Why Every Solvable Sudoku Needs At Least 17 Clues
Type a puzzle in from a newspaper page or an app and hit a wall with no valid number left to try, and the obvious question is whether you mistyped a clue or the puzzle itself is broken. Usually it's the typo. But the second possibility isn't as far-fetched as it sounds: a Sudoku with only 16 numbers filled in never has exactly one solution. Not because nobody's found one. Because in 2012, a team of mathematicians used a supercomputer to check every possibility there is and proved it can't be done.
The proof that took roughly a million CPU-hours
Setters had suspected for decades that 17 clues was the practical floor for a proper Sudoku, the kind with one and only one valid completion. Nobody had ever managed to publish a working 16-clue puzzle, but "nobody's found one" isn't the same as "one can't exist." In 2012, researchers Gary McGuire, Bastian Tugemann, and Gilles Civario settled it properly: they exhaustively checked every possible 16-clue arrangement using a technique called hitting set analysis, spread across a large computing cluster, and confirmed that not a single one produces a unique solution. Every 16-clue grid either has no solution at all or has more than one. Meanwhile, tens of thousands of distinct 17-clue puzzles are already known to work, so 17 is exactly the line.
Why fewer clues breaks uniqueness
The intuition doesn't require the supercomputer. A finished Sudoku grid has a lot of built-in symmetry: swap every 4 with every 7 across the whole board, for instance, and you still get a valid, complete grid. If a puzzle's clues never happen to include a 4 or a 7 anywhere, that swap works on the puzzle too, silently producing a second valid solution alongside the first. A puzzle needs enough clues, placed carefully enough, to block every one of these swaps. With too few numbers on the board, some symmetry almost always survives, and the "one true answer" a proper Sudoku promises quietly stops being true.
The puzzle's name is younger than the puzzle itself
Sudoku reads as distinctly Japanese, but the grid was invented in the United States. Dell Magazines published it as "Number Place" starting in 1979, most likely designed by a freelance puzzle constructor named Howard Garns. The Japanese publisher Nikoli picked it up in 1984 and gave it the name that stuck: Sudoku, short for suuji wa dokushin ni kagiru, roughly "the numbers must remain single," a nod to the rule that each digit appears only once per row, column, and box. The puzzle only became a worldwide phenomenon two decades later, in the mid-2000s, after a retired judge from New Zealand wrote a computer program to generate puzzles by hand-checkable rules and pitched the format to newspapers in the UK.
How a computer actually finds the answer
Solving a Sudoku by computer doesn't mean trying random numbers until something sticks. A good solver picks the emptiest-feeling cell first, specifically the one with the fewest numbers it could possibly be, fills it with a candidate, and keeps going; the moment a choice makes some other cell impossible to fill, it backs up and tries the next candidate instead. That combination, always attacking the most constrained cell first and backing out fast from dead ends, is why a well-built solver can crack even a famously hard puzzle in a few milliseconds instead of grinding through the trillions of raw possibilities a 9x9 grid allows.
That speed is also what makes a solver useful for the "did I mistype something" question above. A properly published puzzle, with 17 or more clues, is built to have exactly one valid completion, so if a solver comes back with no answer or flags a contradiction, checking the entered digits against the original grid is almost always faster than assuming the puzzle itself is at fault.
Try it yourself: enter what you know and let the Sudoku solver fill in the rest, or ask it for just one more number.