Eight boundaries that do notchange between runs.
We derived tier boundaries from the population distribution that published research expects (IAM RoadSmart, Roadcraft, GDE, FIA, ESRA) — not from who takes the test with us. The boundaries are frozen: a re-test will not change where they sit. What changes is only where you sit.
T1 and T8 are rare not because we want them to be, but because published research describes the population exactly like this. The boundaries will not change even if we add another 5,000 people to the dataset — that would break the retroactive comparability of results.
A weighted sum, noblack boxes.
We compute the score for both axes as a weighted sum of answers to questions from our database. No neural networks, no ML, no "proprietary AI".
Raw scoring rule (simplified)
S = Σᵢ (wᵢ · aᵢ) / Σᵢ wᵢ
P = Σⱼ (wⱼ · aⱼ) / Σⱼ wⱼ
tier = fixed band from the table above
wᵢ, wⱼ are the sub-dimension weights from the current rules version. aᵢ, aⱼ is the sub-dimension score 0–100 after modifiers. Band boundaries never change retroactively.
Modifiers
Modifiers adjust the score for context the questions themselves do not capture: accidents and incidents (−5 each, at most −15), two or more completed courses (+5), high annual mileage over a longer driving history (+5), experience with several transmission types (+3). They are part of the rules version the result was computed against.
OE (open-ended) questions
Text scenario questions do not count towards your score in Beta. We ask them and store your answer, but no model evaluates it and it has no effect on your tier. Rubric-based evaluation — a scoring framework describing what a good answer should contain — is ready; we will switch it on once we can substantiate how accurate it is.
What we do notknow and where we are unsure.
No measurement is perfect. Here are the things we admit:
Known DriverRank limits
- The test measures knowledge and reactions, not real behaviour in traffic. A T7 driver can still be dangerous when tired or impaired.
- Self-report questions (driving frequency, annual mileage, number of fines) are prone to social desirability bias — we acknowledge it, we do not eliminate it.
- OE scenario questions do not enter your score — in Beta we collect them but do not evaluate them. The tier rests on the MCQ questions, self-report and measurements.
- The population norm is SK / CZ / AT / DE. For drivers outside this region the norm is less relevant.
- We do not measure driving — we measure knowledge about driving. A track-day racer can score higher Performance than a defensive-driving instructor, even though the instructor is better on the road.
- A tier is a band, not a point. A score close to a band boundary can tip one tier either way on a re-test — that is not an error, that is the resolution of the test.
- The tier is an automated calculation, but not a decision with legal effect (Art. 22 GDPR does not apply). Nobody grants or withdraws insurance, a job or a licence on the basis of it.
We do not list these limits because we are pessimists. We list them because without them the interpretation of the result would be misleading.
Weversion the scoring rules.
Whenever we change a question weight, a tier boundary or a modifier, a new version of the scoring rules is created (for example 2026-05-v1). Your result is always stored with the version it was computed against — we never re-shuffle old results retroactively.
Current rules version: 2026-05-v1. Every result carries the version it was computed against — you can find it in your history.
What we version
- Tier boundaries on both axes
- Sub-dimension weights
wᵢ, wⱼ - Modifiers (additions, removals, thresholds)
- Reaction-test bands