About PuzzleType
A free puzzle site with no accounts, no adverts and no question bank — every puzzle is built when you open the page, and every answer comes with the reasoning behind it.
How This Site Started
It started with an irritation rather than a business plan. Every puzzle site ends the same way — you finish, you press Show Answer, and you are told the thing you already wanted to know. Nothing explains why, and nothing tells you what was wrong with the option you nearly picked. The sites that do explain have a person writing each one by hand, so their libraries stop where the writing stopped.
The fix turned out to be structural. If the site generates the puzzle, it already knows the rule it applied and the specific error behind each wrong answer — the explanation is a by-product of construction rather than a separate writing job. That one decision is why there are 38 puzzle types here instead of three, and why none of them run out.
Who Makes This
Matthew HarrisFounder and maintainer
Matthew runs PuzzleType and maintains the engines behind it. Every puzzle type here ships with its own test suite, and every answer is derived twice by methods that share no code — that is the standard the site is held to, and the reason its explanations can be checked rather than taken on trust.
Corrections are welcome and get read: if a puzzle here looks wrong, the seed in the board's corner identifies it exactly. Report it.
What PuzzleType Is
PuzzleType is a collection of 38 puzzle types, each one generated rather than stored. There is no library of finished puzzles behind the site. When you open a page, the puzzle you see is constructed at that moment from a named pattern — and because the site built it, it knows exactly why the answer is the answer, and exactly what is wrong with each of the other options.
That last part is the whole reason the site exists. Most puzzle sites end an attempt with Show Answer, which tells you what you already wanted to know and teaches you nothing. Here, finishing a puzzle produces the working: which attribute changed, which rule applied, and — for every wrong option — the specific mistake that would have led you there.
Why The Puzzles Are Generated
Hand-written puzzles run out. A site with a thousand of them has a thousand, and once you have worked through them the site has nothing left to offer. More importantly, hand-written explanations do not scale: somebody has to write each one, so in practice most puzzles ship with a bare answer and no reasoning at all.
Deriving the puzzle solves both problems at once. The supply does not run out, and the explanation is a by-product of construction rather than a separate writing job. It also makes a promise possible that a stored puzzle cannot make: the wrong answers are not plausible-looking decoys, they are the results of specific, named errors.
How Correctness Is Established
A generated puzzle can be confidently, silently wrong, so nothing here is trusted just because the code that made it agrees with itself. Where an answer is decidable, it is decided twice — by a second implementation that does not share the first one's method.
| What is derived | The number that comes out | Why it is the proof |
|---|---|---|
| The nets of a cube | 11 | They are found by folding all 35 hexominoes, not listed from memory. Getting 11 means the folding is right. |
| Reachable sliding-tile boards | 181,440 | Exactly half of all arrangements, which is what the parity rule predicts. |
| The furthest 3×3 board | 31 moves | The known diameter of the 8-puzzle, reached by exhaustive search. |
| The tangram square | found in ~3ms | Where the seven tans sit is nowhere in the source. That a tiling exists is the proof the shapes are right. |
Every puzzle type also ships with a test harness that generates thousands of puzzles and checks the things that make a puzzle answerable: exactly one correct option, four visually distinct options, a named reason on every wrong answer, and the same seed always producing the same puzzle. Those harnesses have caught real bugs — puzzles with two valid answers, reflections that equalled their own original, and counts that were wrong in the same way twice.
What The Site Is Held To
Three rules decide what ships and what does not. They are not slogans — each one has thrown away work that had already been built.
One Answer, Provably
A puzzle with two valid answers is not a puzzle. Every grid is checked for a unique solution and for being solvable without a guess, by a second method that shares no code with the first. Puzzles failing either check are discarded rather than shipped.
Derived, Never Recalled
Counts and sets are computed, not written from memory. The eleven nets of a cube are found by folding all 35 hexominoes; that the answer comes out at eleven is the proof the folding is right. Four counts once written from memory were all wrong.
Limits Stated, Not Hidden
Where a type cannot be generated honestly, the page says so. Star battle stops at 7×7 and kakuro ships 60 stored boards rather than unlimited ones — both because the guarantees would otherwise stop being true. Two types were abandoned outright.
What The Site Deliberately Does Not Do
- No accounts and no sign-up. Nothing is gated, so there is nothing to log in to.
- No timer, anywhere. A clock measures how fast you already are; it does not help you get better. Both clocks the site once had were removed.
- No exam framing. These are puzzles to play. Where a puzzle type happens to appear in assessments, that is context, not the purpose.
- No hand-curated question banks. If a puzzle type cannot be generated with a derivable explanation, it does not ship here — which is why kakuro, despite a working engine, is not on the site.
Contact And Corrections
If a puzzle looks wrong, it may well be — a generated puzzle can be wrong in ways a written one cannot. Reports are genuinely useful, because a single bad seed usually points at a whole class of bad seeds. See the contact page.