How scoring works

The formula in full, and what it is quietly training you to do.

Prepare's score is a rating computed from your last hundred ranked attempts, not a total you accumulate. It can go down.

What one attempt records

Every answer you give in the route choice game writes one row: whether you were right, how long you took in milliseconds, which route you picked, which one was correct, how many routes were on screen, and whether the attempt counts toward the global leaderboard.

Correct means shortest. The answer is the route with the smallest measured length, taken from the lengths stored with the map. The main flag in the map file is a drawing-order label and does not indicate which route is shortest.

The first three challenges in a session are a warm-up. They show whether you were right and are not recorded.

The formula

score = (1000 ÷ speed) × accuracyMultiplier × 100 × recency

speed is your weighted mean response time in milliseconds, over correct attempts only; recency is explained below

The two inputs are not symmetrical. Speed enters linearly: halving your response time doubles the base. Accuracy enters through an exponential multiplier, which spans more than an order of magnitude.

The accuracy multiplier

multiplier = e^((accuracy − 50) × 0.035)

k = 0.035; accuracy is a percentage from 0 to 100

Fifty per cent gives a multiplier of exactly 1, which is what guessing returns on a two-route challenge. Above it the multiplier rises exponentially, and below it falls the same way.

Accuracy	Multiplier	What it means
100 %	5.75×	Every ranked answer right
90 %	4.06×	Strong: one mistake in ten
75 %	2.40×	Solid; roughly the ceiling of intuition on hard legs
60 %	1.42×	Better than guessing, but not by much
50 %	1.00×	Baseline: a coin flip on two routes
25 %	0.42×	Systematically wrong: you are reading something backwards
0 %	0.17×	Floor

Below 50 %

On two-route challenges, sustained accuracy under 50 % is worse than guessing. The multiplier falls steeply enough that a fast response time cannot compensate for it.

A worked example

Two runners, same number of attempts. One answers in 1.4 s at 72 % accuracy; the other answers in 0.9 s at 58 %.

Runner A   1400 ms, 72 %
  base       1000 / 1400            = 0.714
  multiplier e^((72 - 50) * 0.035)  = 2.16
  score      0.714 * 2.16 * 100     = 154.3

Runner B    900 ms, 58 %
  base       1000 / 900             = 1.111
  multiplier e^((58 - 50) * 0.035)  = 1.32
  score      1.111 * 1.32 * 100     = 147.0

Runner B answers 36 % faster and still scores lower.

The rolling window

The rating is computed from your 100 most recent ranked attempts: the most recent hundred, not the best hundred. All-time totals are kept separately and shown on your profile; they do not affect your rank.

A bad session pushes older, better attempts out of the window, so the rating describes recent form.

Time decay inside the window

Within those hundred attempts, each one carries a weight based on its age:

Age of the attempt	Weight	Effect
0 to 30 days	1.00	Full value
30 to 120 days	1.00 → 0.00	Linear fade over 90 days
120 days or older	0.00	Counts for nothing

Accuracy is the decay-weighted share of correct attempts; speed is the decay-weighted mean response time over the correct ones. Both are computed from the same rows.

A rating decays rather than freezing

The decay weights also feed a factor called recency: the share of your attempts' weight that is still alive, from 1 (everything inside 30 days) to 0 (everything older than 120 days). The score is multiplied by it. Weighting alone would change nothing, because the weight sits on both sides of an accuracy ratio and cancels, so recency is what makes a rating sink from day 30 and leave the board by day 120. Regular short sessions hold a rank; one long session every few months does not.

Speed is measured on correct answers only

A wrong answer costs accuracy, and its response time is discarded. Answering instantly when unsure therefore cannot lower your average response time; it only lowers your accuracy.

Which attempts are ranked

An attempt counts toward the global leaderboard only when all four of these hold:

Condition	Why
Official maps, not your own or community uploads	Everyone is scored on the same set of challenges
“All maps” selected, not one map	Practising one map you know well is training, not a comparable score
Multi-select off	Same reason: a hand-picked subset is not the common pool
Filters at their defaults, including 100 % route length	Filtering to the shortest, easiest legs would inflate accuracy

Anything else still plays, still reports whether you were right, and still adds to your all-time totals, but does not move your rank. While they are off, the game says Unranked above the challenge, and the i beside it explains why.

Reading the leaderboard

The leaderboard sorts by combined score by default, and can be re-sorted by accuracy alone or speed alone. There is also a country filter. The single-axis sorts show which half of your score is the weaker one.

Rank-change arrows appear only on the combined sort; accuracy and speed have no stored historical rank to compare against.

The leaderboard sorted by Overall with accuracy and speed — The board, with Accuracy, Speed and Overall.

Your all-time totals, which do not affect your rank.

What this is training

The two axes are multiplied rather than added because in a sprint they are not independent: time spent deciding at a junction is time not spent running, and a slow decision can cost more than the route it saves.

What follows from the shape of the formula:

Accuracy moves the score further than speed. Going from 65 % to 80 % multiplies the score by about 1.7×; matching that on speed alone would mean cutting your response time by roughly 40 %.
The clock runs until you answer. A leg you find yourself re-measuring is one to revisit in Analysis mode afterwards.
Analysis mode holds the map so you can zoom in and find the feature you misread, such as a fence, a passage or a set of steps.
Decay and the rolling window both reward regularity.Twenty attempts every other day hold a rating that two hundred once a month does not.

Compare your two-route and four-route accuracy

Two routes can be answered by instinct; four generally cannot. The gap between the two figures indicates how much of your route choice comes from reading the map.

The other modes score differently

The formula on this page is the route choice game's. The other modes have their own:

Route Finder scores a drawing by its length against the optimal path, with penalties for missing the endpoints or cutting through impassable terrain.
Route Navigator scores how much longer your walked route is than the shortest one, from 0 to 100, with 70 as a pass. Time is shown but does not enter the score.
Duel scores per round, live against an opponent: 100 points for a correct answer plus a speed bonus of up to 50 relative to the other correct answers in that round.