Verbraid

How to play Verbraid

Four secret five-letter words sit in a chain, and the last letter of each word is the first letter of the next. Solve all four and you have unwoven the braid. A new one appears every day, the same for everybody, and it changes at your local midnight.

Type a five-letter word and press Enter to play it. Switch between links by clicking them, by pressing 1 to 4, or with the ← and → keys.

Every guess is scored twice

The first score is the one you already know from other word games. A green tile means the right letter in the right place, yellow means the right letter in the wrong place, and grey means the letter is not in that word at all. Repeated letters behave properly: guessing a word with three of the same letter against an answer that has one will light up exactly one of them.

The second score is what makes this a different game. Alongside the tiles, each guess gets a heat reading — how close your word is to the answer in meaning, from Freezing up to Scorching. The two scores are independent. A guess can come back with five grey tiles and still be the most useful move you have made, because Hot tells you that you are thinking about the right kind of thing.

One pool of guesses for all four boards

You start with nine guesses in total, not nine per board. Every link you crack hands two back, so an efficient run has roughly fifteen to work with and a slow one runs dry with boards still hidden. This is where most of the real decision-making lives: not only what to guess, but which board to spend it on.

The chain does the work

Because each answer's last letter is the next answer's first, cracking any link pins a letter on both of its neighbours — the link before it learns its final letter, and the link after it learns its opening one. Information travels in both directions, which means attacking the boards out of order is a genuine tactic rather than a mistake.

A pinned first letter is typed in for you at the start of each guess and shown on a dashed tile to mark it as given rather than chosen. It is a head start, not a constraint: ⌫ clears it whenever you want to ignore it.

Every board tells you its category

Each link shows a clue naming the kind of word it is hiding — "a bird", "clothing", "a number". It narrows the field to a rough category and never to a single word. The clue is what makes a pool of nine playable instead of a lottery, and the two were tuned against each other: before clues existed, the pool was twelve.

Three tactics that actually help

1. Buy information with a deliberately wrong word. Guesses are not required to respect the letters you already know. Throwing a word with the right meaning and the wrong shape at a board tells you a great deal through the heat reading alone. The shared pool is what stops this being free.
2. Solve the cheapest board first. Two guesses come back for every link you crack, so an early easy solve effectively funds the rest of the puzzle. A board whose category you recognise immediately is worth attacking even if it is not the one you find most interesting.
3. Read the clue before the tiles. The category rules out most of the dictionary in one step. Letters narrow a large field slowly; a category narrows it instantly.

Where the heat reading comes from

There is no server here and no language model. Meaning is worked out from a hand-authored map of roughly a hundred categories covering about eight hundred common five-letter words. Each word is filed under the categories it genuinely belongs to, and because those categories nest, relatedness falls away gradually rather than all at once. A zebra and a horse are both mammals; a zebra and an eagle are both animals; a zebra and a table have nothing in common.

Pure hierarchy was not enough by itself. It scored a piano against music at zero — one is an object, the other an art form — so the model also carries categories that cut across the hierarchy and group words by the situation they belong to: musical, nautical, culinary, clinical. That is why a guess can read Warm against an answer it shares no obvious category with.

The model is written and corrected by hand, which means it has errors in it. If a clue does not fit its answer, that is a real bug in the data and worth reporting — several have been fixed that way already. There is a fuller explanation on the about page.

Looking for something different?

Weave is the other daily game here. Sixteen words hide four groups of four, and unlike most grouping puzzles a wrong answer still tells you something: every miss is scored on how closely related your four words actually were.