Collapse — order from constraints
Every cell below begins holding every tile at once— a grid of pure possibility. The solver finds the cell it's most certain about, collapses it to one tile, and lets that choice ripple outward, deleting every neighbour that no longer fits. There is no picture stored anywhere. The pattern is discovered, tile by tile, from a single rule: edges must match.
This is Wave Function Collapse, the algorithm behind a lot of procedurally-generated game worlds. It borrows its name and its logic from physics: each cell is a superposition of states, and observing one (collapsing it) constrains everything entangled with it. The solver always collapses the lowest-entropycell first — the one with the fewest options left — because committing where you're most sure keeps contradictions rare.
When a contradiction does happen — a cell painted into a corner with no legal tile — the run simply restarts. That counter ticks up as you watch. It's a fair confession: this kind of constraint solving has no guarantee of finishing, and the honest fallback is to try again.
The other demos here measure entropy — Sediment weighs it, Crumbs predicts against it. This one runs the idea in reverse: entropy minimised on purpose until something coherent falls out. Same mathematics, pointed at making instead of reading. No model, no network, no stored image — just six tiles and one matching rule, resolving live in your browser.