One puzzle a day, at midnight UK time
Each of the daily games on games.cyber-jack.uk changes over at midnight UK time, following the clocks when they go forward and back. Our three original games, Letterly, Cluster and Rankle, live on their own sites and change over at midnight UTC instead, which is 1am in the UK during British Summer Time.
So that nobody loses a puzzle they started at 11.55pm, the server keeps accepting yesterday's puzzle for 30 minutes after midnight.
Every puzzle is made in advance by a generator that is seeded with its date, so running it again gives exactly the same puzzles. Once a day has been played, it is pinned: its puzzle is copied word for word into a pinned file and can never change, even if we improve the generator later. New generator versions only ever start from the day after the last pinned day.
Keeping the answers secret
The schedule of future puzzles lives only on our server, never in the code your browser downloads. When you open a game, the server sends that day's puzzle and nothing else: the letters, the grid or the clues, but not the answer. Games with a clock go a step further and do not send the puzzle at all until you press start, which is also when your time begins.
Every page of past answers goes through a single gate in our code. A day only counts as finished once the UK date, measured 45 minutes in the past, is already later than it: that is the 30-minute grace period plus a further 15 minutes of safety margin. In practice, yesterday's answers can appear from 12.45am UK time, and today's never can. The page builder checks every day against this gate, and an answer page for an unfinished day stops the whole build with an error rather than being published.
Where the words come from
Our word games check your words against a dictionary built from SCOWL, the Spell Checker Oriented Word Lists by Kevin Atkinson and contributors, version 2020.12.07. SCOWL sorts words into size levels from very common to obscure. We accept words up to size 70, so unusual but real words still count when you type them, and we record each word's level.
The words we choose for puzzles come from much further up the list. Daily start and target words are usually drawn from the most familiar levels, with British spellings. Some examples of how each game uses the list:
- Stepwise works out its par route using only common words, so every rung of the par ladder is an everyday word. You can still use any accepted word, and beating par with a rarer one is allowed.
- Eightfold checks that its grid hides exactly one accepted word of eight or more letters, so there is never a second answer to argue about.
- Daisy's start words are familiar words of four to seven letters, and we leave out gloomy ones, because nobody wants to start their morning chain on a word like abuse.
- Warmer measures how close your guess is in meaning using the GloVe word vectors published by Stanford, limited to words in our list. The rankings are worked out in advance and kept on the server.
Keeping rude words out
We use two blocklists that we wrote ourselves, plus the single-word entries of the List of Dirty, Naughty, Obscene, and Otherwise Bad Words, published by Shutterstock and contributors under a Creative Commons licence.
- The hard list covers slurs and explicit obscenities. Those words are taken out of the dictionary entirely, along with simple forms such as plurals and -ing endings, so they are never accepted and never shown.
- The soft list covers words that are fine to type but that we would not choose. They are accepted if you enter them, but never used as a puzzle word and never shown in public answer lists.
Public lists of what players answered get one more layer of care. Our answer pages only show a player answer on days with at least 20 players, and only if it was given by at least three players or 5% of players, whichever is more. For word games, it also has to be a common word. That stops one person from planting a word on a public page. Area names in Shires and every line of text in Nosy are checked against the same blocklists when the puzzles are built.
Logic puzzles with exactly one solution
A logic puzzle with two solutions is broken, so Sown, Bimble and Twill are each checked twice before they are scheduled. First, an exact counter searches every possibility and proves there is only one solution. Second, a separate technique solver works through the puzzle start to finish using only the kinds of steps a person would make. If either check fails, the puzzle is not used.
The technique solver also sets the difficulty. In Sown, for example, the solver knows a ladder of ideas, from the simplest, where a row, column or plot has exactly as many free cells as seeds it still needs, up to short chains where a seed in one cell leads to a dead end a few steps later. A puzzle's difficulty comes from the hardest idea it needed and how often. The week is planned around that: Monday and Tuesday are easy, Wednesday and Thursday medium, Friday and Saturday hard, and Sunday is a bigger twin garden with two seeds per row.
Nosy, our village whodunnit, has the strictest check of all. Each case must have exactly one solution, must be solvable by deduction without a guess, and every clue must matter: take any single clue away and the case has more than one answer. Nibble, the little crossword, is proved the same way from its clues. Considering only words that match each clue, there must be exactly one way to fill the grid.
Quotes, and how we credit them
Four games use quotations: Unquote, Marquee, Prompter and Hearsay. Puzzles before 18 October 2026 used only public-domain lines whose authors died more than 70 years ago. Since then we have also used short lines from films, TV series, books and plays under the UK quotation exception in copyright law, which allows brief quotation with proper acknowledgement. Every quote is credited with its character, work and year, and the author for books, once you solve it. We do not use song lyrics.
Every quote's wording is checked against a recorded source, such as Wikiquote, a published script or the text of the book. A script fetches that source and confirms that the quote's letters appear in it in the same order. Capitals, punctuation and spacing can differ, but any changed, added, dropped or reordered word fails, and a quote that fails is not used. Each quote is used by one game only, ever. We also try not to repeat a film or book within 14 days across the quote games, or within 30 days in the same game.
If you hold the rights to a line and would like it removed, our credits page explains how to ask, and we will take it out.
Data, people and tools
Shires uses boundary data from the Office for National Statistics under the Open Government Licence, which our own code projects, simplifies and measures. No area comes back within 120 days. Everything else in the games is our own work or openly licensed, and the full list, including every font, is on the credits page.
We are open about where we have used AI help. Rarest's prompts and answer lists were drafted with AI assistance and then reviewed and edited by hand. Nibble's clues are original, AI-assisted and reviewed against each answer's meaning. Choosing and checking quotes was AI-assisted too, with every wording verified mechanically as described above. No AI service runs when you play the daily games on this site, and none is used when their puzzles are generated. Letterly, on its own site, is different: it uses an AI language model to check whether each answer really belongs to its category.
Quick summary
- New puzzles at midnight UK time for the dailies, midnight UTC for Letterly, Cluster and Rankle.
- Puzzles are generated in advance from the date, and played days never change.
- Your browser only ever receives today's puzzle, never an answer.
- Answer pages appear from 12.45am UK time for yesterday and never for today.
- Logic puzzles are proved to have one solution and solved without guessing before they are used.
- Quotes are checked word for word against a source and credited every time.