026. Mean signed bias across the suite is
minus 1.2%, so the model is not leaning long or short. This exact table regenerates from the corpus with one
script, and the 23 failing benchmarks stay in the corpus and in the score.
Why 15% and not 1%: two identical cars on the
same afternoon stop metres apart, and published repeat tests differ by several percent run to run. A model
claiming decimal-point accuracy against field tests would be describing its own curve fit, not the road.
The tolerance matches the noise in the underlying data.
5 · The edgesWhat it is not
Not a lab instrument
It predicts typical instrumented-test
results. Your car, your road and your reaction time will differ, and the gap between them is the
lesson the tool exists to show.
Not brand-specific
It models grades, categories and
condition, not individual tyre models. Two Grade A tyres can differ within the grade band.
Not legal advice
Results are educational. WoF rules and
the Road Code govern what is legal on the road.
6 · SourcesThe research base
Standards and formulas
NASA TN D-2056, Horne and Dreher, hydroplaning critical speed (1963)
EU Regulation 2020/740 and UNECE R117, wet grip grading and its test procedure
SAE J2246, tyre load sensitivity · SAE J2452, rolling resistance
PIARC (1995), international skid-resistance harmonisation · ASTM E1960, friction index
AASHTO Green Book, stopping sight distance and grade effects
Test programmes in the corpus (218 named sources; the largest)
ADAC summer, winter and all-season programmes (2023 to 2025)
Auto Bild 52-tyre tests, winter qualifying, all-season (2024 to 2025)
Tekniikan Maailma and Vi Bilägare Nordic winter programmes (2024 to 2025)
Car and Driver instrumented tests · TyreReviews independent programmes
TÜV SÜD worn-tyre study · Continental Contidrom tread-depth programme
NZTA Research Reports 293, 244 and 382, NZ skid resistance and crash risk
First-party data
Our own workshop dataset: 1,600+ documented tyre inspections with 45,000+ photographs,
published in the Workshop Tyre Report
7 · Verify itFor journalists and researchers
Every number on this page reproduces from two files: the corpus (one CSV, one row per test, source cited
on every row) and the scoring script (no dependencies). Both are available on request, email with your
outlet or institution. If you find an error in a corpus row we want to know about it: the row will be
corrected or excluded with a written reason, the same as the other 103.
Cite this work
Houghton, T. (2026). UBPS braking simulator: methodology and validation corpus. Tyreloop,
Te Puke, New Zealand. 974 catalogued tests from 218 published sources; 848 of 871 testable benchmarks
within tolerance (97.4%), mean absolute error 6.5%.
Contact and credit
The research, corpus and engine are the independent work of Taylor Houghton,
developed under Tyreloop and presented here by Tyre Dispatch.
Media enquiries: contact form, answered directly.
Common questionsFAQ
Did you run these braking tests yourselves?
No, and that is the point. The corpus is built from published instrumented tests by independent
organisations. We collect, standardise and cite; the measurements belong to the testers.
Why do some benchmarks fail?
23 of the 871 sit outside tolerance: a handful of exceptional snow performers, some US
performance-car dry tests, and low-speed studded-ice runs. They stay in the corpus and in the score.
Removing awkward rows is how validation numbers stop meaning anything.
What happened to the excluded tests?
103 entries are marked untestable: a missing water depth, an unverifiable figure, or a source that
published a range too wide to score. Each carries a written reason in the corpus file.
See what your own tyres would do
The simulator runs every factor on this page live: your speed, your tread, your grade, your weather.
Manager, Tyre Dispatch · Warehouse Manager, Traction Tyres
Petitioner to Parliament · tyre labellingTyrewise launch videoTe Puke, Bay of Plenty workshop
Taylor Houghton has spent his working life in the tyre trade, beginning in his father's tyre business at the age of five and handling import containers through his school years. He developed Tyre Dispatch, the retail arm of Traction Tyres, and now manages it alongside the Traction Tyres warehouse in Te Puke, with responsibility for buying and brand selection. He holds no ownership interest in either business.
Both brands the company distributes exclusively in New Zealand, Anchee and Predator, were his selections, each assessed on factory accreditation, compound composition and independent test data rather than name recognition. More than 100,000 Anchee tyres have since been sold here, and both brands hold five star review averages.
The guides, calculators and tools on this site are his own work, as is the workshop dataset behind them, which covers 1,700+ documented tyre inspections and 45,000+ photographs and represents more than a thousand hours, the majority outside business hours. He is the petitioner to Parliament for mandatory tyre performance labelling and minimum wet-grip standards in New Zealand, and appeared in the launch video for Tyrewise, the national tyre stewardship scheme.
Photography, measurements and advice on this site are first-hand; where a figure comes from elsewhere, it is attributed.
