How a Puzzle Proves It Has Exactly One Solution

When you solve a sudoku, you are relying on a promise nobody prints on the page: that the grid has exactly one valid completion. Without it, half the standard techniques are invalid, and the puzzle stops being a deduction problem and becomes a search problem.

Uniqueness is the contract. It is also the single most common thing an amateur-made logic puzzle gets wrong.

Why it matters more than it sounds

Consider a sudoku with two valid solutions differing in four cells. Every technique that reasons about what must be true fails on those four cells, because nothing must be true there. A solver applying correct logic will grind to a halt and conclude they have made an error, when the error is in the puzzle.

Worse, uniqueness is itself a solving technique. Advanced sudoku solvers use patterns that are only valid because the grid is guaranteed unique - if two candidates in a rectangle could be swapped without breaking any constraint, then neither can be right, because a unique puzzle cannot contain that ambiguity. That deduction is invalid the moment uniqueness is not guaranteed.

Uniqueness is not a nice property of a good puzzle. It is the axiom the solver is reasoning from.

How setters actually check

Not by solving it. A human solving the puzzle once proves a solution exists; it proves nothing about whether another one does.

The standard method is an exhaustive solver: a program that explores every consistent assignment and counts how many complete ones it finds. If the count is one, the puzzle is unique. If it is zero, the puzzle is contradictory. If it is more than one, the puzzle needs another clue.

For sudoku and its relatives this is fast, because constraint propagation prunes the search tree aggressively. For puzzles with weaker constraints - some logic grid designs, some nonograms - the search space is larger, but still tractable at puzzle sizes.

The deadly pattern

In sudoku, one specific configuration destroys uniqueness reliably. If four cells sit at the corners of a rectangle, span exactly two boxes, and all four contain only the same two candidate digits, then those digits can be swapped diagonally to produce a second valid solution.

This is called a deadly pattern or unavoidable set, and setters check for it explicitly. It is the reason a sudoku cannot be arbitrarily stripped of clues: some clue configurations look sparse and elegant while quietly containing a rectangle that can flip.

The mathematics here is well studied. It has been proved by exhaustive computer search that no valid sudoku puzzle with 16 clues exists - the minimum is 17 - which is the kind of result you can only reach by taking uniqueness seriously enough to enumerate every case.

What to do when a puzzle is not unique

Add a clue, and add the right one. The instinct is to add a clue in the ambiguous region, which works, but often over-constrains the puzzle and makes it easier than intended.

The better move is to find the minimal addition: test each candidate clue, keep the one that restores uniqueness while leaving the solve path longest. This is the same removal-and-test loop that produces good logic grid puzzles, run in reverse.

The crossword exception

Crosswords are the odd family out. A crossword grid has exactly one intended fill, but nothing about the structure enforces it - two different words could in principle satisfy every crossing. What prevents this is the clue list, not the grid.

This is why crossing two unfamiliar entries is such a serious fault. It is the crossword equivalent of a deadly pattern: a square where the solver cannot determine the letter from either direction, and the puzzle's uniqueness guarantee quietly evaporates.

The general principle

Every deduction puzzle is, formally, a constraint satisfaction problem with the additional requirement that the solution set has size exactly one. That requirement is what makes the puzzle solvable by reasoning rather than by search, and it is what the solver is trusting when they write a digit in ink.

Setters who verify it produce puzzles people finish. Setters who do not produce puzzles people abandon while blaming themselves, which is the worst outcome available in this craft.

Background on the mathematics is easy to find: the sudoku entry covers the enumeration results and the minimum-clue problem, and Nikoli, which still hand-makes its puzzles, is the standard reference for what verified-unique construction looks like as a practice rather than a proof.