Test Original Puzzle Clues for Ambiguity

The AZET team is the team behind AZET, the operator of azet.io. This illustrative guide is for reference and creative exploration, not a demonstrated AZET feature; every puzzle and review example below is hypothetical.

As of 2026-10-12

A homemade clue says, “I have teeth but never bite.” The answer card says COMB. A player writes SAW. Before marking that answer wrong, read the clue again. The writer's intended answer is information the player never received.

Our preference is to put that disagreement on paper before decorating the puzzle. A useful review asks which answers the wording permits, what a hint actually adds, and where the setter still needs judgment. An AI response can supply candidates to examine. It cannot establish that no other answer exists.

Give the clue a small contract

For this hypothetical puzzle, the setter wants a four-letter everyday object, singular noun, with no hidden-letter trick. Those rules belong beside the clue if players need them. Rules left in the setter's notebook do no work for the solver.

Start a review card with five fields: clue wording, visible rules, intended answer, plausible alternatives, and revision decision. Keep the answer separate during the first pass. Otherwise a reviewer may explain why COMB fits instead of looking for rivals.

“Everyday object” still leaves room for interpretation. Do not silently narrow it to bathroom objects after someone answers. If the category matters, state it before play.

Work through the hypothetical rival answers

The first version has no length instruction. SAW is a reasonable candidate because its teeth do not literally bite. ZIPPER deserves discussion too. Neither becomes a bad answer just because it differs from the key.

Adding “four letters” excludes those spellings, but FORK still invites examination. A fork's prongs might be called teeth in ordinary speech. Whether that usage is acceptable depends on the puzzle's declared language rules, not on how confidently a generated review rejects it.

A revised clue might read: “A four-letter tool drawn through hair, with teeth along one edge.” COMB is now a more directly supported answer. This wording also makes the puzzle easier. That is a deliberate tradeoff between surprise and specificity, not an invisible improvement on every dimension.

The revision record could look like this:

Hypothetical clue review, version 2
Visible rule: four-letter singular object; literal description.
Candidate answer: COMB.
Earlier rival: SAW; excluded by length and hair-use wording.
Earlier rival: FORK; length fits, intended use does not.
Open question: could another ordinary term fit these exact words?
Decision: try with readers; uniqueness has not been established.

Notice the last line. A short candidate list is a record of a search, not a complete map of the language.

Write hints that disclose one constraint at a time

Keep the first clue and its hints in separate boxes. If all the hints repeat “teeth,” a stuck player receives more text without more direction.

For the hypothetical original clue, a hint ladder could be: “Think of a handheld object,” then “It is used for grooming,” then “Its teeth pass through hair.” Each step changes the available evidence. The final hint is intentionally close to the answer.

Decide when a player may request each hint. Also decide whether taking a hint changes anything about scoring. For an informal homemade puzzle, simply marking “solved with hint two” may be enough. Make that choice before seeing who struggles.

A prompt for a bounded critique

Use a prompt like this with your own draft material:

Hypothetical puzzle review. Read only the clue and visible rules below. Suggest up to six plausible answers, including awkward alternatives. For each, quote the wording that supports it and identify any rule it violates. Separate clear fits from stretches. Do not infer the intended answer or claim uniqueness. After that pass, review the supplied hint ladder for new information and accidental answer disclosure.

Then provide the clue and rules without the key. In a separate pass, reveal the intended answer and ask what revisions would exclude the strongest rival. Keeping the passes separate makes the review easier to inspect, though it does not turn the model into an independent test panel.

Watch for a review that defends the setter

The main failure mode is answer loyalty: every objection gets explained away because the answer card already says COMB. Another is an invented convention, such as claiming that “teeth” can refer only to a grooming tool.

Ask a human reader to solve the revised clue without commentary. Record their exact answer and the phrase they relied on. Show the first hint only if requested. Repeat with a reader who has not heard the earlier discussion.

If an alternative fits the stated rules, either accept it, revise the clue, or explain a previously missing rule and test again. Do not count a solver's agreement after hearing your explanation as a successful blind solve.

FAQ

Can a model prove my clue has only one answer?

No. A generated list may miss alternatives or apply unstated assumptions. Use it to find objections, then retain uncertainty and review the actual wording with people.

Must every homemade riddle have a unique answer?

Only if that is part of your format. You can explicitly invite several defensible answers and ask players to explain them. The answer key then becomes an examples sheet.

What if the clearer clue feels too easy?

Keep the precise version as a later hint, and test an earlier clue with fewer constraints. Preserve fairness by publishing the rules and accepting answers that fit them.

Our October 3 maker introduction records an email waitlist for the AI assistant. To follow the project, visit azet.io for the waitlist and current notices; this article does not verify October 12 availability.