Strategy · essays

Do these puzzles ever require guessing?

The honest answer is: no, and also sort of — and the difference between those two answers is the most interesting thing in the genre.

The "no"

A valid board in this family — Meowdoku, Queens, Queenzle, 1★ Star Battle, all of them — has exactly one solution, and that solution is always reachable by deduction alone. This is not marketing copy; it is a mathematical property. If a board had two solutions, no chain of logic could distinguish them, and any "solution" you reached would involve an actual coin-flip. Puzzle setters (and generators) therefore verify uniqueness before publishing, and uniqueness guarantees that from any correct position, at least one square is provably a cat or provably empty. Guessing is never required.

A one-step deduction: the green region has a single open square left, so its cat goes there. Every board guarantees some proof exists — it does not promise the proof is this short.

The "sort of"

Required and humanly findable are different things. Deductions come in sizes. "This region has one square left" is one step. "A cat here would suffocate the pink region" is two or three. But the guarantee above only promises that some proof exists — it says nothing about how long that proof is. On some boards, the shortest available proof for the next move is a what-if chain five or six forced moves deep: suppose a cat here, watch three regions collapse in sequence, reach a contradiction, conclude the square is empty.

Technically that is still pure logic. Practically, no human scans a 10×10 board testing hypotheticals six moves deep — a person in that position isn't deducing anymore, they're guessing and backtracking, which feels (and functionally is) the thing the puzzle promised you'd never need. That threshold — where the shortest proof outgrows what a person would realistically find — is what we call the human line.

A chain move: suppose a cat on the highlighted square, follow the forced consequences three or four regions deep, and reach a contradiction. Still pure logic — but past what a person would realistically scan.

Why endless modes cross it

Handcrafted levels almost never cross the line, because a human setter solves their own puzzle before shipping it — the proof-length filter is built into the process. Computer generators are different: they verify that a unique solution exists, but most don't measure how findable the path is. Meowdoku's endless mode is the notorious example — it will occasionally serve a board whose midgame requires a chain that would take a patient human many minutes of systematic trial to uncover. The board is fair in the mathematician's sense and unfair in every other sense.

This is worth knowing because of what it does to your confidence. The worst outcome isn't the hard board — it's concluding that you missed something simple, auditing a perfectly clean position for ten minutes, and quitting. Sometimes you didn't miss anything. Sometimes the board is just past the line.

How the solver handles it

MeowSolver's hint engine climbs the same ladder of techniques a person uses, from easiest to hardest. When it reaches the top — when a contradiction chain is genuinely the only move available — the hint is flagged as beyond the human line, and the engine tells you how many consecutive moves stay that way. You get an honest choice instead of false encouragement:

• Walk the chain. The hint draws the full hypothesize-and-collapse sequence on the board, step by step, so you can follow the proof rather than reconstruct it.
• Skip the trap. One tap places a guaranteed-correct cat — verified against the board's unique solution — and you resume normal solving on the other side of the rough patch.

Whether skipping "counts" is a matter of taste. Our view: the generator broke the social contract first.

What to do when a board fights back

1. Audit before escalating. Most "impossible" positions are an upstream mistake, not a hard board — the common mistakes guide covers the usual suspects.
2. Climb the ladder in order. Confinement and squeezes are missed far more often than they're absent. A position that "needs" a deep chain usually just needs a locked set you haven't spotted.
3. Then trust the diagnosis. If the position is clean and the simple techniques are genuinely silent, you've met the line. Import the board and let the solver either walk you through the chain or hand you the skip — both are more honest than pretending the board is something it isn't.

← Back to the strategy guide