What makes a sudoku puzzle hard?

Counting the givens tells you less than you would think

A grid with thirty givens can be trivial and a grid with thirty givens can be genuinely difficult, and the difference has nothing to do with the count. What matters is where they sit. Twenty-eight digits clustered into three boxes leave the rest of the grid barely constrained; the same twenty-eight spread evenly cascade, each one narrowing several others. Two puzzles with identical counts can be a five-minute solve and a forty-minute one.

There is a genuine floor, and it is worth knowing because it is widely misquoted: a classic 9×9 sudoku needs at least seventeen givens to have a unique solution. That was an open question for years and was settled by exhaustive computer search in 2012. It is a statement about what is possible, not a difficulty rating — plenty of seventeen-given puzzles are more tedious than hard, and plenty of twenty-five-given puzzles are harder than any of them.

Difficulty is the hardest technique the grid forces on you

The measure that actually tracks how a puzzle feels is the set of solving techniques it requires. Puzzle setters grade by the most advanced move you cannot avoid.

At the easy end there are naked singles — a cell with only one candidate left — and hidden singles, where a digit has only one place it can go in a row, column or box. A puzzle solvable by these alone is an easy puzzle regardless of how few givens it has, because you never have to hold more than one cell in your head.

Medium puzzles add pairs and triples: two cells in a unit that between them can only hold the same two digits, which eliminates those digits everywhere else in the unit. This is the first technique that requires reasoning about cells you are not filling in, and it is where most casual solvers stall.

Hard puzzles need pointing and claiming — where a candidate confined to one row within a box lets you eliminate it from the rest of that row — and beyond that, chains: X-wings, swordfish, forcing chains, patterns spanning the whole grid. A puzzle graded "evil" is usually one that cannot be finished without at least one such move, which is why the jump from hard to evil feels like a wall rather than a step.

Why one solution is the property worth insisting on

A sudoku with two solutions is not a hard sudoku. It is a broken one. Somewhere in the grid sits a pair of cells that can be filled either way, and a solver who reasons perfectly will still reach a point where logic runs out and only guessing remains. There is no way to tell from the page that this has happened — the grid looks exactly like a difficult one, so the natural conclusion is that you are not good enough at sudoku.

This is the commonest defect in free printable sudoku, because generating a grid is easy and verifying one is not. Producing a filled grid and removing digits takes a moment; proving that what remains has exactly one completion means solving it exhaustively and confirming no second solution exists. The check costs real work, so it is often skipped, and the failure is invisible until somebody hits it.

It is worth checking whether a source states this. The sudoku generator here verifies uniqueness for every puzzle it prints, and its answer key is the solution that was verified rather than one solution among several.

Picking a level that fits

For a classroom, the useful question is not which level is hardest but which one ends. A puzzle that stalls half the room ten minutes in produces nothing but frustration, and the natural response — telling them to guess — trains exactly the wrong habit, because guessing works and then quietly stops working. Easy and medium grids, solvable by singles and pairs, keep the reasoning visible and finishable.

For a solver working alone, the level to choose is the one just above where you are comfortable. Techniques are learned by needing them: the first pair you spot yourself is worth more than a dozen puzzles finished without one. And a puzzle that turns out to need a chain is a good reason to stop and look at the answer key, not a failure — reading the move you missed is how the next one gets found.

Two practical notes on paper. Print at a size that leaves room for pencil marks in the corners of each cell, because from medium upward you will need somewhere to write candidates, and a grid sized to fit two-up on a page usually does not have it. And print the answer key on a separate sheet rather than the same page, or it will be read before it is wanted.

The honest limits

Difficulty grading is not standardised. One publisher's "hard" is another's "medium", and the same grid can be graded differently by two solvers depending on which technique each happens to spot first. A label is a rough signal about which techniques are likely needed, not a measurement, and it is not comparable across sources.

Nor does verified uniqueness make a puzzle good. It rules out the specific defect of being unsolvable by logic; it says nothing about whether the grid is interesting, whether its givens are pleasingly symmetric, or whether the solve has a satisfying shape. Uniqueness is the floor a puzzle has to clear before any of the rest is worth discussing.