Sudoku Book Difficulty Explained

For anyone compiling a sudoku book, or simply curious why one puzzle with 26 clues is trivial and another with 26 clues is brutal. Difficulty in sudoku is about which techniques a solve requires, not about how many numbers you start with.

ZillaKit team · Published September 2026

Open two sudoku books and you will find that a puzzle labelled "hard" in one is comfortably "medium" in the other. There is no governing body and no standard scale. But underneath the inconsistent labels there is a real, objective structure, and understanding it lets you grade a book sensibly rather than by feel.

Clue count is mostly a red herring

The intuition is that fewer starting numbers means a harder puzzle. It is weakly true and often wrong. A puzzle with 24 clues placed helpfully can fall apart in five minutes, while a puzzle with 30 clues arranged so that no cell can be resolved without a chain of inference can take an hour.

What clue count does control is whether a puzzle is valid at all. Below a certain number of clues, no sudoku can have a single unique solution. That floor is 17: there exist valid puzzles with 17 clues, and it has been proven that none exist with 16. So 17 is the hard mathematical boundary, not a difficulty rating. Almost no published puzzle sits anywhere near it, and most "expert" puzzles in books carry somewhere in the low to mid twenties.

Where the clues sit matters more than how many there are. Clues clustered in three boxes leave the rest of the grid to be derived slowly; clues spread evenly give the solver footholds everywhere.

Uniqueness is not optional

A legitimate sudoku has exactly one solution. This is the single most important property in the whole puzzle type, and it is where amateur generators fail.

If a puzzle has two or more valid solutions, a solver following correct logic will reach a point where two different numbers are equally defensible, and no amount of reasoning resolves it. They must guess. A reader who guesses, completes the grid, and then finds it does not match your answer key concludes the book is broken, and they are right.

Generating a unique puzzle is done by starting from a complete valid grid and removing numbers one at a time, checking after every removal that the puzzle still solves to exactly one answer, and putting the number back if it does not. It is not difficult to do correctly but it is easy to skip, because skipping it is much faster. The Sudoku Generator enforces uniqueness, which means every puzzle it produces is solvable by logic alone with no guessing required.

There is a useful consequence: because uniqueness is guaranteed, a solver can legitimately use it as a solving tool. Certain patterns can be eliminated purely on the grounds that allowing them would create two solutions. Purists dislike this style of reasoning, but it is valid in any properly generated puzzle.

Difficulty is defined by required techniques

The honest definition of difficulty is: what is the simplest technique that will get you all the way through? A puzzle is only as hard as the hardest step it forces.

Easy: singles only

A puzzle solvable entirely with naked singles, where a cell has only one possible number left, and hidden singles, where a number has only one possible cell left in a row, column or box. No note taking required. Most solvers finish these in five to ten minutes.

Medium: pairs and pointing

Now you need pencil marks. Naked pairs, where two cells in a unit share the same two candidates and therefore exclude those candidates elsewhere. Pointing pairs, where a candidate within a box is confined to one row or column and can therefore be eliminated from the rest of that row or column. These techniques are learnable in an afternoon and are the backbone of the medium grade.

Hard: triples, box-line reduction, X-wing

Naked and hidden triples extend the pair logic to three cells. Box-line reduction is pointing pairs in reverse. The X-wing is the first technique that reasons across the grid rather than within a single unit: when a candidate is restricted to the same two columns in two different rows, it can be eliminated from the rest of those columns. This is the step where casual solvers usually stop.

Expert and beyond: chains

Swordfish, XY-wings, colouring and forcing chains. These require tracking an implication through several cells, and many solvers reasonably regard them as the point at which sudoku stops being relaxing. Puzzles requiring them should be clearly labelled, because a reader who bought a book expecting a bus ride puzzle will be annoyed.

Grading a book

Once you accept that difficulty is technique based, grading becomes a mechanical question: run each generated puzzle through a solver that only permits a given set of techniques, and record the hardest one it needed. That is your grade. Any generator that grades by clue count alone will produce a book with wildly inconsistent difficulty inside each section, which is the complaint you see most often in reviews of cheap puzzle books.

How to order the puzzles

Whichever you choose, weight the book towards the easier end. Most buyers of a general sudoku book are medium level solvers, and a book that is two thirds expert will disappoint far more readers than a book that is two thirds easy.

A practical worked sequence

  1. Decide the book size, for example 120 puzzles: 40 easy, 40 medium, 30 hard, 10 expert.
  2. Generate each band and spot check a few by solving them yourself. If an "easy" puzzle makes you reach for pencil marks, the grading is off.
  3. Lay out one puzzle per page for a large format book, or two per page for a compact one, using the Paperback Typesetter.
  4. Number every puzzle, and repeat that number on its answer grid.
  5. Put the full solution grids at the back, six or nine to a page, in order.
  6. Assemble front matter, sections and keys with Merge PDF into a single interior file.

A one page explanation of the basic techniques at the front adds real value and costs you almost nothing. Readers who learn hidden singles from your book tend to keep it.

Honest limits

Technique based grading is objective but it is not the same as perceived difficulty. A solver who has never learned the X-wing will find a hard puzzle impossible rather than hard. A solver who knows it well may find the same puzzle easier than a medium one that requires a lot of tedious bookkeeping. Grading captures the logical ceiling, not the experience.

Grid layout also nudges perception. A puzzle with symmetric clue placement feels more elegant and, for many solvers, slightly easier, even when the required techniques are identical. Symmetry is an aesthetic convention rather than a difficulty property, but it is worth keeping because readers notice its absence.

Lastly, do not stretch your labels to make a book sound more impressive. Calling a medium puzzle "diabolical" wins nothing. Readers who finish it in eight minutes will simply trust the rest of your labels less.

Tools mentioned

Related guides