Datasets:

LKForge
/

maze-solver-benchmark

license: mit
pretty_name: Maze Solver Benchmark (BFS shortest path by maze size)
language:
  - en
tags:
  - maze
  - bfs
  - shortest-path
  - recursive-backtracker
  - pathfinding
  - benchmark
task_categories:
  - other
size_categories:
  - n<1K
configs:
  - config_name: bfs_by_size
    data_files: data/maze-benchmark.json

Maze Solver Benchmark — BFS Shortest Path by Maze Size

By LK Forge: privacy-first browser tools & games, nothing to install, nothing uploaded.

Search effort and path length for the LK Forge Maze Solver, which generates a perfect maze with a recursive-backtracker and finds the unique shortest path with breadth-first search. 200 seeded mazes per size.

How the solver works: https://lkforge.com/tools/puzzles/maze-solver/how-the-maze-solver-works.html
Try it: https://lkforge.com/tools/puzzles/maze-solver/
Producer: LK Forge — client-side AI games, solvers, and tools.

Headline numbers (200 mazes each)

Maze	Open cells	Mean shortest path	Mean cells visited by BFS	% of open cells explored
12×12	287	131	185	64.5%
20×20	799	308	487	60.9%
30×30	1,799	607	1,071	59.5%
40×40	3,199	972	1,784	55.8%

In a perfect maze there is exactly one path between any two cells, so BFS returns the unique shortest route. It explores a fairly stable ~56–65% of all open cells before reaching the exit, and both path length and search effort scale roughly linearly with the number of open cells.

Files

data/maze-benchmark.json — per size: open cells, mean shortest-path length, mean cells visited, % explored.

Method

Engine: the recursive-backtracker generator and BFS solver copied verbatim from the shipped maze-solver.js (DOM stripped, Math.random → seeded RNG, size parametrized).
Cells visited = BFS dequeue count (machine-independent search effort).

Reproduce

node scripts/benchmark-maze.mjs

Citation

LK Forge (2026). Maze Solver Benchmark — BFS Shortest Path by Maze Size. https://lkforge.com/tools/puzzles/maze-solver/

Everything runs client-side, no login, no tracking.