Safe building and quality checklist

Everyone, every level, every build.

• 1. No student names or data in any prompt, file, file name, screenshot or tool. Use made-up examples. Treat anonymised student work as student data too. And the tool you build: it never asks students to type names or personal details.
• 2. Nothing student-facing calls AI. Students use the finished page, never the AI tool. Every build today uses no AI and no keys. Think (AI inside the tool) is beyond this session: see card 18.
• 3. Nothing is saved, unless it's a Remember build. A Remember build may store progress in this browser only, never what students type, with a Clear button. Test it where students will open it: browsers differ for a file opened from the computer, and a private window forgets it.
• 4. No calls to other websites in the self-contained file. Test the final saved file offline, then ask a helper to review its code and browser Network requests across normal and error cases (card 05). Offline success or a search for http alone cannot establish this rule: requests can be relative, conditional or built from strings. An SVG namespace is not a request. Canva needs its hosting service; check the published page has no added AI or data-service features instead. Canva builds need internet; skip the offline test.
• 5. Every fact, answer and formula checked by hand, by you, against your own scheme of work or textbook.
• 6. A public link is public. Anyone who has it can open it and pass it on. A GitHub Pages site can also be found by search engines, and its repository through your GitHub profile (card 17).
• 7. Logins, keys or student data need school IT sign-off. That is Step 5, Connect, on the skills ladder (card 18), and it is not taught here.
• 8. Your school's rules. Check your school's AI policy and your regulator's guidance (such as KHDA, ADEK or MoE in the UAE; DfE in England).

• The law. Information that identifies a real student is personal data, and the law protects it. In the UK: UK GDPR and the Data Protection Act 2018. In the UAE: the Personal Data Protection Law (Federal Decree-Law No. 45 of 2021). Ask whoever leads on data protection at your school how it applies to you.
• Your accounts. Use the accounts your school approves. Turn off any setting that lets the tool train on your chats.
• Age limits. Rule 2 keeps students off the AI tools. If your school ever lets students use one, check its age limit first:

Apply them to classroom learning tools before students see them. For a teacher workflow such as menu item 9, use the safe-building checks, accuracy and accessibility checks, plus your workflow success checks. A preparation checklist does not need retrieval questions or right/wrong feedback. From DEEP's review of the learning evidence for each build type (27 Sep 2026).

• 1. One learning goal, written first. I can say what students will know or do better. Coherence and seductive details (Mayer; Sundararajan & Adesope)
• 2. Students have to think, not just click. At least one step where they predict, explain, choose and justify, or create. ICAP (Chi & Wylie); generative learning (Fiorella & Mayer); Crouch et al.
• 3. Retrieval is built in. It ends with, or is, a low-stakes question answered from memory. Roediger & Karpicke; Adesope et al.; Yang et al.
• 4. Feedback explains. Every answer shows why it is right or wrong and what to do next. No praise-only feedback. Hattie & Timperley; Shute; Van der Kleij et al.; Butler & Roediger
• 5. Guidance matches the learner. Novices get a worked example or a short explanation first. Stronger students get fewer steps. Kirschner et al.; Alfieri et al.; Atkinson et al.
• 6. Nothing decorative. No confetti, fun facts, stock images, points or timers unless they serve the goal. Seductive details; Hanus & Fox
• 7. Visuals are integrated. Labels sit next to parts. Motion is used only for change. One idea per screen. Mayer; Höffler & Leutner; Tversky et al.
• 8. Few controls. A simulation or model has only a few variables (two or three is a rule of thumb) and responds instantly and visibly. PhET research (Podolefsky et al.; Adams et al.)
• 9. I checked the facts. Answer key, formulas, units, dates, spellings and references, every one. Walters & Wilder
• 10. Accessible. Contrast, more than colour, keyboard use, an adjustable timer, alt text, phone width. WCAG 2.2
• 11. Reused. I know when students will meet it again, spaced out and mixed with other topics. Cepeda et al.; Dunlosky et al.
• 12. I'll know if it worked. At least one check of learning, beyond "they enjoyed it". Deslauriers et al.; Roediger & Karpicke

Three quick checks from item 10, on your Respond build.

• With the mouse put away, I could reach and choose every answer using Tab, the arrow keys, Space and Enter, and could always see where I was.
• Right and wrong are shown in words, not colour alone.
• With the browser window narrowed to its smallest, or on my phone, I could still read everything, with no sideways scrolling.

Also a Level 3 stretch (card 14). Log each line in your run log (card 05).

• With the mouse put away, I reached and used every control in a sensible order, and could always see where the focus was.
• I checked every text and background colour pair in a contrast checker. Each is at least 4.5:1.
• Right and wrong, and any other status, are shown in words as well as colour.
• If I published it: I repeated the keyboard check on the live link on a laptop, and opened the live link on a phone.