Appendix A · What this demonstration does

A six-slide concept pitch with a scripted, 78-second learning scenario on slide 3. It illustrates the proposed experience; it does not call a live AI model or execute generated p5.js.

0–12 s · Intention: “As the mouse moves right, make my square grow.”
12–28 s · Candidates and prediction: present three code variants with outputs hidden.
28–50 s · Compare: the same mouse input makes A grow, B shrink and C move.
50–66 s · Modify: change A’s maximum size from 100 to 160.
66–78 s · Explain: ask for an explanation and a new intention without a supplied answer.

Implementation: HTML, CSS, JavaScript and SVG. Code and shapes share the same numeric mappings. A: size 20→100; B: size 100→20; C: x-position 30→270, size 60. Input range 0→300; drawing uses centre alignment. Pause, replay, keyboard navigation and a manual reduced-motion sequence are available.

Appendix B · Evaluation, boundaries & assistance

Research question: Does comparing implementations selected for a learner’s own intention improve prediction, explanation and independent revision compared with fixed examples?

Proposed evaluation: recruit first-year Design learners with consent, compare matched tasks with AI-selected and fixed examples, then ask for a prediction, an explanation and a small new change without AI. Report reasoning and task outcomes rather than relying on a polished final animation. No learner study or learning benefit is claimed for this demonstration.

Four-week plan: build checked examples and comparison first; add bounded model suggestions; run learner trials; refine and evaluate. The future system would use an instruction-following language model, a curated example library and a browser coding environment with isolated execution and time limits. This is proposed architecture, not an implemented AI backend.

Risks and responses: check suggestions by running code; require prediction and independent revision to address dependence; minimise stored learner data; support keyboard access and reduced motion. Recruitment, teaching unit, model choice and interface remain to be validated.

AI assistance acknowledgement: OpenAI Codex assisted with presentation code, visual composition, reference formatting and English rehearsal wording under the student’s direction and iterative feedback. The AI dialogue shown on slide 3 is authored as a scripted illustration. Research findings are attributed to the cited authors; no fabricated learner results are presented.