A maze can be thought of a series of binary choices, so you'd get a different distribution if you spawn particles from successful ones than from a random walk in 2D. However, I don't think this gets you any better a model of scientific discovery than existing statistical models of combinatorial innovations (https://www.newthingsunderthesun.com/pub/2elwp33m/release/2#...), and it may be the same as some others.