How Our Puzzles Are Made and Checked

What does every puzzle have to get right?

Four things. The board must be fresh, so Next puzzle rarely feels like a repeat. It must be solvable. It must land near the level you chose. And the same puzzle number must give the same board every time, so a shared link works. Each requirement is easy alone. The work is meeting all four at once, which is why each game has its own checks.

Why isn't a board just random?

A random Queens grid can have no solution or several. A random Zip line can leave cells stranded. A random Waffle grid may not spell six real words. Chance gives variety, but rules give quality. So our generators build a board from a valid answer, then test the result and throw it away if it fails.

From puzzle number to board

Every unlimited puzzle has a number such as M-7Q2K: a level letter and four characters, about a million numbers per level. Daily boards use the date on your device instead, like M-D20261107. The game turns the number into a seed. A 32-bit hash of the text feeds a seeded random generator, and the board code never uses chance outside it. The same number always gives the same board, which is why a shared link opens exactly what you were playing.

Which games build boards and which draw from a set?

Some games build every board from scratch: Pips, Queens, Tango, Zip and Colorfle. Others draw from a prepared set and shuffle it. Waffle holds 1,500 finished grids, and each puzzle number scrambles one differently, as in the Waffle swap grids. Connections holds 300 puzzles, 100 per level, so familiar sets return with the words dealt in a new order. Tradle, Factle, Costcodle, Betweenle and Colordle draw from data tables, and Size It Up pairs things from a library of well over a hundred objects.

Checking that a board can be solved

Queens plants the queens first, grows the regions around them, and counts solutions with a solver. It returns a board only when the count is exactly one. In Queens, tests count the solutions of 150 boards, and a separate brute-force count rechecks the smaller ones. Every one of the Queens boards here passes that count. Tango removes clues one at a time and keeps each removal only if plain logic still finishes the board, so you can play Tango with no daily limit and never guess. Zip begins with a line through every cell, so a solution always exists, and the Hard walls never cross it; see the hedge-wall boards in Zip. In Waffle, the minimum swap count is calculated exactly and checked against a brute-force search. Connections puzzles go through a validator that rejects near-duplicates, then a review round that rewrote 47 of the 300 where a word could fit two groups or a fact was off.

When is more than one answer fine?

Queens and Tango have one answer each. Pips and Zip do not: any layout that satisfies every rule wins. A Zip hint follows the route we stored. If your own line has drifted off it, the hint trims the line back to where it still matches and adds one cell.

Setting the difficulty

Difficulty is a set of numbers per game, not a feeling. Pips uses 4, 7 or 10 dominoes. Queens grows from 6 by 6 to 10 by 10. Waffle gives 12, 15 or 12 swaps, with a minimum of 6, 10 or 10. Zip needs at least 55% of its cells to be bends and no straight stretch longer than 4. Factle requires ranks one to six to differ by at least 1.5%. Size It Up starts the shape at least 1.5 times off the truth.

Testing the hints

In most games, each press of Hint moves you exactly one step forward, and the tests press it until the board is solved. Queens finishes in 6, 8 and 10 hints, and Connections in 16. Hints in Tradle, Factle, Betweenle and Waffle cost no guess. After three hints, or the last one a game offers, a Show answer button appears. A board where you showed the answer does not count toward your streak or best time.

Where does the data come from?

Tradle uses export figures from Growth Lab at Harvard's CC0 release, derived from UN Comtrade data, with country positions from Natural Earth. Factle uses World Bank indicators under CC BY 4.0 and state figures from the U.S. Census Bureau. The prices in Costcodle's price records come from Open Prices by Open Food Facts under the ODbL. Size It Up measurements come from reference pages, each stored with the exact quoted line it came from. Colordle names come from the xkcd color survey and the color-name-list project.

Cleaning word lists and color names

Waffle and Betweenle take five-letter words from the SCOWL word lists. That list has no offense flag, so we keep our own block list for slurs, swear words, and sexual and violent terms. Waffle also drops plurals and past tenses of shorter words. Colordle runs a similar block list, plus a list of brand names. Betweenle's hidden words come from the common end of the list, and the harder levels widen it.

What do the numbers in trivia games mean?

They are records, not promises. Each Costcodle price was typed in by a contributor at one store on one day, so it may differ from what you pay, and the page shows the city, state and month. Tradle shows shares of classified exports for one year. Factle shows the real numbers, the year and the source when a question ends. Factle only asks a question when every member of the group has a value.

What tests run before a game ships?

Each game has its own test file. They check that the same seed gives the same board over 50 seeds on every level, that the stored answer passes the solved check, and that a board survives being saved and reloaded. Generation must take under 50 milliseconds at the 95th percentile. Counts run from 16 tests for Zip to 37 for Colordle. Data tables are re-validated on every import, which stops on any error.

Reporting a mistake

Use the contact page. Tell us the game, the level and the puzzle number from the link, and what looked wrong. A wrong measurement, price or word is a row in a table, so it can be corrected. For other picks, see our guides to daily games like Wordle and daily guessing games.