How the eight differ
All of them produce a perfect maze: every cell reachable, exactly one route between any two cells, no loops. Every one is a spanning tree of the grid. What differs is which spanning tree, and the character that gives the result. These numbers are for the current size and seed, so they update as you change them — press the row to load that algorithm.
Why the dead-end count is the number that matters
A maze that is 30% dead ends feels completely different from one that is 10% dead ends. The recursive backtracker carves as far as it can before backing up, so it produces long snaking corridors and relatively few junctions — it is the one that feels like a proper labyrinth, and it is what most games use. Randomized Prim grows outward from a single point like frost, which produces many short stubs and a maze that feels busy but shallow. Both are correct; neither is better in the abstract.
Uniform is a specific thing, and only two of these are it
Wilson’s algorithm and Aldous–Broder both sample uniformly from all spanning trees of the grid — every possible perfect maze equally likely. Nothing else here does. The backtracker is heavily biased toward long corridors, binary tree towards one diagonal, sidewinder towards horizontal runs. If you need a maze with no algorithmic fingerprint — for a study, a puzzle set, a benchmark — those two are the only correct choices, and Wilson is the one to use because Aldous–Broder spends most of its time wandering over cells it has already visited.
The two fast ones give themselves away
Load Binary tree and look at the top row and the right column: both are a single uninterrupted corridor, every time, on every maze it will ever produce. That is not a bug — the algorithm makes one independent decision per cell, carving either north or east, so the cells on the top edge can only go east and the ones on the right edge can only go north. It runs in one pass with no memory at all, which is why it exists, and the diagonal texture is the price. Sidewinder is the same idea done one row at a time; it fixes the right column but the top row is still a single corridor.
Braiding
Turn up Braid and dead ends get opened into loops. The maze stops being a tree, so “the” solution becomes “a” solution and there are now many routes. This is what most real mazes — hedge mazes, game levels — actually are, because a perfect maze is solvable by simply keeping one hand on the wall, and a braided one is not.