Skip to content

Learn · Difficulty

What actually makes a sudoku hard

Almost every site labels puzzles Easy / Medium / Hard / Expert. Almost none of them can tell you what the label means. We measured it instead.

The clue-count myth

The intuitive story is that fewer given digits means a harder puzzle. It is true at the extremes and false almost everywhere else. Here is the actual distribution across the 893,916 puzzles in our public source, grouped by the difficulty the source assigned:

Source tierPuzzlesMean cluesMedian10th–90th pctNeed chains
Easy100,00030.33025–360%
Medium352,64327.82826–300%
Hard321,59227.82826–3037%
Extreme119,68127.62825–30100%

Read the median column. Medium, Hard and Extreme all sit at 28 clues. You cannot tell a hard puzzle from an extreme one by counting digits. What changes is which technique the puzzle forces you to use — and that is what we grade.

The technique ladder

Human solving techniques form a ladder. Each rung is harder to spot than the last, and — importantly — each rung can be given a cost. We use the same scale as the classic Sudoku Explainer, because it is the only one that is publicly documented and reproducible:

TechniqueWeightWhat you do
Full House1.1the last empty cell in a unit
Naked Single2.0a cell with one candidate left
Hidden Single2.6a digit with one home in a unit
Locked Candidates3.0pointing / claiming eliminations
Naked / Hidden Pair3.0 / 3.4two digits locked into two cells
Naked / Hidden Triple3.6 / 4.0three digits, three cells
X-Wing4.2a rectangle across two lines
Swordfish4.6the three-line version
XY-Wing4.8a pivot and two pincers

A puzzle's score is the sum of the weights of the steps needed to finish it. But the sum is not the whole story: a puzzle that needs forty singles is easier than one that needs a single XY-Wing. So the tier is set by the hardest technique required, with clue count used only as a tie-breaker.

What the tiers actually contain

We solved a random sample from each source tier with the technique solver and recorded which techniques were required at least once. This is the table nobody publishes:

TechniqueEasyMediumHardExtreme
Naked Single100%100%94%73%
Hidden Single51%85%100%98%
Locked Candidates0%29%90%91%
Naked Pair0%5%43%33%
Naked Triple0%1%25%21%
Hidden Pair0%0%5%3%
Hidden Triple0%0%2%2%
X-Wing0%0%10%7%
XY-Wing0%0%17%11%
Not solvable by these 90%0%37%100%

Three results are worth staring at:

  • Easy puzzles never need an elimination technique. Not once in the sample: 0% require locked candidates, pairs or anything beyond a single. That is what "easy" should mean, and it is why we grade them 100% naked-single-driven.
  • Locked candidates are the real dividing line. They appear in 29% of Medium puzzles but 90% of Hard ones. If you cannot spot pointing pairs, Hard is where you will stall.
  • “Extreme” is a different kind of puzzle. 100% of the hardest tier cannot be finished with any of the nine techniques — they need forcing chains or search. That is the honest definition of expert: not "fewer clues", but "no single-deduction path to the end".

How our four tiers map

TierWhat it guaranteesWhat it avoids
Easysolvable with full house and naked/hidden singlesany pencil-mark-only technique
Mediumat most locked candidates, pairs and triplesfish and wings
Hardat most one fish or triple-level patternXY-Wings and chains
Expertanything up to XY-Wings; chains allowed—

Every puzzle page shows the techniques it needs and its score, so you can audit the label yourself. If a puzzle's label ever looks wrong, the technique list is right there next to the board.

Reproducing this

The rating is done by the same solver that powers the hint button, so you can check any claim: open a puzzle, press H, and the hint will name the technique it is about to use. The corpus scan behind the tables above runs on every build of this site.