Names and fair play: where the cast comes from and why it is set the way it is

The names in a grid are not real people and no case is based on anyone. But how they are generated, and how they are typeset, decides whether the puzzle is fair — so both are worth explaining.

Matched pairs, not a shuffle

A first name and a surname are drawn together from the same naming tradition, so the cast reads like a population register rather than a random collision of syllables. Around seven thousand distinct full names exist per language, and at least three-quarters of any grid comes from the language's own majority tradition.

Each pair is drawn from one tradition, so first name and surname belong together.

This matters for the puzzle as much as for the reading. A grid where surnames and first names were shuffled independently would make the origin clue meaningless and would make every other clue feel arbitrary.

A tradition describes a name, never a person

One clue type asks whether the culprit's name comes from a particular naming tradition — a Dutch name, a Turkish name, a West African name. That is a statement about the name in the grid and nothing else. There are no nationalities, no biographies and no people in this game: there are strings drawn from pools, and the pools are grouped by where the naming convention comes from.

The phrasing in the game follows that distinction carefully, and the about page says the same thing in the same words. It is a design constraint, not a disclaimer.

Why the names are set in Atkinson Hyperlegible

In most sans-serifs a capital I, a lower-case l and a digit 1 are the same vertical stroke.

A clue may say no L in the name while a card shows a capital I. In most typefaces those are an identical stroke, and the puzzle is then lost to typography rather than to reasoning. Atkinson Hyperlegible was designed for exactly this problem — every ambiguous pair of characters is given a distinct shape.

The font is self-hosted rather than loaded from a font CDN, so reading a page does not announce you to a third party. The rest of the site's headings stay in a serif; only the names and the clues use the legible face.

Bigger pools, fewer repeats

The pools started far smaller, and it showed: a grid of a hundred names would repeat the same first name eight times, and every dossier read like the same handful of people. Widening the pools to thousands of combinations took the worst repeat down to about four in a hundred and lifted distinct first names from roughly a third of a grid to nearly three-quarters.

Some repetition at the top of the ladder is unavoidable. A grid of 250 names that is three-quarters one tradition cannot be filled from forty first names per gender without a name appearing twice — and a grid that quietly dropped the majority share to avoid that would be a different, less honest cast.

What stays fixed

Two guarantees hold regardless of the pools: no full name is ever dealt twice in the same grid, and the middle initial is drawn per name rather than per batch, so a grid never reads as everyone is an A. Both are enforced by tests rather than by inspection.

Generated names, and why they are not random

Procedural generation is easy to do badly. Sample letters at random and you get names no human would carry; sample from one narrow pool and every case starts to feel like the last one. The interesting constraint is that a generated name has to be both plausible and varied enough that a clue about it is informative.

Games have wrestled with this for decades — the procedural generation tradition runs from roguelikes to open worlds, and the recurring lesson is the same: the generator needs rules about what it may not produce, not just what it may.

Here that means separate first-name and surname pools per language, real particles and occupational names in the proportions those languages actually have, and a check that no clue depends on a property only one name in the grid could possibly have.

See a cast in a real case

Where the names come from