Brief it: the spec

Your build menu item (card 13) has a ready-made version of this spec, twice: a Show prompt and a Respond prompt. Fill in their blanks with your class, your topic and your content, checked by you. The blank shape below is for a problem the menu doesn't cover.

• Quick Show build first: keep only the Show part of the "It should" line, delete the "After two wrong tries" line, and use checks you can see at a glance (for example: "When I open it on a phone, I can read every label with no sideways scrolling").
• Then your Respond build: keep only the Respond part and the "After two wrong tries" line, and use your canvas's three checks.

Build a single web page for [who: year group or age] in [subject].
The change I want: after using it, they can [outcome].
It should [Show: show … | Respond: ask … and, after each answer, say whether it is right and explain why].
After two wrong tries on a question, it says "Ask your teacher" or shows a worked example.
The content is: [paste your questions, labels or steps, checked by you].
It must work on a phone and with a keyboard alone, and show right and wrong in words, not colour alone.
It must never: ask for names or save anything; call AI or any website; use timers, points, leaderboards or confetti; [your own red line].
It is working when:
1. [success check 1]
2. [success check 2]
3. [success check 3]
Give me one self-contained HTML file. Before you build, tell me in three lines what you will make.

This spec is your whole prompt. Paste it as it is (card 04).

• Canva builds straight away. Open the preview.
• AI chat tools only: the prompt asks it to tell you in three lines what it will make before it builds. If that matches your spec, reply "Go ahead". If it builds straight away, check what it made against your spec.

Four boxes of your canvas fill five sections: outcome statement (box 2) → section 1; the user (box 3) → 2; red lines (box 5) → 6; success checks (box 4) → 7 and 8. Each success check becomes one of the five acceptance criteria, and its "When I …, it …" becomes a test case. Add two more criteria from card 08, such as keyboard alone and phone width.

Minimum in the session: sections 1, 3, 5, 6 and 7, plus three test cases. The rest is after the session.

Choose the learning move

Not sure? Start with a quiz: of the four, it is the easiest to build and to check.

• Retrieval quiz = retrieval practice. Students answer from memory, with low stakes, and see why each answer is right or wrong. Use it again in later lessons.
• Simulation = guided inquiry. Students predict, change a few variables and explain what happened; you debrief. The highest-risk build: check its maths or science by hand.
• Model = a representation to reason with. Labels on the parts, motion only when the idea is about change, and its limits stated.
• Explainer = a worked example. Step by step, then partly worked, then one to try, with a "why?" at each step.

The template

Copy it into a document or your AI chat and fill in the slots, starting with the minimum above. Include at least one test case where a correct answer must be accepted, not only wrong answers rejected.

TOOL BRIEF: [working name]
1. LEARNING GOAL (one sentence: what students will know or do better): [ ]
2. WHO USES IT: [class and age] / already know: [ ] / go wrong now: [misconception]
3. LEARNING MOVE: [retrieval quiz / simulation / model / explainer]
4. HOW STUDENTS REACH IT, WITH NO STUDENT LOGIN: [.html file on LMS or Teams / published link, as a stretch] on [laptops / tablets / phones]
5. WHAT IT MUST DO: [the one core thing students do, step by step]
- After two wrong tries on a question, it says "Ask your teacher" or shows a worked example.
6. WHAT IT MUST NEVER DO:
- ask for a name or personal information, or save or send what students type
- call an AI service or any other website
- show timers, points, leaderboards, confetti or sound, unless they serve the goal: [how]
- say right or wrong without explaining why
- [your own]
7. ACCEPTANCE CRITERIA (five, each checkable): 1. [ ] 2. [ ] 3. [ ] 4. [ ] 5. [ ]
8. TEST CASES (5-10, written before building): what I do -> what should happen
1. [ ] -> [ ]
2. [ ] -> [ ]
3. [ ] -> [ ]
4. [ ] -> [ ]
5. [ ] -> [ ]
9. NON-GOALS FOR VERSION 1: [ ]
10. OPEN QUESTIONS: [what I still need to decide or check]

Paste it under the build prompt on card 04.

1. Write your own spec, in your own structure. Use the ten headings of the template (version B) as a checklist, not a form: goal, user, learning move and rung, how users reach it, must do, must never do, acceptance criteria, test cases, non-goals, open questions. Your outcome statement and three success checks go in word for word.
2. Draw two constraint cards from the pile (card 14). Write each into your spec as a requirement, with the check that proves it.

Then, at the start of Build Sprint 1, sketch the architecture on paper and have the agent plan before it codes (card 04).

Stuck on a blank template? Let the AI ask the questions. Adapted from DEN Lab Build 01 and 03.

Interview me to turn a classroom need into a brief for a small tool students will use in a web browser. Do not suggest designs or write code until the brief is agreed.

NEED: [the topic or skill students keep getting stuck on]
CLASS: [year group or age, subject, what they already know]
LEARNING MOVE: [retrieval quiz / simulation / model / explainer / not sure]
HOW STUDENTS WILL OPEN IT: [.html file on our LMS or Teams / published link / not sure]

Ask me one question at a time, covering:
1. What should students know or be able to do better afterwards?
2. Where exactly do they go wrong now?
3. What will students do in the tool that makes them think, not just click?
4. What should the feedback say when a student is wrong?
5. What is the smallest version that would be useful in one lesson?
6. What failure would make it unsafe or not worth using?

If I mention a student's name or any personal detail, stop and ask me to remove it. Use only made-up examples.

Then write a one-page brief: learning goal (one sentence); who uses it; learning move; how students reach it with no login; what it must do (always: after two wrong tries on a question, it says "Ask your teacher" or shows a worked example); what it must never do (always: no names or personal information, nothing saved or sent, no calls to AI or other websites, no timers, points, leaderboards or decoration that do not serve the goal); five checkable acceptance criteria; five to ten test cases, including one where a correct answer must be accepted; non-goals for version 1; open questions.

Do not claim the tool will improve learning. End with the one decision I still need to make before building.

Alex's test for a brief: "Ask whether two different teams, given only this document, would build essentially the same thing." (Why a Normal Spec Isn't Enough When AI Builds the Thing). Would two teachers, given only your spec, build the same tool? If not, settle the fork now, or the AI will pick for you.

TOOL BRIEF: Particles check
1. LEARNING GOAL: Students can use the particle model to explain what happens to particles when a solid melts and when a substance is heated.
2. WHO USES IT: Year 7, mixed-attainment science. Taught the particle model last lesson. Go wrong now: think particles in a solid do not move, or that particles get bigger when heated.
3. LEARNING MOVE: retrieval quiz, end of lesson, and again in two weeks.
4. HOW STUDENTS REACH IT: .html file on the LMS. Laptops and some phones.
5. MUST DO: eight multiple-choice questions, one per screen, three options each. A right answer: "Correct" in words, plus one sentence on why. A wrong answer: "Not quite" in words, one sentence on why, and another try. After two wrong tries on a question, it says "Ask your teacher" or shows a worked example. Ends with one question answered in their own words, then a model answer to compare.
6. MUST NEVER: ask for a name or save answers; call AI or any website; timers, points, leaderboards or confetti; "all of the above"; the longest option always being right.
7. ACCEPTANCE CRITERIA: (1) every answer matches my scheme of work, checked by me by hand; (2) every answer shows why; (3) works with the keyboard alone, focus visible; (4) usable at phone width, no sideways scrolling; (5) right and wrong shown in words, not colour alone.
8. TEST CASES:
1. Choose the right answer to Q1 -> "Correct", plus why.
2. Heating question, choose "the particles get bigger" -> feedback names the misconception: particles move faster and further apart; they do not get bigger.
3. Press Next without choosing -> asked to choose; not skipped.
4. Tab through with no mouse -> every option reachable, focus visible.
5. Open on a phone, reload halfway -> no sideways scrolling; restarts cleanly; nothing saved.
6. Choose two wrong answers on one question -> it says "Ask your teacher" or shows a worked example.
9. NON-GOALS: saving scores; a teacher dashboard; random order; sound; images other than one labelled particle diagram.
10. OPEN QUESTIONS: does the heating question need the particle diagram beside it?