Different puzzles have different validation logic. So the logic varies from puzzle to puzzle. Generally we're working backwards from a completed puzzle.