Sudoku Evil: What Makes Expert Puzzles Hard and How to Solve Them

"Sudoku evil" refers to the hardest standard difficulty tier — puzzles that cannot be finished with scanning and singles alone, and that require pattern-based techniques like pointing pairs, X-Wing, XY-Wing, and coloring. The name comes from Web Sudoku's four-level scale (Easy, Medium, Hard, Evil), and it stuck; most sites now use "Expert," "Extreme," or "Diabolical" for the same thing.

Here's the thing that surprises people: an evil sudoku isn't harder because it has fewer clues. It's harder because at some point in the solve, there is no cell with only one candidate and no digit with only one home. Every single-step deduction is exhausted. You have to find a pattern spread across four or more cells that eliminates a candidate somewhere else entirely — and only then does the grid start moving again.

That's a genuinely different skill, and it's learnable in an afternoon. This guide covers what actually separates evil from hard, the six techniques that clear almost every expert puzzle, and the notation discipline without which none of them are visible.

What "evil" actually means

Difficulty ratings are generated by running a solver that only knows a fixed list of techniques, then recording the hardest one it needed. Roughly:

Clue count correlates with difficulty, but weakly. A 26-clue puzzle with generously spread clues can be gentler than a 30-clue puzzle where all the clues cluster in one band. The technique requirement is the real measure.

For the record: the minimum number of clues a sudoku can have and still possess a unique solution is 17, proved by exhaustive computer search in 2012. Those puzzles are mathematical curiosities and aren't typically harder than a normal evil grid — sparse doesn't mean deep.

Before any technique: fix your notation

This is not a warm-up section. The number one reason people fail at sudoku evil puzzles is incomplete or stale pencil marks, not missing knowledge. Every technique below is a visual pattern in the candidate grid. If the candidates are wrong, the patterns are invisible or, worse, fake.

Four rules:

1. Fill in every candidate before you get stuck, not after. On an evil puzzle, do a complete candidate pass as soon as scanning stops yielding placements. Marking up a half-solved grid under pressure is where errors creep in.
2. Keep candidates in fixed positions inside the cell. 1 top-left, 2 top-center, through 9 bottom-right. Consistent placement means you recognize a {4,7} pair by its shape without reading it.
3. Erase the instant you place. When you write a digit into a cell, immediately strike it from every candidate list in that row, that column, and that box. All three. Every time.
4. Never write a candidate you haven't checked against all three units. A single phantom candidate can make a valid X-Wing look invalid, and you'll stare at a solvable grid for twenty minutes.

If you're solving on paper, use a mechanical pencil with a fine lead and a real eraser. If you're solving digitally, use the auto-candidate feature — there's no honor in hand-marking, and the point is the deduction, not the bookkeeping.

The six techniques that clear a sudoku evil grid

Learn these in order. Each one builds on the last, and the first three will get you through most of a typical expert grid before you need the heavier machinery.

1. Pointing pairs and triples

If all the candidates for a digit within one box lie in a single row or column, that digit can be eliminated from the rest of that row or column outside the box.

The logic is airtight: the digit has to go somewhere in that box, and every available cell is in the same row. So the digit is definitely in that row, somewhere in that box — and therefore definitely not in that row anywhere else.

Example. In box 7 (rows 7–9, columns 1–3), the candidate 9 appears in only two cells: R7C1 and R7C3. Both are in row 7. Therefore 9 is in row 7 within box 7, which means 9 can be erased from R7C4, R7C5, R7C6, R7C7, R7C8, and R7C9.

Six eliminations, and one of them is very often the one that unlocks a hidden single.

2. Box/line reduction (the mirror image)

If all the candidates for a digit within one row or column lie inside a single box, that digit can be eliminated from the rest of that box.

Same reasoning, run the other way. If the only places a 4 can sit in row 5 are R5C4, R5C5, and R5C6 — all inside box 5 — then the 4 for row 5 lives in box 5. So box 5's 4 is accounted for, and 4 is erased from R4C4, R4C5, R4C6, R6C4, R6C5, and R6C6.

Pointing pairs and box/line reduction are the workhorses. Run both passes over the whole grid before reaching for anything fancier — on many evil puzzles, alternating between them cracks the entire thing.

3. Naked and hidden subsets

A naked pair is two cells in the same unit with identical two-candidate lists; those digits belong to those cells and vanish from the rest of the unit.

If R4C2 and R4C7 both read {2,9}, then between them they own the 2 and the 9 for row 4. Every other cell in row 4 loses both digits.

Naked triples work the same way but are easier to miss because the three cells don't need identical lists — they just need to draw from the same three digits. {3,6}, {6,8}, and {3,6,8} in one unit is a valid naked triple on {3,6,8}.

A hidden pair is the inverse: two digits in a unit that can only go in two specific cells. Those cells might have five candidates each, but if 4 and 7 appear nowhere else in the box, then those two cells are 4 and 7 — and every other candidate in them can be erased.

Hidden pairs are the technique experienced solvers most often forget to look for, because your eye is drawn to short candidate lists and hidden pairs hide inside long ones.

4. X-Wing

An X-Wing is four cells at the corners of a rectangle: a digit is a candidate in exactly two cells of one row, exactly two cells of another row, and those cells share the same two columns. The digit can then be eliminated from the rest of both columns.

Concrete version. Candidate 6 appears in row 2 only at R2C3 and R2C7. Candidate 6 appears in row 8 only at R8C3 and R8C7. Those four cells form a rectangle across columns 3 and 7.

Why it works: row 2's 6 is at C3 or C7. Row 8's 6 is at C3 or C7. Whichever way it falls, one of them is in column 3 and the other is in column 7 — they can't both be in the same column, because then the other column would have no 6 for those rows. So columns 3 and 7 each have their 6 locked into row 2 or row 8.

Therefore 6 is erased from every other cell in column 3 and column 7 — up to ten eliminations from one pattern.

X-Wings work identically with rows and columns swapped: two columns, two shared rows, eliminate from the rows.

How to hunt for one: pick a digit, then look for rows where it has exactly two candidate cells. Write down the column pairs. Any two rows sharing a column pair is an X-Wing. This is a mechanical search that takes about ninety seconds per digit, which is why it's worth doing on the digits that appear least often in the grid.

5. XY-Wing

An XY-Wing uses three cells with exactly two candidates each: a pivot with candidates {X,Y}, and two pincers with {X,Z} and {Y,Z}, each sharing a unit with the pivot. Any cell that sees both pincers cannot contain Z.

Worked example:

• Pivot at R2C2, candidates {4,7}
• Pincer A at R2C8, candidates {4,9} — shares row 2 with the pivot
• Pincer B at R5C2, candidates {7,9} — shares column 2 with the pivot
• Here Z = 9

Run the two cases. If R2C2 is 4, then R2C8 can't be 4 (same row), so R2C8 = 9. If R2C2 is 7, then R5C2 can't be 7 (same column), so R5C2 = 9. Either way, one of the pincers is a 9.

So any cell seeing both pincers can't be 9. R5C8 sees R5C2 along row 5 and R2C8 down column 8. Erase the 9 from R5C8.

XY-Wing is the technique that most often separates people who finish evil puzzles from people who don't, because it's the first one where the pattern isn't geometrically obvious — you find it by scanning bi-value cells, not by looking at shapes.

6. Simple coloring

Coloring tracks a digit that has exactly two possible cells in several units, chaining them into a network of "if this one is true, that one is false" links until a contradiction appears.

Pick a digit — say 3. Find a unit where 3 has exactly two candidate cells. Mark one blue and the other green: exactly one of them is a 3. Follow the chain. Wherever a blue cell shares a unit with another cell that's part of a two-candidate link, that partner becomes green, and so on.

Two payoffs:

• Two cells of the same color in one unit means that color is entirely false. Every cell of that color loses the digit.
• A cell that sees both a blue and a green cell can't be the digit at all, since one of those two is definitely true.

Coloring is more work than the others and it's easy to make a bookkeeping mistake, so save it for when everything else is exhausted. But it clears grids that nothing simpler will touch.

The order to attack an evil puzzle

Working randomly is how a 40-minute puzzle becomes a 2-hour one. A repeatable sequence:

1. Scan for singles first. Crosshatch every digit, most-common first. Place everything easy.
2. Full candidate pass. Every empty cell, every unit checked.
3. Naked and hidden pairs. Sweep all 27 units.
4. Pointing pairs, then box/line reduction. Both directions, all nine digits.
5. Back to singles. Steps 3 and 4 usually generate new ones. This loop — subsets, locked candidates, singles — clears most evil puzzles on its own.
6. X-Wing, then XY-Wing. Only when the loop stalls completely.
7. Coloring, unique rectangle, chains. Last resort.

After every successful elimination, go back to step 5. The loop is the method; the exotic techniques are just how you restart it.

Two more patterns worth having

Unique rectangle (Type 1)

If four cells form a rectangle spanning exactly two boxes, and three of them contain only the candidates {a,b} while the fourth contains {a,b} plus extras, the fourth cell cannot be a or b.

The reasoning is unusual: it relies on the puzzle having a unique solution. If that fourth cell were a or b, all four corners would hold only a and b, and you could swap them diagonally to produce a second valid solution. Since a properly made sudoku has exactly one answer, that arrangement is impossible — so the fourth cell must be one of its extra candidates.

This feels like cheating the first time you use it. It isn't, as long as the puzzle is genuinely unique, which any reputable generator guarantees.

Swordfish

Swordfish is X-Wing extended to three rows and three columns. A digit is confined to at most three columns across three rows — say columns 2, 5, and 9 in rows 1, 4, and 7 — and the digit can then be eliminated from those three columns everywhere else.

It's rarer than X-Wing and considerably harder to spot by eye. Most puzzles labeled evil don't need it; puzzles labeled extreme sometimes do.

What actually trips people up

Having watched a lot of people hit the wall on expert grids, the failure modes are remarkably consistent.

• Stale candidates. Said three times now because it causes more failed solves than everything else combined. When a grid feels impossible, re-derive the candidates in the region you're stuck on from scratch before assuming you need a new technique.
• Guessing, then losing track. If you do guess — and plenty of strong solvers do as a last resort — mark the branch point clearly and use a different color for everything after it. An unmarked guess that turns out wrong costs you the entire grid.
• Hunting X-Wings before running locked candidates. Pointing pairs are faster to find and more common. People skip them because X-Wing sounds more impressive.
• Only scanning rows. Every technique described here works in columns and boxes too. A solver who only sweeps rows will miss roughly two-thirds of the available patterns.
• Quitting at the first wall instead of the second. The first stall on an evil puzzle almost always yields to a hidden pair you didn't look for. Do a full 27-unit subset sweep before deciding you're stuck.
• Marathon sessions. Accuracy falls off a cliff after about 45 minutes. Put the grid down, come back, and the pattern is usually sitting there in plain sight.

Building up to evil difficulty

If evil puzzles are currently out of reach, the fastest route isn't grinding evil puzzles. It's grinding hard ones until pointing pairs and hidden pairs are automatic — because on an evil grid, those techniques do most of the work and you need them to be free.

A practical drill: work hard puzzles with full candidate notation even when you don't need it, and after each solve, name the technique that broke the biggest logjam. Naming it is what makes it retrievable next time.

For volume, our free puzzle maker generates unlimited classic sudoku at easy, medium, and hard, prints one to six per page with an answer key, and guarantees a unique solution on every grid — which matters more than it sounds, because uniqueness is what makes techniques like the unique rectangle legitimate and guarantees the puzzle really is solvable by logic alone.

Frequently asked questions

Is sudoku evil difficulty the hardest level there is?

On most sites, yes — "evil" is the top of a four-tier scale that runs Easy, Medium, Hard, Evil, popularized by Web Sudoku. Some publishers add tiers above it under names like Extreme, Diabolical, or Nightmare, which typically require Swordfish, forcing chains, or uniqueness techniques. There's no industry standard, so an "evil" puzzle on one site may be a "hard" on another.

Can evil sudoku puzzles be solved without guessing?

Yes, provided the puzzle has a unique solution — and any properly generated one does. Every placement in a unique-solution sudoku is logically derivable; the deductions just get long. What varies is whether you know the required technique. If you're stuck and considering a guess, the honest reading is usually that a pointing pair or hidden pair is sitting on the board unseen.

How many clues does an evil sudoku have?

Typically 22 to 26 given digits, versus 36 to 45 for an easy puzzle. But clue count is a poor difficulty measure. Difficulty is determined by which solving techniques the grid forces you to use, and a well-placed set of 30 clues can produce a harder puzzle than a poorly placed set of 24. The proven minimum for any unique sudoku is 17 clues.

What's the difference between X-Wing and XY-Wing?

X-Wing concerns a single digit across four cells forming a rectangle, and eliminates that digit from two full lines. XY-Wing concerns three digits across three bi-value cells that don't form any particular shape, and eliminates one digit from the cells that see both pincers. They're unrelated despite the similar names — X-Wing is a shape you spot visually, XY-Wing is a relationship you find by checking bi-value cells against each other.

How long should an evil sudoku take?

For a solver comfortable with the full technique list, 20 to 45 minutes is normal. Someone learning the techniques should expect an hour or more on their first several, and that's fine — the time goes into searching for patterns, not into the deductions themselves. Speed comes from recognition, and recognition comes from repetition.

Why do I keep making mistakes on evil sudoku?

Almost always candidate bookkeeping rather than faulty logic. The two big culprits are forgetting to erase a candidate after placing a digit, and doing an incomplete candidate pass in the first place. Both produce grids that look solvable but aren't, and the error surfaces twenty moves later when you can no longer trace it. Slowing down the notation pass is counterintuitively the fastest way to finish.