Loading…
Sixteen words are laid out in a grid. They hide four groups of four. Find all four groups before you make four mistakes. A new grid appears every day, the same for everybody, and it changes at your local midnight.
Pick four words that you think belong together and press Submit. Get it right and the group is revealed along with the category that ties it together. Get it wrong and it costs you one of your four mistakes.
The arrow keys move around the grid, Space picks a word, Enter submits, and Esc clears your selection. The Shuffle button rearranges the tiles without costing anything, which genuinely helps — a set you cannot see in one arrangement often jumps out in another.
This is the part that makes Weave different from other grouping games. A miss is not simply wrong. Every wrong submission comes back with a coherence reading: how closely related your four words actually are to one another, expressed as a percentage of how related a real group on that grid is.
The reading comes from the same hand-authored model of meaning that powers the heat gauge in Verbraid. It is not a hint about which group you should try next; it is a measurement of the guess you actually made.
The four groups are ranked by difficulty, which is what the four colours mean once they are revealed. The ranking is not decorative. Each group contains one word chosen precisely because it looks like it might belong to a different group on the grid — it scores highest in meaning-similarity against the other three.
That word never actually belongs elsewhere. Every word on the grid has exactly one correct home, and the generator enforces that before it picks a single word: a word that genuinely fits two of the four categories would give the grid two correct answers, which is not a hard puzzle but a broken one. The pull is real, the ambiguity is not.
There is no server and no editor. Grids are generated from a hand-authored map of roughly a hundred categories covering about eight hundred common five-letter words — the same model that scores Verbraid's guesses on meaning.
That is the reason this game can exist at all. A grouping puzzle normally needs a person to write a fresh one every single day, which is why the genre has so few good imitators. What a generated puzzle gives up is wit: a model of meaning produces categories, not puns. What it gets back is a mechanic no hand-built grouping game has, because measuring how close a wrong answer was requires a number, and a human editor does not have one.
The model is written and corrected by hand, so it contains mistakes. If a group seems unfair, or a word looks like it belonged in two places at once, that is a real bug and worth reporting. The about page explains the model in more detail.
Verbraid is the other daily game here. Four secret words sit in a chain where each one's last letter is the next one's first, and every guess is scored twice: once on spelling and once on meaning.