ThePuzzleLabs
The DailySunday, September 6—day
streak
New in—:—:—Daily Hub →

Eliminate the Impossible: Sherlock Holmes's Method as an Actual Technique

Red-Handed guide · 5 min read

Sherlock Holmes gets quoted a lot and used rarely. The lines about eliminating the impossible and about never guessing usually turn up as atmosphere, printed over a deerstalker on a puzzle box.

They are better than that. In a properly constructed deduction puzzle, they are not flavour, they are the algorithm.

Rule one: eliminate the impossible

The famous version is that when you have eliminated the impossible, whatever remains must be the truth, however improbable.

Read as advice about life that is close to useless, because real situations do not come with a finite list of possibilities. Read as a solving procedure it is exact, and it tells you three things.

It tells you what to enumerate. Not "who seems guilty" but the complete list of candidates. In a Red-Handed case that is every suspect, because any of them could be the one who lied. Six suspects, six hypotheses, and you must be willing to entertain all of them including the ones that feel absurd.

It tells you what counts as progress. Ruling somebody out is not a consolation prize for failing to find the answer, it is the mechanism. Three eliminations on a four-suspect case is a complete solve. If you only feel like you are getting somewhere when you confirm something, you will find this style of puzzle frustrating for no good reason.

It tells you the standard. Eliminated means impossible, not unlikely. A suspect is cleared when assuming their guilt produces a flat contradiction, such as two people in a one-chair room. "It doesn't feel like them" is not an elimination.

What makes this work on a puzzle rather than on life is that the list really is finite and the contradictions really are decisive. That is a property somebody has to build in deliberately, which brings us to the second rule.

Rule two: never guess

Holmes calls guessing a shocking habit, destructive to the logical faculty. This is the rule people skip, and it is the one that does the structural work.

A puzzle that permits guessing is a different object from one that does not, even when both have exactly one answer. Uniqueness is a weak property. It says the answer exists and nothing else fits, but it says nothing about whether you can get there by reasoning.

You can build a puzzle with exactly one solution where the only way to find it is to try something, follow it a long way, hit a wall, back up, and try something else. Uniqueness is satisfied. The experience is search, not deduction. And you cannot tell the difference by looking at the puzzle, which is why this matters.

What that guarantee costs

The guarantee we make is stronger than uniqueness. Every case is checked to confirm that assuming any wrong suspect dies of a plain contradiction, and that the surviving assumption then completes the board on its own without further assumptions. No nesting, no guess inside a guess.

That check is expensive to satisfy. When we first ran it against a batch of cases that already had verified unique answers, 29% failed. They were sound puzzles by the usual standard and they were not solvable the way Holmes describes, and we threw them away.

It is also why the harder tiers take so long to build. Expert cases accept at roughly one in every 25,000 candidate boards generated, and most of the rejects die on exactly this test.

The practical consequence for you is a small one, and it is the whole point: if you are stuck, there is a deduction available. Not maybe. You have not run out of information, you have missed a step. That converts being stuck from a reason to quit into a reason to look again, which is a completely different feeling.

The method, step by step

Put the two rules together and you get a procedure.

  1. Start with what cannot be disputed. The forensic lines come from investigators, not suspects, so they cannot be lies. Place everything they settle before reading a single account.

  2. Enumerate the candidates. Every suspect, including the boring ones.

  3. Assume one is guilty and follow it. That puts them alone at the scene and makes every other account true, because only one person lies. Push those facts until either something collides or the board completes.

  4. Cross off and move on. A contradiction is a completed piece of work, not a wasted attempt.

  5. Refuse to guess. If two hypotheses are both still alive, you have missed a deduction inside one of them. Go back rather than picking.

The one addition Holmes does not give you is the shortcut: if two accounts cannot both be true, the liar is one of those two and everybody else is telling the truth. That single observation promotes four or five statements from uncertain to certain at once, and it is worth more than any individual clue in the case.

Why the top tier is called Sherlock

The hardest cases here make two accounts false rather than one. One liar was alone at the scene and did it. The other lied for reasons of their own and was somewhere else entirely.

That is the version where the method stops being optional. Six suspects give thirty ordered pairs to eliminate rather than six candidates, and the leverage of "everybody else is telling the truth" weakens from five accounts to four. You cannot hold it by feel. You have to enumerate, assume, contradict, discard, and keep going.

The tier is named for the standard it is held to rather than the decor. Every case on it is verified reachable by reasoning alone before it ships, which means the deerstalker is doing no work at all. The method is.

Try it on Sherlock, or build up to it from Easy. The rules guide has all seven techniques, and how to solve who-is-lying puzzles works a full case end to end.