The rules, and the part they leave out
Fill the grid so that every row, every column and every three-by-three box contains each of the digits 1 to 9 exactly once. That is the whole game. There is no arithmetic in it — you could play with nine colours or nine letters and nothing would change.
What takes time to learn is not the rules but the order you look for things. A solver who has been at it for years is not calculating faster than you; they are asking better questions, in a better sequence. That sequence is what the rest of this page is about.
Learning to solve, in the order it actually happens
Below is roughly the order most solvers acquire these, each one useful long before the next becomes necessary. Everything here applies directly to the board above.
Squares that only have one answer
Pick any empty square and ask what cannot go in it. Everything already in its row is out. Everything in its column is out. Everything in its box is out. Cross all of those off and sometimes exactly one digit survives.
The trap is doing this square by square across the whole board, which is slow and dull. Work a box at a time instead, starting with the boxes that already hold the most digits — they are the ones most likely to give something up.
Digits that only have one home
Now flip the question. Instead of asking what fits in a square, take a digit and a region and ask where it can possibly go. Take a 4, look along the rows and columns cutting through a box, and rule out every square a 4 cannot occupy. If one square is left, the 4 belongs there — even if that square could also have held three other digits.
Most people find this harder at first, because it means holding a digit in mind while your eyes move rather than holding a square while you run through digits. It is worth the effort: once this is automatic, medium puzzles stop being work.
Sweep one digit at a time across the whole grid rather than wandering. Do every 1, then every 2, and so on. It feels mechanical, but it is how you stop missing the obvious, and it is far quicker than it sounds once you have done it a few times.
When singles run out: pencil marks
Sooner or later the grid stops giving. This is the point where you write down candidates — the small digits that could still go in each empty square. Everything from here on is a statement about candidates, so without them written down there is nothing to reason over.
Pairs, triples, and the squares you can eliminate from
If two squares in a row both hold exactly the candidates 3 and 7, then between them they will use up the 3 and the 7. You do not know which way round, and you do not need to: every other square in that row can lose both digits. That is a naked pair, and the same logic extends to three squares sharing three candidates.
The mirror image is a hidden pair. If 3 and 7 can only appear in two squares of a row, then those two squares are the 3 and the 7, and every other candidate sitting in them can be rubbed out.
Locked candidates
Here is the one that unlocks most hard puzzles. Look at a box and find a digit that can only go in squares sharing a single row. You do not know which of those squares it is, but you do know the digit lives in that box, on that row — so it cannot appear anywhere else along the row outside the box. Cross it off there.
The reverse works too: if a digit in a row can only sit inside one box, it can be removed from the rest of that box. Two directions, one idea.
X-wings and XY-wings
These are the first techniques that feel like proper deduction rather than bookkeeping. An X-wing is four squares at the corners of a rectangle: a digit appears in exactly two squares of one row, exactly two of another, and both pairs sit in the same two columns. Whichever way it resolves, those columns are spoken for, so the digit can be struck from the rest of them.
An XY-wing uses three squares holding two candidates each — a pivot that sees two others, arranged so that whichever digit the pivot takes, one of the other two is forced. Any square seeing both of those loses the forced digit entirely.
Getting the most out of this board
Notes, and letting the board fill them
Notes switches to pencil mode, so tapping a digit writes it small instead of placing it. Fill writes every legal candidate into every empty square at once.
Filling them automatically is a mild shortcut, and whether you use it is a matter of taste. What is not a matter of taste is timing: do it after the singles have run dry, never at the start. Every digit you place first is a set of candidates you would otherwise have written out and then crossed off.
What the hint actually does
It does not hand you a digit. It finds the easiest genuine deduction available in your current position, names the technique, explains it in plain terms, and highlights the squares involved. You can then apply it or wave it away and go and find it yourself.
If the grid has no pencil marks yet it offers to write them first, because an explanation about candidates is impossible to check without them. If you have a wrong digit on the board it tells you that instead, since no technique is trustworthy on a broken grid. And when a position really is past every method it knows, it says so rather than inventing something.
Mistakes, undo, and the reveal button
A wrong digit stays on the board in red and adds to the mistake counter. Nothing is enforced — there is no three-strikes rule and no game over — because the counter exists to inform you, not to punish you.
Undo and redo step through everything, including a whole Fill as one move, so you can try an idea and back out of it cleanly. Reveal asks for confirmation and then completes the entire grid, which ends the puzzle.
Your game saves itself
The grid, your entries, your pencil marks, the timer and the mistake count are stored in your own browser as you play. Close the tab, come back tomorrow, and the same puzzle is waiting. Each difficulty keeps its own game, so trying a harder grid never costs you the one you already had going.
There is no account and nothing is uploaded. The flip side is that clearing your browser data for this site will lose an unfinished puzzle.
Playing on a phone
The board stays square and thumb-reachable rather than being shrunk to fit, with the digit pad directly beneath it. On a desktop keyboard, 1 to 9 place digits, the arrow keys move around the grid, N toggles notes, and Backspace clears a square.
Taking puzzles off the screen
Underneath the board is a pack builder that writes a PDF of fresh puzzles, with an answer key at the back if you want one. The grids are drawn as vectors, so they stay sharp at any size, and the whole thing is produced on your own machine — nothing is sent anywhere to be printed.
How difficulty works here
What "hard" actually means
A puzzle is not hard because it has fewer clues. It is hard because of the deepest technique it forces you to use. Every grid is put through a technique-based solver before you see it, and the level on the button is the hardest rung that solver had to stand on. The line above the board tells you which one.
The level on the button measures the solve ahead of you, not the clues taken away.
Each level has its own page going into what that particular rung asks of you: easy, medium, hard, expert, evil and extreme. Use the selector above to change level here, or those pages if you want the reasoning behind one.
A note on guessing
Every puzzle here has exactly one solution — verified during generation, where a removal that would leave two solutions is rejected on the spot. So guessing always works, in the sense that you can split the position and grind it out. Whether it satisfies is up to you.
Below the very top level you never need to. Most solvers who do guess save it for the end, on a square with two candidates, having exhausted everything else — and there is no shame in that.
Where the puzzles come from
The first four levels are built in your browser the moment you ask for one. A complete grid is generated, squares are removed one at a time in random order, and after every removal the puzzle is re-solved to confirm it still has exactly one answer. If it does not, the digit goes back.
That is why the clue count varies a little from puzzle to puzzle: we would rather hand you twenty-six clues with one solution than twenty-four with two. The top two levels are drawn instead from banks of grids chosen for resisting the standard methods outright.