The maze
Every race gets a brand new maze. It's carved with a recursive backtracker, the classic way to build a maze, and then about 7% more doors get knocked open so there's more than one way through. Those extra doors are why leads change: a racer that guesses wrong early can still catch up.
There are three sizes. Quick is 11 by 15 cells, Normal is 15 by 21, and Epic is 19 by 27.
The racers
Every racer follows the same rule, a randomized depth-first search. At each fork it picks a random direction it hasn't tried. At a dead end it backs up to the last fork that still has an untried path. No racer is smarter or faster than another. The only difference between them is luck.
Each racer gets its own random seed from your browser's crypto.getRandomValues, the same secure source browsers use to make encryption keys. Where a name sits in your list doesn't change anything.
Pacing
The whole race is worked out in an instant before it starts, so the page knows exactly when the winner gets out. Then it's replayed at a speed that fits the length you picked, with the last few steps in slow motion. Races that need everyone to finish, like turn order and teams, fast-forward after the first racer gets out.
Ties
When racers reach the exit on the same step, a secure coin flip puts them in order and the result says PHOTO FINISH. They're common with big lists on small mazes and rare the other way around. Here's how often first place came down to a coin flip in our tests:
| Maze | 2 names | 12 names | 100 names |
|---|---|---|---|
| Quick (11 x 15) | 2.9% | 15.1% | 45.5% |
| Normal (15 x 21) | 0.6% | 5.5% | 20.7% |
| Epic (19 x 27) | 0.3% | 2.6% | 10.9% |
Extra entries
A name listed more than once gets one racer per copy. Its best racer sets its place, and its other racers leave the maze at the same moment. So extra entries raise the odds, but no name can win twice in one race or land on two teams.
How we tested fairness
We simulated 12,000 races with 12 names on the Normal maze and counted how often each spot on the list won. A perfectly even split would be 1,000 each:
| List spot | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Wins | 1005 | 979 | 1000 | 1028 | 950 | 1063 | 973 | 1017 | 972 | 1007 | 1017 | 989 |
That spread is what you'd expect from pure chance. A chi-square test scored it 10.0, well under the 31.3 that would point to a real bias.
The same checks run automatically every time the code changes: winners with 2, 4 and 12 names, every finishing place in a full order, and extra entries (a name with 3 of 6 entries wins about half the time, as it should).
Who makes this
Rando Race is made by Algo Arcade, a channel where sorting algorithms, maze solvers and route planners race each other with chiptune sound. These name races grew out of the channel's maze races. Questions or ideas? Email hello@randorace.com.