Generate puzzles and save them to your library. Difficulty is measured after creation, not requested.
Click the gap between two islands to add a bridge; click again for a double bridge; once more to remove it.
No puzzle loaded. Pick one from the list.
bridges: 0, islands satisfied: 0/0
Everything you have generated, newest first.
Computational solvers are planned. Nothing here runs yet. Create and Play are fully functional today.
An exhaustive search over bridge assignments between islands, used as a correctness baseline. It will confirm solvability and report how many states it explored. It will be far too slow for large boards, which is exactly its job: a reference to measure the other two against.
The published constraint-programming / branch-and-cut approach used for benchmark Hashiwokakero instances. It will encode the degree, non-crossing and connectivity rules as constraints and solve them with a modern combinatorial solver. This is the target for both correctness and speed.
A learned model trained on many generated puzzles — the project's namesake idea. It will learn to propose bridges (or prune the search) from patterns in solved boards and be compared against the exact solvers. Not implemented yet; no model has been trained.