Most sudoku sites rate difficulty by counting clues: fewer givens, harder puzzle. It’s an easy number to compute and a bad proxy for what actually makes a puzzle hard. We don’t use it. This page describes the system we actually run, not a simplified version of it.

Why clue count doesn’t work

A 28-clue puzzle and a 24-clue puzzle can require exactly the same logic, or wildly different logic, depending on which cells are missing rather than how many. A puzzle can lose clues until no more can be removed without creating a second solution — genuinely minimal — and still be solvable with nothing but naked and hidden singles. Clue count tells you how aggressively a puzzle was dug, not what’s needed to finish it.

What we grade on instead

Every puzzle we generate gets run through our own logical solver: a program that applies named techniques — naked single, hidden single, locked candidates, pairs, triples, X-Wing, swordfish, XY-Wing — in order from simplest to most advanced, always taking the easiest applicable step at each point, the same way a careful human solver would. The puzzle’s difficulty is the hardest technique that solver ever needed to use, not the easiest one that happened to work somewhere.

That produces five bands:

  • Easy — naked and hidden singles only.
  • Medium — adds locked candidates (pointing and claiming).
  • Hard — adds naked and hidden pairs, and naked triples.
  • Expert — the hardest requirement is exactly X-Wing.
  • Evil — needs swordfish, XY-Wing, or something beyond all of those.

A side effect worth knowing about

Hitting a band exactly isn’t evenly likely. In testing, randomly-dug puzzles land on “easy” more often than any other band — simple techniques alone solve a surprising number of minimal puzzles — while needing X-Wing as the exact ceiling (not easier, not needing something harder too) turned out to be rare, on the order of one in a few hundred attempts. Evil, by contrast, is comparatively common, since “needs something beyond our ten techniques” is a wide net. That’s why generating an Expert puzzle can take noticeably longer behind the scenes than generating an Easy one — we’re not picking a 28-clue grid and calling it a day, we’re searching until we find a puzzle whose logical ceiling actually matches the label.

Why this matters while you’re solving

Every puzzle on this site tells you which techniques it actually needs, and every hint is generated by the same solver, in the same order — so a hint on an Easy puzzle will never casually drop an X-Wing on you. See how our puzzles are generated for the other half of this pipeline: how a puzzle gets down to a unique solution in the first place.