The Puzzle You Solve Is the Last Draft

A good puzzle arrives with no visible seams.

The grid is clean. The instructions fit in a paragraph. The answer appears to have been waiting inside the clues all along. Even a particularly devious turn can feel inevitable once you have seen it. Of course that was what the title meant. Of course the empty squares made a map.

The maker’s desk tells a less dignified story. There are rejected themes, duplicated answers, clues that point two ways, rules that made sense only to the person who wrote them, and a test solver who spent twenty minutes studying the wrong corner. Sometimes the elegant final mechanism began as three less elegant mechanisms standing on one another’s shoulders.

Puzzle making is often described as an act of inspiration. Publication is mostly an act of revision.

The first draft knows too much

Every puzzle author begins with a private disadvantage: they know the answer.

That knowledge reaches backwards. It makes an ambiguous instruction look precise and a large leap look like a small one. If the intended route is to take every third letter, the author sees those letters almost glowing on the page. A new solver sees rows, colours, word lengths, punctuation and perhaps an acrostic that happens to spell something tantalising but useless.

This is the central difficulty of constructing for other people. The author must arrange a discovery, then forget enough to judge whether it can actually be discovered.

Crossword constructors meet the problem one clue at a time. They know that BARK is supposed to mean a dog’s noise, so the tree feels like a neat diversion rather than a competing answer. As A Crossword Clue Is a Tiny Promise explains, the misdirection is allowed only if the clue still accounts for the answer. The constructor can check the machinery. They cannot, on their own, experience the surprise.

Logic-puzzle authors have the same problem at a larger scale. A deduction may be valid yet practically invisible. A rule may be concise yet invite a plausible misreading. A puzzle can contain everything required and still fail to put those things where a mind would think to look.

That is why a serious test begins before anyone asks whether the answer is correct. The first question is simpler: what does an unbriefed person actually do?

Someone has to get stuck on your behalf

In an ordinary solving session, being stuck is the problem. In a test solve, it is evidence.

The 2021 MIT Mystery Hunt writing team used a deliberately elaborate process for its roughly two hundred puzzles. A first idea was assigned two editors before it was fully written. Only after their approval did it reach test solvers, who might solve it or fail, then record feedback and rate both difficulty and fun. A troubled puzzle returned to revision. The team aimed for two clean test solves before the work moved on to solutions, web production, fact-checking and copy-editing.

That sequence matters. Test solving is not a ceremonial run-through performed after all the interesting decisions are over. It is where the author’s imagined solving path meets an actual mind.

A tester who takes a wrong turn can reveal an accidental pattern stronger than the intended one. A group that reaches the answer without using half the material may have found a shortcut. Three people who understand a rule in three different ways have not displayed charming personal styles; they have found a sentence that needs rewriting.

The useful moments are often undramatic. A pencil stops. Someone rereads the title. A solver returns to a step the author considered obvious and tries to make it mean something else. It can be hard for a maker to sit through this without offering a rescue. Yet an early hint spends the very information the session was meant to collect.

The tester is not a specimen, of course. One person’s blind spot proves little. Patterns across several sessions are more persuasive: the same mistaken interpretation, the same abrupt halt, the same late realisation that produces relief in one group and annoyance in another.

Editing means touching the puzzle

People often picture an editor correcting spelling around the edge of a finished object. Puzzle editors sometimes do that. They also ask for a new corner of the grid.

A standard 15-by-15 American crossword has roughly 75 clues. Raymond Simon, writing about his work for Games World of Puzzles, describes checking each for factual errors, house style, suitability for the intended audience and the small traps created by obscure fill. At a large crossword desk, acceptance is still not a promise to print the submission verbatim. Clues are rewritten; dated references are replaced; weak entries may force a section of the grid to be rebuilt.

This is not interference with the puzzle’s pure original form. There may be no such form. Theme, grid, fill and clues press against one another. Improve a crossing answer and two clues change. Remove a stale reference and the difficulty shifts. Repair one unfair square and a prized piece of wordplay may lose its home.

An editor also represents the publication’s memory. The constructor knows this puzzle intimately; the editor knows the last fifty. They can notice that a delightful device appeared three issues ago, that a supposedly easy clue assumes rather a lot of opera, or that a punchline depends on a cultural reference the audience is unlikely to share.

The work is less like polishing a stone than adjusting a small clock while it is running.

Difficulty has no dependable dial

Puzzle difficulty is usually printed as a tidy label: easy, medium, hard. Behind it sits a rather untidy sample of human beings.

Solvers bring different vocabularies and habits. One spots hidden words instantly and misses every homophone. Another has met the logic technique but not the theme. A tightly practised team can make a puzzle-hunt clue look trivial by dividing the work six ways. None of them is a universal measuring instrument.

Editors therefore look beyond elapsed time. Where did progress happen? Did one discovery unlock the next, or did the solver have to start a complete search again? Was the long pause spent on the intended hard step? Did the answer produce the clean click of recognition, or merely end the ordeal?

This is the workshop version of why some puzzles feel fair. Fairness is experienced by the solver, but it is built from decisions: which information arrives first, which ambiguity remains, how much confirmation a correct idea receives. Difficulty can survive revision. Pointless uncertainty should not.

Nikoli makes the same distinction when it defends its handmade logic puzzles. The Japanese publisher says that its editors review every submission and reject work even from experienced creators. Its case for handmade Sudoku is not that computers cannot produce valid grids. It is that an author should shape the order in which a solver encounters resistance and release. The Quiet Architecture of a Sudoku lives in that order, not merely in the completed square.

The claim comes with a publisher’s understandable pride. The practical insight is sound: uniqueness can be verified; pleasure has to be observed.

The wastepaper basket is part of the method

Revision is easy to admire in the abstract and unpleasant in the particular.

The clue that took longest to write may be the one that goes. A visually beautiful grid may depend on an answer nobody can reasonably infer. A puzzle-hunt mechanism can be ingenious, thoroughly tested and still make the whole event too long. Editorial work includes recognising when effort already spent is not a reason to spend a solver’s time as well.

Nikoli’s description of its own process is unusually blunt: even experienced authors are rejected. The MIT account is gentler but equally revealing. Ideas move through review, writing, testing, revision, fact-checking and copy-editing; at several points, the puzzle can go backwards. The arrows on a real production chart do not all point towards publication.

What survives is rarely the work of a solitary flash. It is the constructor’s idea after editors have worried at it, testers have misunderstood it, facts have been checked, instructions have been shortened, and somebody has noticed that the clever title gives away the answer.

The solver should not feel this machinery. They should meet a puzzle that is clear enough to enter, stubborn enough to resist and precise enough to reward attention. All the discarded versions have done their job when the final one feels as though it could never have been otherwise.

By publication day, the comments are resolved and the red ink is gone. That is the editor’s neatest trick.

Sources and further reading

• CJ Q., “Two hundred puzzles, fifty weeks later”, MIT Admissions, 6 February 2021, on the editorial, test-solving and production pipeline for the 2021 MIT Mystery Hunt.
• Nikoli, “Nikoli Puzzles”, on collaboration between creators and editors, submission review and rejection.
• Nobuhiko Kanamoto and Maki Kaji, “Why hand made”, Nikoli, on designing a Sudoku’s solving process.
• Raymond Simon, “More than a clue”, ACES: The Society for Editing, 6 May 2021, on crossword editing and audience calibration.
• Deb Amlen, “On Crosswords and ‘Wordplay’”, With a Side of Knowledge, University of Notre Dame, on theme, grid, fill, clues and working with editors.

Image credit

Photograph by cottonbro studio via Pexels. Used under the Pexels License. The hosted image is resized and compressed by Pexels; no editorial alterations were made.