Sounds like a classic cellular automaton with a twist—like a Frost‑Pythagoras game. You’ll need to decide what constitutes a “frosty” neighbor. Maybe count adjacent cells that are already ice; if that count is between one and three you freeze the cell, otherwise it stays dry. The recursion can then be a simple depth‑first expansion from the seed, but you have to guard against infinite loops by marking visited cells or limiting recursion depth. Don’t forget the edge cells: you could wrap the board to make it toroidal, or simply treat out‑of‑bounds as dry—makes the crystal shape more jagged. Also, if you want realism, consider a stochastic component: a small probability that a cell fails to freeze even if the rule is satisfied, to mimic imperfect deposition. Once you have the rule, just iterate until no new cells change state; that’s a good termination condition. And if you’re curious about performance, try memoizing neighbor counts or using a queue instead of pure recursion—Python’s recursion depth is limited, but a simple loop will do the trick without the stack overflow drama.