Puzzles that are provably fair
Every generated puzzle is proven to have exactly one solution before you see it, and difficulty is measured from the reasoning it actually requires. No account, no scores sent anywhere, nothing to install.
Nothing leaves your device Puzzles are generated in your browser, progress is saved locally, and no score, time or answer is ever sent anywhere. There is no leaderboard, no username and no account.
Play
Deduction you can check, with hints that teach the next step.
Create
Printable sheets with answer keys, for classrooms, books and long flights.
Being built next
The shared foundations — seeded generation, the publish-and-verify pipeline, and the difficulty framework — are done and tested, so the remaining work is per-puzzle engines rather than new architecture. Next up: Cryptograms, a Word Search generator and a Maze generator, each held to the same rule that nothing ships until it is proven to have one answer.
They are named here rather than shown as cards you cannot click.
The rule everything here follows
A puzzle whose rules imply one intended answer must be proven to have exactly one, and "a solver found an answer" is not that proof. It establishes only that at least one exists. Proving uniqueness means continuing the search after the first solution and finding no second — which is what this site does, after every single clue removed, solving from the givens exactly as you receive them.
It matters because the failure is invisible. A Sudoku with two answers looks completely normal, plays normally, and only betrays itself when someone fills the grid correctly and is told they are wrong. Most puzzle generators on the web do not check.
Difficulty is measured, not guessed
The usual shortcut is to label a puzzle by how many clues it has. That is very nearly meaningless — a 26-clue Sudoku solvable by the simplest technique is easier than a 32-clue one needing an X-wing. Here a second solver works the way a person does, applying named techniques in order of difficulty, and the label comes from the hardest one the puzzle genuinely requires.
Two puzzles generated here have 31 clues each: one is beginner, the other easy. Same count, different work. That is the whole argument for measuring.
What we will not build
No leaderboards, usernames, chat or public feeds. No streak pressure, loot boxes or chance-driven retention. No claim that a score measures intelligence. No account gate on hints, difficulty levels or exports. Children use puzzle sites, and none of that belongs anywhere near them.