Hangar Rating
How the Hangar Rating works
The Hangar Rating is a chess-style Elo rating for car knowledge. You start at 1500. Every answer you give the first time you see a question moves it, by how surprising the result was.
Last updated:
The formula
E = g + (1 − g) / (1 + e^(d − (R − 1500) / 173.72)) R′ = R + 173.72 × max(0.05, 1 / (1 + 0.1 n)) × (result − E)
Each answer is a small match between you and the question. We work out the chance you had, then move your rating by the difference between what happened and what was expected: a hard question answered right moves you a long way up, an easy one answered wrong a long way down.
The scale is chess's: 400 points is ten to one. Your rating moves fast at first and slows as answers build up: a step is 174 points × 1 / (1 + n / 10), and never less than 8.7 points. The hundredth answer moves you about a tenth of what the first did, and a settled player moves steadily without being thrown around by one round.
Speed plays no part. Devices and networks differ, and fast fingers are not knowledge.
Where a question's difficulty comes from
A question starts from its writer's estimate: easy, medium or hard. Then it learns: when people who are expected to get it right miss it, it becomes harder; when they get it, easier. It learns slowly, and more from settled players than from newcomers, so a batch of new accounts answering badly on purpose has little effect on it. Guests do not move it at all.
What counts, and what never does
Only the first time you see a fact counts, in a daily, practice or extended round. Your rating changes when the round closes.
- Never: a question you have seen before, or its twin in another game.
- Never: duels, custom rounds, links to someone else's round, archive rounds, chapters of the Road Book or relaxed rounds. Someone could have read those answers first, or the round is teaching you.
- Never: an answer faster than anyone can read, or any of a round flagged for a person to review.
- Walking away does not help: what you answered still counts, and the question you left open counts as wrong.
Does it work?
We tested the formula on thousands of simulated players of known skill, on the same code the game runs, and the script is part of our source. After 100 answers, a rating was on average within about 70 points of the player's true skill, and the order of players matched their true order very closely (rank correlation 0.95). After 500 answers it was within about 40 points (0.985).
Ratings did not drift: a panel of players whose skill never changed moved by less than 10 points over 3,000 answers, with new players arriving throughout. That is why we show a provisional rating below 100 answers, and rank nobody until then.
A simulation only proves the formula on its own assumptions. Real people learn, and real questions do not all behave alike. We will check it against real answers as they build up.
What it can and cannot catch
We catch answers faster than a person can read, and hand those rounds to a person. We stop duels and shared rounds counting. No answer leaves the server before you commit to yours.
We cannot see someone looking answers up in another tab at a normal pace, or a friend who played today's daily first. A second account could read a daily too. So a high place on a public ranking is a strong sign, not a proof. If you think a result looks wrong, write to [email protected].
What is public
Your pseudonym, rating, tier, peak, number of rounds and this month's gain, once you have 100 answers. Your first name, email and account are never shown. Other boards can also show the country and club you chose, and which days you played in the last week. You start with a generated pseudonym such as Driver-3A419A; if you choose your own, please do not use your real name.
You can hide from every ranking at any time in Progress, in the app, and deleting your account removes you. Accounts are for people of 16 or older, and guests are never ranked. The privacy policy has the details.
The tiers
A tier says what a rating means: the chance of getting a four-option question right at the start of that band.
| Tier | From | Easy | Medium | Hard |
|---|---|---|---|---|
| RookieStarting out. Hard questions are close to a guess. | <1300 | 49% | 36% | 30% |
| Club DriverKnows the road. Gets most easy questions. | 1300 | 60% | 43% | 33% |
| Gentleman DriverA real enthusiast. Easy questions are safe, medium ones more often than not. | 1500 | 80% | 63% | 45% |
| ProSerious knowledge. Beats the odds on the hardest questions. | 1700 | 92% | 82% | 65% |
| Works DriverRare. Hard questions fall about five times in six. | 1900 | 97% | 93% | 84% |
| Grand MasterAlmost nothing gets past them. | 2100 | 99% | 98% | 94% |
Questions
Is the Hangar Rating the same as chess Elo?
It uses the same scale, where 400 points is ten to one, and the same idea of moving by surprise. The opponent is the question rather than a person, and the question's difficulty is learned from answers. It also allows for guessing, which chess does not need.
Why did my rating go down after a good round?
Because the questions were easy for your rating. At 1900, a right answer to an easy question moves you very little, and a wrong one a long way.
Does my rating fall if I stop playing?
No. There is no decay. You leave the rankings after 90 days without a rated answer and return with your next one, and for a while your rating moves faster, in case you have changed.
Why is my number different for each marque?
Each marque has its own rating from its own questions, as well as the overall one, because knowing Ferrari says little about knowing Bentley. A marque's ranking opens at 150 verified questions.
Why can I run out of rated questions?
Because only a first sight counts. Once you have seen a marque's whole bank, your rating there rests until new questions arrive.