Why Every Quartet and Outlier Puzzle Has Exactly One Answer: Checking Puzzles Before They Ship

Anyone who has played a grouping puzzle has had the argument. You sorted the sixteen words into four groups, the game said you were wrong, and you could make a perfectly good case for your answer. The puzzle was ambiguous. Someone wrote it, checked it against their own reading of the words, and did not notice that a different reading also worked. This article is about how Quartet and Outlier avoid that, which comes down to one rule: no puzzle ships unless it has been shown to have exactly one answer.

Where puzzles come from

Both games are built on a category database: a list of words, each tagged with every category it belongs to. Dolphin is tagged mammal and sea creature. Crane is tagged bird and construction equipment. Hammer is tagged tool and nothing else. Outlier's database is 1,482 words across 99 categories — mammals, musical instruments, gemstones, currencies, planets, spices, tools, weather, continents and so on. Quartet's is broader in the kinds of category it allows, because a good grid needs at least one group that is not a simple noun class: words that precede ball, things that can be broken, anagrams of a colour.

A candidate puzzle is generated by picking categories and drawing words from them. That is the easy part. The generator produces far more candidates than the games contain, and most of them are thrown away by what follows.

The checks Outlier runs

An Outlier round is five words: four from one category and one from another. Two tests decide whether it is fair.

The outlier must share no category with the four. This is the obvious test and it catches the obvious failure. If the intended odd word is dolphin among four fish, the round is invalid, because dolphin is tagged sea creature and so, in the database, are the fish. The odd word has to be odd on every axis, not just the one the setter had in mind.

No three of the four may form a tighter group of their own. This is the subtle test and it is where hand-written puzzles go wrong. Suppose the four are violin, cello, guitar and drum, and the odd one is radio. The intended answer is radio, because the other four are instruments. But violin, cello and guitar are also all string instruments, which makes drum a defensible answer too. Now the round has two odd words and the player who picks drum has been cheated. The check looks at every three-word subset of the four and rejects the round if any of them share a category the fourth word lacks.

Ten thousand rounds pass both tests. The rounds that failed are not in the app, and the difference is what makes a wrong answer in Outlier feel like a lesson rather than a coin flip.

The checks Quartet runs

A Quartet grid is sixteen words in four groups of four, and it has more ways to go wrong. Three checks are applied.

No word may belong to two of the chosen groups. If the grid contains a sea creatures group and a mammals group, dolphin cannot appear in either, because in the database it belongs to both. It can appear in a grid where only one of those groups is present — and that is precisely when it becomes a good decoy, which we will come back to.

No alternative sorting may exist. The checker searches for any other way of dividing the sixteen words into four groups of four such that each group shares a category. If one exists, the grid is ambiguous by definition and is rejected. This search is expensive in the naive form, but the category database makes it tractable: only groupings where all four words share at least one tag need to be considered, and there are far fewer of those than there are ways to partition sixteen items.

No hidden fifth group may be lurking. Even if the only complete sorting is the intended one, a grid can contain four words that share a category the setter did not intend. A player who spots that group first will submit it, lose a life, and feel — correctly — that the puzzle lied to them. The check looks at every four-word combination across the grid and rejects any grid where a combination shares a tag that is not one of the four intended categories.

Decoys are the point, ambiguity is the bug

It would be easy to read all this as an attempt to make the puzzles easy. It is the opposite. The checks are what make it safe to be difficult.

A decoy is a word that looks like it belongs to a second group but, within this grid, does not. Dolphin in a grid that has a sea-creatures group and a mammals group is ambiguous and forbidden. Dolphin in a grid that has a sea-creatures group and no mammals group is a fine, honest word with one home. Dolphin in a grid with a mammals group, sitting next to three fish that belong to some other category entirely, is a decoy — the player's eye groups it with the fish, the checks have guaranteed that grouping is wrong, and the moment the player realises why, the puzzle opens up. Every hard Quartet grid has two or three of these. The difference between a decoy and an ambiguity is that a decoy has exactly one correct reading and an ambiguity has two, and the checks are what tell them apart.

Where difficulty comes from

With ambiguity removed, difficulty has to come from somewhere legitimate, and in both games it comes from the same place: how close the categories are. The word map described in the first article gives every category a position — the centre of its words — and the distance between two category centres is a good predictor of how hard players find the distinction. Tools and fruit are far apart, so a hammer among apples is a beginner round. Crustaceans and fish are neighbours, so a crab among fish is an expert one. Outlier orders its rounds by that distance. Quartet's four groups are colour-coded from easiest to hardest on the same basis, with the hardest group usually a wordplay category that the map does not place well, which is why it comes last.

What the checks cannot catch

The database is the limit. A category the database does not know about is invisible to the checker. If four words in a grid happen to be the names of characters in a novel the setter has not read, the checks will not see the hidden group and a player who knows the novel will find it. Proper nouns, regional usage and slang are all places where this happens, and it is the reason both games lean on ordinary nouns and well-established categories. Feedback from players goes back into the database, and a grid that turns out to have an unforeseen reading is fixed or retired.

That is the honest version of “exactly one answer”: exactly one answer within everything the database knows, which is a great deal and not everything. It is still a far stronger promise than a puzzle written by hand and checked by the person who wrote it can make.