How We Make the Daily Puzzles

Every day we publish a fresh set of puzzles that thousands of people solve at once. Once a board goes out there is no correcting it, and a broken one costs somebody twenty minutes of their evening that we cannot give back. So the process behind them is more careful than the puzzles look.

The generator never checks its own work

Every board is built by one program and then checked by a second, separate one.

That sounds like belt and braces, and it is the most important rule we have. A generator that marks its own homework will always agree with itself, including when it is wrong, because the same mistaken assumption sits on both sides of the test. Ask it whether the board is fair and it says yes, for the same reason it built the board that way in the first place.

So the checker is written independently and uses different methods to reach the same answers. It re-solves every board from scratch, proves there is exactly one solution, and re-rates the difficulty on its own terms. When the two disagree about anything, the board is thrown away rather than patched, because disagreement means one of them is wrong and we do not get to assume which.

For the grid games we run a third check as well, with a different search strategy again. Agreement between two unrelated methods is decent evidence. Agreement between three is the standard we would want if we were the ones solving.

Exactly one solution, always

A puzzle with two answers is not a harder puzzle. It is a broken one, because somewhere in it sits a point where reasoning runs out and you simply have to pick. The pleasure of finishing rests on the answer having been findable all along, and a second solution quietly takes that away.

So uniqueness gets proved for every board, on every day, before it is ever scheduled. Never assumed.

You should never have to guess

The harder promise, and the one we care about most, is that logic alone is enough to finish.

A board can have exactly one solution and still be unfair. If the only way forward is to pick a cell, assume something and hope, the puzzle has stopped asking you to think and started asking you to be lucky. Keeping that out is most of the work.

For the number games it is built into the rating: the checker solves each board using only techniques a person actually uses, and anything that cannot be finished that way does not qualify. Balance is different, because there you sometimes do have to test an assumption, so we cap how deep the test is allowed to run. An assumption you can disprove in a few steps is reasoning. One you keep holding because nothing has broken yet has turned into a guess, and those boards get rejected.

Which is why being stuck on our puzzles is safe, if occasionally infuriating. There is a next move and it is findable. Had there not been, the board would never have passed.

What makes a Thursday harder than a Monday

The grids stay the same size all week, and Sunday plays by exactly the same rules as Monday. Two other things change instead.

The first is how much of the board comes free. Early in the week most moves sit where you would naturally look; later they do not. The second is how deep the reasoning runs once the easy moves dry up. A Monday rarely asks you to hold two ideas at once. A Sunday will, at least once.

Every board gets rated on both before it is scheduled, then placed on the day whose promise it fits. A board too easy for its slot fails as surely as one too hard, and both get discarded. Most candidates never make it at all: for every board you solve, a great many were generated, rated, and thrown out for landing on the wrong day.

We time them too

Difficulty is only half of what a day promises. The other half is how long it takes.

So each board carries an estimate of how long an unhurried average solver will need, and that estimate has to land inside the day's target before it can be scheduled. It is the reason a weekday puzzle is meant to fit inside a coffee break while a Sunday is meant to be sat with. The full day-by-day breakdown is in how the daily week works.

We adjust it when we get it wrong. When players told us the midweek grids were running past twenty minutes, we retuned the whole set rather than defending the numbers. The fix was to hand you more starting clues while leaving every deduction in place, because a shorter puzzle should mean less filling in rather than less thinking.

Why we bother

There are quicker ways to run a daily puzzle. Generate a board, look it over, ship it. That works most days.

We would rather be the site where it is right on the days nobody is checking. Somebody who hits a wall on a Saturday is trusting that the wall has a way through, and if that turns out to be false even once, for one person, we have lost something worth more than the board.

Frequently asked questions

Is there always exactly one solution?

Yes, on every board, proved by a program written separately from the one that generated it.

Will I ever have to guess?

No. Every board is checked to be solvable by reasoning alone before it is scheduled.

Are the puzzles the same for everyone?

Yes. One board per puzzle per day, worldwide, changing at midnight UTC.

Do the puzzles ever change after they are published?

No. Once a day's puzzles go out they are frozen, and the archive serves the original boards forever. Improvements only ever apply to future days.

Who checks the answers?

Programs do the proving, since they can re-solve thousands of boards exhaustively in a way people cannot. People decide what the difficulty should feel like, and retune it when players tell us it is off.