A tyre on ice · Zero-Trust Physics

Whose measurements these are

Schleinin, Savitski, Jimenez, Sandu and Ivanov measured one tyre on ice 252 times, and published every run.

The tests were made in April 2016 on the Terramechanics Rig of the Advanced Vehicle Dynamics Laboratory at Virginia Tech, for the European project EVE, and the authors published the raw recordings of all 252 runs under a licence that lets anyone build on them. Three loads, three pressures and three repeats of each are what made it possible to hold whole conditions back from a model.

The authors: Dmitrij Schleinin, Dzmitrij Savitski, Emilio Jimenez, Corina Sandu and Valentin Ivanov
Where the tests were made: The Advanced Vehicle Dynamics Laboratory at Virginia Polytechnic Institute and State University, on its Terramechanics Rig, in April 2016
Made for: The European project EVE, European Commission grant 645736, Horizon 2020
The dataset: Dataset of tyre tests on ice, Parts 1 to 3. Zenodo, 2017. Licensed CC BY 4.0.
Where it is published: Part 1, doi.org/10.5281/zenodo.1009017. Part 2, doi.org/10.5281/zenodo.1009023. Part 3, doi.org/10.5281/zenodo.1009025.

What happened

How much grip does a tyre have on ice?

In April 2016 a rig at Virginia Tech rolled one tyre along six metres of prepared ice, 252 times, to find out.

The test, drawn and not to scale. The rig presses the tyre onto the ice with a set load and moves it along at a creep. The wheel is turned a little faster or slower than rolling, and the ice pushes back: that push is the grip.

The rig holds a wheel over a track of ice, presses it down with a set load and moves it along at a creep, about six centimetres a second. Then it turns the wheel a little faster or a little slower than rolling would, so that the tyre slips, and measures how hard the ice pushes back.

Each run stepped through four amounts of slip. The runs covered three loads and three tyre pressures, driving and braking, three times each. A further set added a film of water and a steered wheel, and is not part of this page.

What came out is a number most drivers have felt and few have seen written down. On this ice, at this crawl, the tyre could push with a little over a fifth of the weight resting on it.

What the model did

The model answered for ninety of their runs it had never seen, and met every mark.

Three constants were fitted to seventy-two of the runs. Ninety were held back: every run at the heaviest load, heavier than anything the model had been fitted on, and every run at the middle tyre pressure. The passing grade was fixed before any of them was opened.

On the held-back runs the measured grip averaged 0.218 of the load, and the model’s answers were typically within 0.034 of what was measured. All five marks were met.

What does the model know, then? Its description is plain about it. The ice itself changed from one session to the next by about as much as the model’s error, so any model that is not told the session lands in the same place, and a flat average of the opened runs comes almost as close.

Met

One square for each run.

Where it stops

It speaks for this tyre, on that rig’s ice, in that fortnight, at a creep. It says nothing of a tyre on a road.

What it was tested for

One tyre, on one rig and one track of ice, in one campaign, at a creep of 0.06 metres a second.
Slips of 5 to 20 percent, driving and braking.
A load above every fitted load, and a pressure between the two fitted pressures.

Past that, and what its own description says of itself

Slips under 5 percent, and the rise of the force from rest, are outside the model.
In the first five seconds after a change of slip, and after twenty-one, it stops and says why.
Nothing here tests another ice temperature, another speed, another tyre or another laboratory. The files record no ice temperature and no carriage speed, and the tyre’s make is not published.
It cannot say which way the ice of a given session will sit. Its description says to carry about ±0.03 on the ratio for the session.
The friction of ice changes with sliding speed, so its figure is not one for a tyre on ice at road speed.

For the reader who wants every number

In the verifier’s own words.

Everything above is this page’s telling. What follows is the verifier’s own, exact and whole: four parts of the description of the model materials/tire_eve_r19_ice, as read on the morning of 11 October 2026. Where the verifier says card it means a model, where it says line it means a pass mark, and where it says refuses it means the model stops there and says why.

The real thingobject

One 235/55 R19 all-season sport-utility tyre of the European project EVE, rolled on a prepared ice track by the Terramechanics Rig of the Advanced Vehicle Dynamics Laboratory at Virginia Tech in April 2016, at a creep of about 0.06 m/s: the ratio of its longitudinal force to its normal load at a held slip, driven and retarded.

The passing grade, fixed firstpass_line

Fixed before any held-back run was opened (pre-registration SHA-256 84a380a5c948cd741ac91845fbfeac7ed1b60ed3c6a01855bcb99c996e7782c1), on the 60 held-back cells, each the mean over three repeats of the last nine seconds of a slip stage at 5, 6, 7, 12, 15 or 20 percent slip: root-mean-square error of the friction ratio at most 0.045 over all 60 cells, over the 24 cells at 2.5 bar and over the 36 cells at 6.5 kN, and over the 60 cells smaller than the error of a ratio of 0.1 that uses no measurement of this tyre and smaller than the error of the opened runs' mean with no slip term. That mean is the geometric mean of each mode, least squares on the logarithm, as the fit is; the arithmetic mean would have given 0.038, and the line is met either way.

The resultresult

Compared with the 90 held-back runs on 2026-10-09, and all five lines fixed beforehand were met. Root-mean-square error of the friction ratio: 0.034 over the 60 held-back cells against a line of 0.045, 0.040 over the 24 cells at 2.5 bar and 0.029 over the 36 cells at 6.5 kN. A ratio of 0.1 that uses no measurement of this tyre gave 0.122, and the opened runs' geometric mean with no slip term (least squares on the logarithm) gave 0.036. The mean measured ratio was 0.218, and the card read high by 0.019 on average. Two held-back conditions taken alone were above 0.045, traction at 2.5 bar at 3.5 kN and at 5 kN, with errors of opposite sign; no line was set on a single condition. One held-back run, number 153, is a faulty record: it rolled for about 90 s like the others, and its channels are on other scales. Its vertical force reads 1,849 N at a nominal 3,500 N, and its wheel angle spans 0 to 11.75 where every other file spans 0 to 360. It was scored with the rest, as the protocol says; without it the three errors are 0.031, 0.034 and 0.029, and those are the figures that describe the tyre. The card read high on all six conditions at 6.5 kN, by 0.023 on average. On the 48 opened cells the error was 0.030. The binary that made the scored runs (SHA-256 7a2f3cc8e75aadbbdaf4f74289803ddf1149dbe882d94cadc96f668061b943c7) was overwritten by a later build; the later build returns all 108 saved answers byte for byte. A second scoring of 2026-10-09, on the same machine and on the scorer's own reduction of the raw files, found the same verdicts: evidence/checks/second_party/V-33/.

Its limitslimits

The three constants are fitted to this tyre's own runs; nothing is derived from the rubber or the ice. The held-back runs at 2.5 bar test interpolation in pressure and those at 6.5 kN test extrapolation in load, on the same rig, the same ice and in the same campaign as the fitted runs: nothing here tests another ice temperature, another speed, another tyre or another laboratory. The files record no ice temperature, no carriage speed and no rolling radius, and the tyre's make is not published. The load and the pressure do not enter the friction ratio: on the opened runs a ratio that varied with them predicted a left-out condition worse than one that did not, and the differences between conditions, about 0.03, were three times the scatter between repeats, so part of what a condition shows is the state of the ice on the day. Slips under 5 percent and the rise of the force from rest are outside the card. The speed is a creep of 0.06 m/s: the friction of ice changes with sliding speed, and the card says nothing of a tyre on a road. The error is between conditions and not between repeats: a condition sits 0.01 to 0.06 from the card as a block, and the card cannot say which way the ice of a given session will sit. The error follows the test block, not the load or the pressure: the 18 condition means of the measured ratio, in six blocks, run from 0.185 to 0.287, and the standard deviation between them is 0.029, against 0.005 to 0.011 between repeats. The ice changed between sessions and within the first block, where the ratio rose by 0.010 to 0.024 from the first repeat to the third. Any model that does not know the block lands near 0.03, and the card is at that floor. A user should carry about ±0.03 (one standard deviation) on the ratio for the session. The figure of 0.23 to 0.24 is this tyre on that rig's ice in that fortnight at 0.06 m/s; it is not a figure for a tyre on ice at a stated temperature or at road speed. The slip term is real on the held-back runs and small: the exponent fitted to them alone is -0.107 ± 0.014, where the card holds -0.133, and the term is worth 0.002 to 0.003 of the 0.03 error. The longitudinal force at rest before a run, about -62 N with a standard deviation of 49 N, is not subtracted from the measured forces. The ledger of the three works closes for any constant, because each work is proportional to the force: it can catch a corrupted step, and it cannot catch a wrong constant, a wrong exponent or a wrong ice. A run whose end in decimal is exactly the track's 6 m lands every step: 1,000 steps of 0.1 s at 0.06 m/s, rolled as runs of 16 s each started from the distance the last returned, land 1,000 and read 6.000000000000005 m. Through engine 0.5.0 the last was refused, because the distance is summed in binary, and through engine 0.6.0 a start handed back from a run of small steps could be refused on one step to the end that small steps reached; the allowance at the end of the track is now the same at every step. The time after a slip change was not part of the registered comparison. Compared on 2026-10-10 by the protocol's cells and lines, with the change placed where the wheel speed steps, every one-second window from 5 s to 20 s after a change met all five lines, and every one before 5 s failed one; the card refuses a step before 5 s and after 21 s (evidence/research/tire_eve_235_55r19_ice/RESULTS.md). Engine 0.7.0 changed every answer of this card: a start states slip_held_s, the state returns it, and the step ceiling is 16 s. One party wrote the card and ran it. It was scored twice, the second time by a second scorer, on one machine.

Hand this one to your agent.

https://mcp.zerotrustphysics.com/mcp

Open to any agent.

Something to paste

Connect to https://mcp.zerotrustphysics.com/mcp and read the model materials/tire_eve_r19_ice. Then tell me what it is, and why it might matter to me.

Give it this page too.

What it will find

This model’s own description: what it takes in, where it stops, the runs it was held to and how it did, in its own words.

If these are your measurements.

The model’s account of your tyre is there for any agent to read, in its own words. It could hold whole conditions back only because you measured a full grid and published every run. John Kruze would be glad to hear what it got wrong about the real tyre and the real ice.

Give the address and this page to your own agent first. Then write or call.

Write: [email protected]
Call: 928-264-7883