How a Dayzle puzzle is made, and how we know it is fair
Every puzzle on Dayzle is built by a program before you ever see it, and a second program then proves that it can be solved. This guide explains how that works, why we bother, and what it means for a daily puzzle you are meant to attempt only once.
People wrote the rules, the difficulty tables and the filters, but the daily work of producing a fresh puzzle for every game is done by code that can be tested.
A generator proposes, a solver checks
Each game has its own generator, kept apart from the code that runs in your browser. Given a game and a date, it builds a candidate puzzle. The randomness is repeatable: the seed is the game id plus the date, something like stamp:2026-10-18, so the same inputs always give the same puzzle. If a candidate fails a check, the generator tries again with a numbered sub-seed until one passes.
The second half is the solver. In the games scored against par, it is an exhaustive search designed to be complete: it explores every branch that could matter and drops only the ones it can prove are no better. When it reports the smallest number of moves, nothing smaller exists.
Two things come out of that. First, proof that the puzzle has a solution at all. Second, the par, the true minimum. When a game says par is seven stamps, seven is not an average of what testers managed. It is the best anyone could do.
Two examples: Tickwork and Overstamp
In Tickwork you turn every gear's arrow upwards, and each push turns the pushed gear a quarter turn one way and its touching neighbours the other way. The key observation behind its solver is that order never matters: pushing A then B leaves the same board as B then A, and four pushes make a full turn. So a solution is just a count from 0 to 3 for each gear. Each gear's final angle gives one equation modulo 4, and par is the smallest total push count that satisfies the whole system.
The solver is a branch-and-bound search over those counts. It checks each equation as soon as its gears have a value, and abandons a branch once its total passes the best found so far.
Overstamp works the other way round. You reproduce a picture with coloured stamps, and a new stamp covers whatever lay beneath it. The solver therefore peels backwards: the top layer was stamped last, so it looks for a stamp that could have gone on last, removes it, and repeats. Its lower bound never overestimates the work left, so the first answer it reaches is a minimum. We validated it against two independent methods.
Both solvers run on a budget of search steps, not a stopwatch. A time limit would let a fast laptop and an old phone reach different puzzles from one seed.
Filters: rejecting the dull and the guessable
A solvable puzzle can still be a bad one, so every candidate also passes game-specific filters. Tickwork rejects a mechanism in which some gear sits idle in the best solution, and any locked gear needs at least two neighbours, because with one neighbour that neighbour's moves would be forced. Overstamp requires that every stamp appears in some par solution, and that at least two cells are overlapped by different colours, so the order of stamping really matters.
Gatelit shows the idea most clearly. You flip switches to light exactly the lamps a circuit of logic gates asks for. The filters brute-force every combination, at most 256, and require that exactly one lights the goal. They also require that flipping any switch or gate changes some lamp somewhere, and they reject circuits you can solve one lamp at a time: if each lamp alone pins down all the switches, the puzzle is thrown out.
The week: easiest on Monday, longest on Sunday
Difficulty is set by weekday, not by whim. Each game has a table with three levels. Monday and Tuesday are level 1, Wednesday to Saturday are level 2, and Sunday is level 3. Tickwork starts Monday as a chain of 5 gears and reaches 11 by Sunday. Overstamp goes from 5x5 boards with 3 stamps to a 7x7 board with 5 stamps on Sunday.
We have already corrected this once. In the first weekend of October 2026 the Saturday and Sunday puzzles felt far harder than Monday and Tuesday. From Monday 12 October the curve is gentler, and the tables used before that date are kept untouched so past days still generate exactly as they did.
Cubecast shows how the numbers push back: a 5x5 board has a minimum par of 13 under the filters, so Saturday could not be made easier without changing the board, and it became 4x4.
If you open a game for the first time on a Sunday, the game offers a warm-up: Monday's easy puzzle, which does not count towards streaks or rankings.
One puzzle for everyone, and solutions held back
Everyone who opens a game on the same date gets the same puzzle. The date comes from your own device's calendar, so a new puzzle arrives at your local midnight. The site is rebuilt automatically twice a day, and tomorrow's puzzles are already published, because the earliest time zone, UTC+14, is a day ahead of us.
Solutions follow a stricter rule. A solution file for a date is published only once that date is two days behind in UTC, because the last time zone on the planet, UTC-12, is still on yesterday in UTC. The build fails if a solution for a still-playable day appears in the output.
Once a day is closed everywhere, it goes in the archive. You can replay it and watch an animated solution, shown step by step. The build replays each solution against the published puzzle and fails if it does not reach the best result.
What par means for your score
In par games, par earns three stars. In Tickwork, one or two pushes over earns two stars and anything above that earns one. Only your first attempt on the day's puzzle counts for rankings. The server rebuilds that day's puzzle with the same generator, replays your moves and confirms the result, so what is ranked is what actually happened on the board. Our own scoring notes admit the limits: the engines are deterministic, so someone could write a solver, and the realistic defence is a minimum human time rather than a promise of no cheating.
Not every game uses par. Tipscale and Gatelit count tries out of six. Pingwreck and Ropecall are scored against a bot that always chooses the most informative move. That bot is not a proven optimum, so beating it is possible, and you will see a trophy when you do.
When you share a result, it contains your stars and counts, and never the solution. A friend can see the score to beat and still solve it from scratch.