Launch somethingGuide

Create a print-and-play logic puzzle with one checked solution

Draft a small original puzzle, count every possible solution and keep the answer key separate from the player sheet.

Illustrative resourceNot tool-testedPeople creating a small printable puzzle for personal play or a fictional activity pack

What you’re working towards

A complete fictional four-person logic puzzle, printable HTML player sheet, separate explanatory answer key and a local validator that finds exactly one solution. The fixture is checked by enumeration; the consumer AI workflow and browser/physical-print behavior remain untested.

Download playbook .md

Your choice also saves this guide’s current worked-example checkmarks. Only this browser; no account or sync. View saved guides

Saving needs JavaScript. Reading and downloading this guide does not save it.

Included example · illustrative HTML file

Download the printable puzzle

Complete script-free fictional player sheet with five clues, explicit rules, three solving grids and a blank answer table. The solution is not embedded. Browser and physical-print checks remain unperformed.

Not tool-tested. Review and adapt before using it for a real project.

Open example Download example

Authored puzzle-sheet preview titled The cabinet of small wonders. It shows rules, five clues, people and objects, four quarter-hour arrival times, three blank logic grids and a blank final-answer table. No solution appears. — Inspected code-native page preview. Browser rendering and physical-print checks remain unperformed.

Source files to work with

Open the answer key (spoilers)
The complete matching and a five-step deduction path, with exact reproduced rules and clues. Keep separate from the player sheet.
Download the puzzle brief
Complete fictional design brief, rules, authored clues and verification boundaries; no answer table.
Download the optional validator (spoilers)
Python standard-library checker with two exhaustive enumerators and exact-edition mutation tests. Its diagnostic controls can reveal solution details. Inspect before running; the default reads companion files and prints a report.
Download the scoped check report (spoilers)
Machine-readable counts, mutation outcomes, hashes and unperformed checks. Diagnostic details can reveal parts of the solution; keep separate from the player sheet.
Download the layout preview
Inspected code-native 794 × 1123 preview of the blank puzzle sheet. An authored illustration of the layout, not a browser screenshot.
Download the editable layout preview
Self-contained original SVG for the same puzzle-page layout preview. Contains clues and blank grids; no answer.

Before you begin

A bounded puzzle brief: categories, exact one-to-one rules and an intended audience
Fictional names and objects or permission to use supplied material; no real personal data or franchise characters are needed
A place to run and inspect a small exhaustive checker, or help from someone who can; a fluent written answer alone is insufficient
A browser and printer only if you want to check and print the HTML player sheet

Possible tools

A text AI assistantA user-chosen text interface · required
Access and plan limits unverified
No named product, account plan, coding runtime entitlement or free route is verified. Use tools you already have access to; do not assume an AI assistant can execute code or inspect a browser.
A Python 3 runtimeA user-chosen local or authorized coding runtime · required
Access and plan limits unverified
No named product, account plan, coding runtime entitlement or free route is verified. Use tools you already have access to; do not assume an AI assistant can execute code or inspect a browser.
A plain-text editorA user-chosen local editor · required
Access and plan limits unverified
No named product, account plan, coding runtime entitlement or free route is verified. Use tools you already have access to; do not assume an AI assistant can execute code or inspect a browser.
A browserA user-chosen browser opening the local HTML file · suggested
Access and plan limits unverified
No named product, account plan, coding runtime entitlement or free route is verified. Use tools you already have access to; do not assume an AI assistant can execute code or inspect a browser.

Use the sample first. Remove private information before taking a prompt into another tool.

Make a start

One step at a time.

5 steps

Start with the synthetic worked example or other non-sensitive inputs. Run one prompt at a time in your chosen AI tool. Replace any bracketed placeholders with the requested material, check each reply, and supply the source or earlier outputs that step needs. Paste them again if the tool lacks that context.

Set up the worked example

Open: Open a fresh text-AI chat and a plain-text editor. You also need an available, permitted Python 3 runtime to establish a solution count; a chat that only returns code has not run it. Python is a programming runtime, not a feature every AI chat provides. A browser is needed only for your later local page/print-preview check. Keep spoiler files separate from the player sheet.
Add the sample: Download and open small-wonders-source-brief.md; it adds the fixed five Authored clues to the design brief in ‘The starting point’ (Synthetic input). Paste the complete source brief into step 1. For a replay of the supplied edition, keep those exact rules and clues through steps 2–5 and compare the separate design answer with small-wonders-answer-key.md only when ready for spoilers. In step 2 add ‘For this worked edition, retain the source brief’s five Authored clues verbatim.’ If different clues are drafted, label it a new unverified edition and do not apply the sample’s fixed checker or removal counts. Replace later placeholders with the actual specification, clues, separate key and checked files. ‘The intended result’ (Illustrative output), HTML, PNG and existing report are references; the PNG is a code-native preview, not a browser screenshot. For the optional advanced sample exercise, paste the complete baseline rules, exact visible clues, separate key, checker source and actual complete-solution report into its placeholder, even if they appeared earlier. If starting a new chat, supply all of them again; if any is missing, retrieve it or keep that exercise pending rather than infer the earlier result.

Start with the synthetic example, then adapt each prompt to your task.

Fix the rules before writing clues

Choose small, explicit categories and say whether every value is used once. Four people, four objects and four times create a manageable 576-assignment space.

Your starting prompt

Turn this brief into a precise logic-puzzle specification. List every category and value, state the one-to-one rules, define any time/order language, and identify assumptions that would make a clue ambiguous. Use only fictional non-sensitive details. Do not draft the player sheet yet or claim a unique solution. Keep any proposed hidden target answer in a separate design section.

PUZZLE BRIEF:
[paste the complete brief]

Expect: A precise specification for four fictional people, four objects and four times, with one-to-one assignment rules and no uniqueness claim yet.
Check: The people are Aya, Bram, Cleo and Dax; objects are compass, lantern, ribbon and shell; times are 09:00, 09:15, 09:30 and 09:45 on one fictional morning. Each object and time is used once. Four factorial means 4 × 3 × 2 × 1; 24 object orders × 24 time orders gives 576 candidates.
If it differs: If it adds another time, ambiguous ‘shortly after’ language or outside knowledge, return to the exact brief. Keep any hidden target answer separate and marked as spoilers; choosing one target is not proving uniqueness.

Draft clues from a complete target answer

A target answer helps you write consistent clues. It does not establish that the clues exclude every other answer. Keep it out of the player version.

Your starting prompt

Choose one complete assignment satisfying this specification and label it DESIGN ANSWER: SPOILERS. Draft a small set of precise relational clues that are all true of that assignment. State every clue in plain language and also as a proposed predicate. Prefer exact minute differences where needed. Include no clue based on unstated knowledge. Return the design answer, clues and predicates separately. Do not say the puzzle is unique until exhaustive checking has been run.

SPECIFICATION:
[paste]

Expect: A spoiler-labeled complete design answer and a separate list of clues with proposed code conditions. For this worked edition, the five source clue sentences stay unchanged.
Check: Compare the five precise relations: lantern 30 minutes before compass; Cleo 15 minutes after compass; Bram before Dax; Aya neither lantern nor shell; shell before ribbon. Check each is true of the design answer, then compare with the separate reference key if needed. This consistency check does not rule out other answers.
If it differs: If a clue or category changes, stop treating it as the supplied edition. Either restore the exact source for this replay or use the own-input route to adapt both checkers, key and all files together. Never repair a contradiction by hiding an extra clue.

Count all solutions, then check the key

Run the supplied checker for the included edition. For a new puzzle, write and inspect new predicates and enumerate the full space; stop neither at the target answer nor at the first match.

Your starting prompt

Build a Python standard-library checker for this bounded puzzle. Enumerate all one-to-one assignments and count every solution. Independently express the clues a second way, using a different representation, and compare the complete solution sets. For the four-by-four-by-four fixture, each implementation must examine all 4! × 4! = 576 candidates. Reject counts of zero or greater than one; compare a unique result with the supplied key. Test deleting each clue, adding a known contradiction, changing a key while preserving one-to-one assignments, and changing visible clue wording. Return runnable source, exact run instructions and actual results only if you executed it. If execution is unavailable, explicitly say the count is unverified.

RULES, CLUES AND SEPARATE DESIGN ANSWER:
[paste]

Expect: Runnable checker source and instructions, plus actual complete solution-set results only when code was really executed. Otherwise the solution count remains unverified.
Check: For the supplied edition, inspect small-wonders-validator.py and place it with the downloaded HTML, answer key, source brief, SVG and PNG in one local folder. If Python 3 is available, run python3 -I small-wonders-validator.py from that folder; it reads sibling files and prints a report by default. Two separately expressed enumerators each inspect 576 assignments. Baseline count is 1; deleting clues 1–5 gives 15, 12, 2, 3 and 3, and the stated contradiction gives 0. These are fixed-source checks only.
If it differs: If no permitted runtime is available, leave execution and uniqueness unverified and get help from someone who can inspect/run it. If it fails, save the failure before any edit. Do not overwrite originals, use --report-file on a source artifact, or claim an existing reference report is a new run.

Lay out the puzzle and separate the spoilers

Use the included HTML and Markdown as complete examples. The player sheet needs room to solve, not just attractive clues. Never hide the key in an HTML disclosure or source comment.

Your starting prompt

Produce one self-contained printable HTML player sheet and a separate Markdown answer key from these checked rules, clues and solution. Include all category values, exact bijection rules, numbered clues, three blank 4-by-4 pairwise grids and a blank final-answer table. Use readable type, dark ruled cells, useful print CSS and a narrow-screen layout. Put no answer in player-page text, comments, attributes, scripts, hidden content or disclosures. Keep the answer in the separate spoiler-labeled file, with a deduction path and reproduced rules/clues. HTML must have no scripts, forms, remote assets, dependencies, tracking or network endpoints; add noindex/nofollow and a restrictive CSP. Do not claim a browser or print test from source review.

CHECKED MATERIAL:
[paste]

Expect: A complete player-only HTML file and a separate spoiler-labeled Markdown key with a deduction path. Save generated files into a new folder, preserving the downloaded edition.
Check: The player file needs all rules/categories/clues, three blank 4-by-4 pairwise grids and eight blank final-answer cells. No answer belongs in visible text, comments, attributes, scripts, hidden sections or disclosures. Check there are no scripts, forms, remote resources or tracking, and that noindex/nofollow, restrictive CSP and print CSS are present.
If it differs: If only file text is returned, save the HTML alone with .html and the key alone with .md, without code fences; then reopen them in the editor. Noindex/CSP text does not replace source inspection. Keep any unchecked generated edition labeled unverified rather than copying old validation claims into it.

Recheck the delivered edition

Compare the exact visible wording to the checker and key after layout edits. Open the page yourself and inspect print preview before distributing it.

Your starting prompt

Audit these final files as one versioned puzzle edition. Compare each visible rule and clue verbatim across player HTML, answer Markdown, source brief and the intended checker predicates. Re-run complete enumeration if you can and ensure the final key matches the sole solution. Check the player source for leaked answer data, scripts, forms and external resources. Separately list actual visual observations and unperformed browser, print, assistive-technology and consumer AI checks. Give a manual checklist for legible clues, writable cells, no clipped rows, sensible page breaks and keeping the answer page separate. Do not call a static illustration a browser screenshot or a local validator run an AI workflow test.

FINAL FILES:
[attach or paste all files]

Expect: An audit of the actual saved edition, plus separately labeled observations from opening its HTML in your browser and checking print preview if you perform those steps.
Check: Reopen all files first. Compare visible rules/clues with source and both code implementations, rerun the matching checker if available and compare the sole solution with the key. The supplied fixed checker may reject a differently structured generated page; it is not a general puzzle or HTML validator. Inspect legible clues, empty writable cells, page edges, breaks and absence of spoilers yourself.
If it differs: If layout or wording changes, preserve the failed version, fix a new version and recheck every file and solution set. A PNG inspection is not HTML rendering or printing. Keep browser, physical-print, accessibility, originality-search and difficulty claims pending unless separately established.

Optional: improve the result

Perturb one clue and measure the difference

Treat clue editing as a controlled experiment. Count first, then decide whether the revision preserves one solution or changes the intended answer.

Advanced prompt

Make a clue-perturbation audit of this puzzle. Baseline: record the full solution set and verified key. Remove each clue individually, then change one relation or time offset at a time; count all solutions for every variant using both independently expressed implementations. Report zero, one or multiple solutions and whether the original key still matches. For the supplied five-clue edition, deletion counts should be 15, 12, 2, 3 and 3. Do not silently add a new clue to repair a failed variant. Propose a clearly labeled revision only after the counts are known, update the visible sheet and separate key together, and rerun text parity plus complete enumeration. Keep browser/physical-print review separate.

BASELINE RULES, CLUES, KEY AND CHECKER:
[paste]

Optional: use your own input

Create and check your own small fictional puzzle

Use three equally sized categories with exact one-to-one rules and a size small enough to enumerate completely. Do not paste private details or copied puzzle content. Review the opening specification, then send:
“Choose a complete target assignment for MY SPECIFICATION, labeled DESIGN ANSWER: SPOILERS. Draft precise relational clues and a plain-language-to-code condition map. Check every clue against the target without claiming it excludes alternatives. Define every ordering/time word from my listed values, with no unstated knowledge. SPECIFICATION: [paste].”
“Create a standard-library Python checker for MY actual categories/rules/clues. Enumerate all assignments, with the expected search size derived from my category sizes; for three one-to-one categories of size n it is n! × n!. Independently express the clues in a second representation without sharing predicate logic, compare full solution sets and compare any sole result with MY key. Require exactly one solution. Include changed-key, contradiction, clue-removal and visible-text-change tests whose expected outcomes are derived for my edition, not copied from another puzzle. Provide complete source and exact instructions. Do not execute source data as code or access a network. State what actually ran; if no runtime exists, leave uniqueness unverified. SPECIFICATION, CLUES, TARGET AND AVAILABLE RUNTIME: [paste].”
Inspect the generated code before any permitted execution. Save it as my-puzzle-checker.py in a new folder, reopen it and have it checked if you cannot assess it. Save actual output separately. If execution is unavailable or results disagree, stop the uniqueness claim.
“From MY CHECKED MATERIAL make a self-contained player-only HTML and a separate spoiler-labeled Markdown key with a deduction path. Size the three blank pairwise grids and final-answer table from my categories. Preserve exact visible rules/clues. No answer anywhere in player source, no scripts/forms/remote assets/tracking; include noindex/nofollow, restrictive CSP and print CSS. Avoid unperformed validation claims. CHECKED MATERIAL AND ACTUAL CHECK REPORT: [paste].”
Save each complete file without fences, reopen the actual saved files, then send:
“Audit MY SAVED EDITION against MY ORIGINAL SPECIFICATION and two checked predicate implementations. Compare exact rules/clues/key, enumerate all assignments again if available, and ensure the key matches the sole solution. Check player source for leaked answers and active/external content. Give a manual browser/print-preview checklist and keep actual observations separate from unperformed tests. Do not certify general correctness from a fixed-source checker. SAVED FILES, SOURCE AND ACTUAL REPORTS: [paste permitted text/files].”
Open your saved player HTML in your own browser and inspect the actual print preview if available. Correct a new revision, reopen and rerun after every clue/key/layout edit. Your categories, search size, unique answer, grid sizes and negative-case outcomes must change together. The original advanced prompt and bundled checker remain sample-scoped; do not reuse their deletion counts or hard-coded controls for your new puzzle. No publication or distribution is included.

Your own-input prompt

Help specify a small original fictional logic puzzle with three equally sized categories and exactly one use of each value. First list MY categories and values, one-to-one rules, intended audience, every ordering/time definition and unresolved ambiguity. Derive the complete search-space size from these rules and warn if it is too large for the available checking route. Stop before choosing a hidden design answer or writing a player sheet. A uniqueness claim requires actual complete enumeration and independent condition implementations, not a plausible answer. No existing puzzle, franchise or private personal data is needed.

MY PUZZLE BRIEF
[Give three equally sized small categories, their exact fictional values, purpose/audience, exact assignment constraints and any time/order interpretation. State whether you have a permitted Python 3 runtime and a local browser/editor; if not, say unavailable. Do not claim a unique solution or a measured difficulty.]

Check your own result

Every clue means exactly the reviewed rule, without outside assumptions; the complete search size comes from your actual categories.
Both separately expressed implementations examine all candidates and agree on the complete set; uniqueness remains unverified without execution.
Your key matches the sole solution, and visible rules/clues, predicates, negative controls and grid sizes change together for each edition.
The player source contains no spoilers or active/external content; actual browser/print checks remain separate from enumeration and fixed-text checks.

Show, then try

See the shape of the result.

This input is synthetic. The output is a written illustration of the intended result, not a verified AI-tool run.

The starting point

Synthetic input

CREATE A COMPLETE FICTIONAL PUZZLE
Purpose: a small print-and-play deduction activity, with no prior puzzle knowledge required beyond the included rules.
Theme: four fictional collectors bring small objects to a display cabinet.
People: Aya, Bram, Cleo, Dax.
Objects: compass, lantern, ribbon, shell.
Arrival times on one fictional morning: 09:00, 09:15, 09:30, 09:45.
Rules: one object and one arrival time per person; each object and time used exactly once; no simultaneous arrivals. Earlier and after refer only to these listed times on that morning. No unstated facts.
Create concise relational clues. Use precise minute differences, not an ambiguous word such as shortly. Choose a complete solution privately during drafting and prove that the final visible clues allow exactly one solution by exhaustive enumeration. Supply a human-readable deduction path that uses only the rules and clues.
Player deliverable: a self-contained HTML puzzle sheet with all rules, categories, clues, three blank pairwise logic grids, a blank final-answer table and print CSS. Keep answers out of HTML source, comments, attributes, scripts and disclosures. No scripts, forms, external assets, tracking or network dependencies; add noindex and restrictive CSP.
Separate deliverables: a spoiler-labeled Markdown answer key; a source brief; a Python standard-library validator with two independently expressed checks; a report that accurately names its scope.
Originality boundary: newly author this fixture using fictional non-sensitive details; do not use a franchise or copy a published puzzle. Do not claim an exhaustive originality search or tested solving difficulty.
Verification boundary: local computation and editorial preview inspection are artifact checks, not an executed consumer AI workflow. Browser preview, physical printing and time savings must remain unclaimed unless actually measured.

Copies text only. For image or file tasks, use the source files named in the steps too.

Download source .txt

The intended result

Illustrative output

PLAYER SHEET: THE CABINET OF SMALL WONDERS
Four fictional collectors brought one small wonder each to a display cabinet. Work out who carried which object and when they arrived.

PEOPLE
Aya, Bram, Cleo, Dax

OBJECTS
Compass, lantern, ribbon, shell

ARRIVAL TIMES
09:00, 09:15, 09:30, 09:45

RULES
1. Each person carried exactly one object and arrived at exactly one of the four listed times.
2. Each object and each time was used exactly once. No people arrived together.
3. Earlier and after refer to the same morning. Use only the facts stated here.

FIVE CLUES
1. The lantern arrived exactly 30 minutes before the compass.
2. Cleo arrived exactly 15 minutes after the person carrying the compass.
3. Bram arrived earlier than Dax.
4. Aya carried neither the lantern nor the shell.
5. The shell arrived earlier than the ribbon.

SOLVING SPACE
The HTML player sheet supplies blank People × objects, People × times and Objects × times grids, plus a final-answer table. Mark a match and rule out incompatible pairs until each logic-grid row and column has one match.

SEPARATE ANSWER
The complete answer and five-step deduction path are in small-wonders-answer-key.md. No answer is embedded in the puzzle HTML or its layout preview. Open the answer file only when you want spoilers.

LOCAL CHECK RESULT
Two independent implementations each enumerate all 576 one-to-one assignments and agree on the complete solution set: exactly one solution, matching the key. Removing clues 1–5 respectively leaves 15, 12, 2, 3 and 3 solutions. Adding a contradictory condition leaves zero. Ten invalid file mutations are rejected, including a wrong but still one-to-one key and changes to visible clue wording.

USING THE FILES
Open the HTML locally and use the browser print command. A4 portrait at 100% is the intended setting; inspect the actual preview and adjust for Letter paper if needed. The supplied PNG is an inspected code-native layout preview, not a browser screenshot or proof of printed output. The Python checker uses only the standard library and reads the files beside it.

UNFINISHED CHECKS
Consumer AI execution, browser rendering, browser print preview, physical printing, assistive-technology behavior, solving difficulty and time savings have not been tested or measured.

Download result .txt

You’re the final check

Check the worked example.

These checks use the worked example and its particular facts. With your own material, check your actual names, numbers, sources and rules; do not reuse the sample’s values as the answer.

The supplied edition has four distinct people, objects and times, explicit one-to-one rules and exactly five clues.Both independently expressed enumerators examine 576 assignments, agree on the full solution set and find exactly one solution. The Markdown key matches it.Removing any one clue makes the supplied puzzle ambiguous; contradictory and mismatched-key variants are rejected. Exact text parity guards against layout-time clue changes.The HTML player sheet and code-native preview contain no embedded solution; the explanatory key is a separate spoiler-labeled file.The HTML has three blank 4-by-4 pairwise grids, a final-answer table, no scripts/forms/resources, noindex and print CSS. Browser appearance is still a separate check.The raster preview has been directly inspected for readable text, intact grids and no visible clipping. It is not a browser screenshot; browser and physical-print tests remain unperformed.

Unsaved checks stay in this tab and reset when you reload. If you choose Save guide, its current example checkmarks are kept in this browser when storage is available.

A few things to watch for.

Finding the intended answer and stopping without checking whether other answers also fit.
Writing a clue such as immediately after without specifying whether it means the next listed slot or an exact time difference.
Sharing one buggy predicate implementation between two counters and treating their agreement as independent confirmation.
Editing a visible clue after running the solver, or leaving an old key beside a new puzzle.
Keeping the answer in a collapsed disclosure, HTML comment or hidden script on the player sheet.
Claiming that attractive layout, a solver pass or an AI-generated explanation proves print quality, player difficulty or originality against every published puzzle.

What has been checked?

Documentation: No dated source review recorded.
Included files: Scoped checks passed on 2026-10-14. File parsing; Fixture checks. See scope below.
AI workflow: Not run in a recorded consumer-AI workflow.
Time savings: Time savings not measured

Read the scope and evidence

small-wonders-puzzle.html

2026-10-14 · parser · passed

Static exact-edition HTML inspection and clue/rule parity; no browser behavior was tested. Three empty pairwise grids and a final-answer table are present.

No active/external tags or resource URL attributes
Restrictive CSP and noindex/nofollow
Exact five clue texts and three rule texts
Blank solving cells: 16, 16, 16 and 8
Explicit print CSS; separate answer filename only