Didn't Do It · 1 September 2026 · 8 min read

The fair-play rules of detective fiction

Lees in het Nederlands

In the 1920s, detective fiction had a credibility problem: authors kept producing solutions the reader could not possibly have reached. The genre's answer was to write down rules. A century later those rules are the specification for a fair puzzle.

How the generator accepts or rejects a clue chainWHAT THE GENERATOR CHECKS BEFORE A CASE SHIPSEnds on exactly one nametwo survivors means the chain is unfinishedEvery clue removes someonea clue that changes nothing is noiseNo clue is redundantdrop any one and more than one name must remainClue tiers match the trackan easy grid never asks you to count vowelsA chain that fails any of these is thrown away and rebuilt — never softened.

The problem the rules were written for

A mystery is a contract. The writer promises that the answer is discoverable; the reader promises to pay attention. Break the first half — an unmentioned twin, a poison unknown to science, a confession from nowhere — and the reader has been reading a magic trick, not a puzzle.

The complaint became loud enough that the writers themselves codified it. What is striking, reading the lists now, is how many of the rules are really one rule: no information may exist that the reader cannot have.

Knox's Decalogue

Ronald Knox published his ten commandments for detective fiction in 1929. Several are of their time and one is openly racist, which is worth naming rather than quietly skipping; the text and its context are easy to find. The load-bearing ones are these:

Strip the period detail and what is left is a definition of solvability: the answer must be in the candidate set, the reasoning must be available, and nothing may be hidden.

Van Dine's twenty rules

S. S. Van Dine's Twenty Rules for Writing Detective Stories (1928) is stricter and more mechanical. Rule one is the whole philosophy: the reader must have equal opportunity with the detective for solving the mystery. All clues must be plainly stated and described.

Van Dine also insists on exactly one culprit and one solution. That is not a literary preference; it is a uniqueness constraint, the same one a logic puzzle needs to be well-posed.

The Detection Club oath

The Detection Club, founded in 1930 by Christie, Sayers, Chesterton and others, made new members swear an oath promising that their detectives would solve crimes by "wits which it may please you to bestow upon them and not placing reliance on nor making use of Divine Revelation, Feminine Intuition, Mumbo Jumbo, Jiggery-Pokery, Coincidence or the Act of God".

It is a joke with a serious core. Every item on that list is a way of producing an answer the reader could not have produced.

The three rules that survive translation to a game

  1. Completeness. Everything needed is on screen. No off-screen fact, no hidden attribute, no clue that arrives after you commit.
  2. Truth. Every clue is true of the culprit, without exception or trick wording. A clue that is 'mostly' true is a lie with better manners.
  3. Uniqueness. Applying all the clues leaves exactly one candidate — never two, never zero.

Those three are checkable by a machine, which is the difference between a novel and a puzzle generator: this game verifies them for every case before you see it. The details are in the one-survivor guarantee.

Where the game deliberately breaks with the tradition

Golden-age mysteries hide the culprit in plain sight among suspects with motives. A generated puzzle has no motives, no alibis and no psychology — only names and properties. That loses the human texture and gains something the novels could never offer: a fresh, provably fair case every single day.

The other break is in the wording. Knox's detective withholds; here every clue is stated as innocence, which means the game can never mislead you by omission — it can only be read too fast. More on that in why crossing out works.

See a fair chain in action

Related reading