Why does changing one starting number transform an Easy puzzle into an Evil one? The answer lies in constraint propagation, combinatorics, and the solving techniques your brain must deploy.
Pick up any sudoku book and you will find labels: Easy, Medium, Hard, Expert, Evil. But what actually determines these ratings? Is it simply the number of given digits? The position of those digits? Or something more mathematically subtle?
The answer is both simpler and stranger than most solvers expect. Sudoku difficulty is almost entirely determined by the solving techniques required — not by how many numbers are initially provided. A 22-clue puzzle can be trivially easy; a 30-clue puzzle can be fiendishly hard. Understanding why requires a quick tour of constraint mathematics.
Valid completed sudoku grids
Minimum clues for a unique solution (proven 2012)
Cells in a standard 9×9 grid
Possible arrangements of one row alone (362,880)
Those 6.67 sextillion completed grids might seem to suggest sudoku is infinitely varied. But from any single completed grid, you can derive many different puzzles simply by removing different subsets of clues. The challenge for puzzle designers is choosing which clues to remove to create a puzzle of a specific difficulty — and ensure it has exactly one solution.
Every sudoku starts with a set of constraints: each row, column, and 3×3 box must contain the digits 1–9 exactly once. When you fill in a cell, you eliminate that digit as a possibility from 20 other cells (8 in the same row, 8 in the same column, 8 in the same box, minus overlaps). This is constraint propagation.
Easy sudoku puzzles are those solvable purely through constraint propagation using two simple techniques:
A cell has only one possible digit left after eliminating all digits already present in its row, column, and box. No deduction required — just scan and fill. This is the most basic constraint-propagation step.
A digit can go in only one cell within a row, column, or box — even though that cell still has multiple candidates. You must scan each unit (row/column/box) for each digit to find these. Slightly harder than naked singles, still solvable mechanically.
A puzzle that can be fully solved using only these two techniques is rated Easy by virtually every rating system. The mathematics is straightforward: enough constraints are given that each step uniquely determines the next.
When naked and hidden singles are exhausted but the puzzle is not complete, you need more powerful techniques. These involve reasoning about groups of cells rather than individual cells.
| Technique | Difficulty Level | What It Does |
|---|---|---|
| Naked Pairs/Triples | Medium | Two/three cells in a unit share the same two/three candidates — eliminate those from the rest of the unit. |
| Hidden Pairs/Triples | Medium–Hard | Two/three candidates appear only in two/three cells within a unit — eliminate other candidates from those cells. |
| Pointing Pairs | Medium | A candidate in a box is confined to one row or column — eliminate it from the rest of that row/column outside the box. |
| X-Wing | Hard | A candidate forms a rectangle across two rows and two columns — powerful elimination across the full grid. |
| Swordfish | Expert | Three-row/column variant of X-Wing — more cells, more eliminations, much harder to spot. |
| XY-Wing | Expert | Three bivalue cells in a pivot-wing pattern — eliminates a candidate from cells that see both wings. |
| Forcing Chains | Evil | If digit X goes here, then Y must go there, then Z must be... — extended logical chains through the grid. |
| Trial and Error | N/A (bad practice) | Guessing — logically valid but considered poor form by serious solvers. |
Automated sudoku raters work by attempting to solve the puzzle using techniques in order of difficulty — simplest first, most complex last. The difficulty of the puzzle is determined by the hardest technique required to reach a solution.
The most widely respected open-source rating algorithm, used by many apps and websites, is the SudokuWiki solver developed by Andrew Stuart. It assigns numerical scores to each technique and sums them for a total difficulty score. A puzzle requiring only naked singles scores around 100; one requiring forcing chains might score 2,000+.
Some publishers use a simpler heuristic: count the number of given clues. But this is a poor proxy. Research by mathematicians at MIT has shown that clue count correlates only weakly with human-perceived difficulty. A 25-clue puzzle strategically designed to require X-Wings is harder than a 20-clue puzzle solvable by naked singles throughout.
One of the most elegant results in sudoku mathematics is the proof that you need at least 17 given clues to guarantee a puzzle with a unique solution. This was proven in 2012 by Gary McGuire, Bastian Tugemann, and Gilles Civario after an exhaustive computer search of all possible 16-clue configurations.
Their proof required checking approximately 5.5 billion possible puzzle configurations — a months-long computation. The result: no 16-clue puzzle exists with a unique solution. All 17-clue puzzles with unique solutions were catalogued; there are tens of thousands.
This is a remarkable result because it means difficulty ratings have a hard mathematical floor. No matter how cleverly a constructor places clues, if they use fewer than 17, the puzzle either has no solution or multiple solutions — it is not a valid sudoku.
You may have noticed that published sudoku puzzles often have a pleasing visual symmetry — clues appear in rotationally symmetric positions. This is not a mathematical requirement; it is an aesthetic convention. Setters typically impose 180-degree rotational symmetry on their given clues because it produces visually balanced, satisfying-looking grids.
Imposing symmetry actually makes the setter's job harder — they must find valid puzzles within the additional constraint that clue positions are symmetric. It slightly increases the average clue count needed for a unique solution, typically requiring 22–26 clues for symmetric medium-difficulty puzzles.
Occasionally a puzzle rated "Hard" contains one step that requires an advanced technique, while the rest is trivial. This can feel unfair to solvers — a single evil step buried in an otherwise easy puzzle. This is a known limitation of technique-based rating systems: they rate by worst-case technique, not by the overall experience arc.
The best puzzle constructors balance not just difficulty level but difficulty flow — a gradual escalation from simple steps at the start through increasingly complex deductions toward the end. This is craft, not algorithm.
Most apps rate difficulty by the solving techniques required rather than the number of given clues. Easy puzzles need only naked singles and hidden singles. Hard puzzles require X-wings, swordfish, or forcing chains.
There are approximately 6.67 × 10²¹ valid completed sudoku grids. The number of valid minimal puzzles is still an open research question.
Mathematically proven in 2012: you need at least 17 given clues to guarantee a unique solution. No valid 16-clue puzzle with a unique solution has ever been found.
An X-Wing occurs when a candidate digit appears in exactly two cells in each of two rows, and those cells align in the same two columns. This allows you to eliminate that digit from all other cells in those two columns.
Yes — every well-formed sudoku with a unique solution can theoretically be solved using logical deduction alone, though some puzzles require extremely advanced techniques like Aligned Pair Exclusion or pattern-based forcing chains.