Every Sudoku you have ever played began life as a completed grid. That feels backwards until you think about it: a puzzle maker cannot scatter random digits and hope a solvable puzzle falls out. They start with a full, valid solution, then carefully remove digits, checking after every removal that the puzzle can still be solved in exactly one way. The elegant pattern of givens you see is what remains when that process is done well.
This article follows a puzzle from birth to publication: how solution grids are generated, why symmetry appears in almost every newspaper puzzle, how makers decide which clues stay, and how difficulty is really graded. Understanding it will also make you a better solver, because several “mysteries” of Sudoku, including why some sparse puzzles are easy and some dense ones are brutal, stop being mysterious.
Step 1: Build a complete, valid grid
Software generators start by filling an empty grid so that every row, column and box contains 1 to 9. The common method is backtracking with randomised digit order: fill cells one by one, retreat whenever a placement breaks a rule, and shuffle the candidate order so each run produces a different finished grid. There are 6,670,903,752,021,072,936,960 possible completed Sudoku grids, so no generator ever repeats itself by accident.
Human setters do the same thing more slowly, often starting from a pattern (the digits along the diagonals, for example) and completing the grid by hand. Either way, the finished grid is the answer key, and everything afterwards is subtraction.
Step 2: Remove digits in symmetrical pairs
Now the maker starts deleting givens. Most published puzzles are rotationally symmetrical: if a cell at the top left is a given, the cell diagonally opposite at the bottom right is a given too. Symmetry is not required by the rules, and it makes the maker’s job harder, not easier; it survives because players find it beautiful, and because a symmetrical pattern is a visible promise that the puzzle was crafted rather than dumped out by a script.
Digits are therefore removed in opposite pairs (or as a single centre cell). After each removal, the maker must re-test the puzzle, which leads to the rule that governs the entire craft.
Step 3: Enforce the one-solution rule
A valid Sudoku has exactly one solution. Remove the wrong digit and the puzzle suddenly admits two or more completions, at which point it is broken: a solver following perfect logic can reach a fork that only guessing can settle. So after every removal, a solver program checks the puzzle. This is computationally serious work; a naive checker is far too slow, so generators use fast logical solvers that track how many solutions exist and stop as soon as a second one appears.
If removing a pair breaks uniqueness, the pair goes back and the maker tries elsewhere. The process ends when every remaining given is load-bearing: remove any one of them and the puzzle gains a second solution. Such puzzles are called minimal, and the most admired newspaper puzzles are minimal or close to it.
This is also where the famous number 17 comes from. In 2012, researchers proved by exhaustive computer search that no 16-given Sudoku can have a unique solution; 17 is the true minimum. Almost no published puzzle goes that low, however, because minimal-clue puzzles are usually savage, and because symmetry constrains the pattern. If clue count and difficulty surprise you, our article on whether more clues can mean a harder puzzle explains the mechanics.
Step 4: Grade the difficulty by technique, not by clue count
Once a unique puzzle exists, the maker grades it. Serious generators do not count givens; they solve the puzzle with a library of human techniques, singles first, then pairs, intersections, and on up to chains and fish patterns, and record the hardest technique the solve required. A puzzle solvable with singles alone is easy even if it has only 24 givens. A 34-given puzzle that demands an X-Wing is not.
That is why difficulty labels occasionally feel “wrong” to players: the label measures the peak technique, while your experience measures the whole journey. The full picture is in our guide to what makes a Sudoku puzzle hard.
How Killer Sudoku puzzles are made
Killer generation reuses the same skeleton with one extra layer. The software takes a completed grid, partitions it into cages (usually favouring sizes of two to four cells), computes each cage’s total from the solution, and then runs the same uniqueness checking; there are no givens to remove, so the “removal” step instead tests whether cages can grow or merge while keeping exactly one solution. Cage shape is a design tool: many small cages make an easier puzzle, large sprawling cages make a harder one, which is why beginner Killer grids look like patchwork and expert ones look like continents.
Frequently Asked Questions
Are Sudoku puzzles made by humans or computers?
Nearly all high-volume puzzles today are computer-generated, then filtered and graded by software. Human setters still create puzzles for competitions and books, usually with software checking uniqueness. The honest difference is curation: a good editor rejects valid-but-dull puzzles that a raw generator would ship.
Why do some puzzles have no symmetry?
Symmetry is an aesthetic choice, not a rule. Some apps skip it for speed of generation or to reach lower given counts. An asymmetric puzzle can be perfectly valid; it just gives up some of the crafted look.
Can a puzzle have two solutions and still be published?
It happens by mistake, rarely, and it is considered a defect. Reputable sources, including the puzzles behind our daily Sudoku, verify a unique solution before publication.
So the puzzle on your screen is really a sculpture: a full grid with everything removable carved away, until what is left is the smallest pattern of clues that still forces one single answer. Next time a puzzle feels elegant, that is why. Someone, or some very careful software, removed a hundred digits and kept exactly the right ones.
