At least two risks per layer — six risks total, minimum. For each risk:

• Which layer — purpose, process, or structural.
• The exact lesson activity from Phases 1–4 where this risk applied. Reference it by phase and task.
• The affected student population. Be specific — not “students” but “the three EAL students in my Year 9 class” or “the students in my top set who have learned to optimise for the rubric over the learning.”
• The mechanism by which harm could occur. Walk through it step by step.

Structural findings are the hardest. Start by asking: what did I pay with? Student data, attention, training data. Each of those flows is a structural question if you trace it honestly.

For each risk from Task 1 (or a focused subset), produce a memo entry with:

• Risk name — a short handle you can reference later.
• Mechanism — the step-by-step causal pathway from AI use to student harm.
• Affected students — who specifically.
• Mitigation action — what you will actually do, with a timeline. Mitigations without timelines tend not to happen.
• Disposition — did you revise the unit plan, or accept the risk with documented rationale? Both are valid; accepting must come with reasoning, not a shrug.

Level 3 memos trace at least one mechanism all the way to a specific student outcome with an implementable mitigation. Level 4 memos also reference subject-specific literature or institutional context.

Go back through every AI-use decision you made in Phases 1–4. Identify the ones made on efficiency grounds — “this was faster,” “this was easier,” “this let me skip the hard part.” For each efficiency-driven decision, do one of two things:

1. Revise it — decide the decision was wrong in retrospect, and describe what you would do instead, with the cost in teacher time or effort.
2. Provide a principled argument — defend the efficiency decision on its merits. Principled arguments for efficiency are legitimate and sometimes correct — but they must be arguments, not defaults.

At least one efficiency-driven decision must be identified. An audit that finds zero is a Level 2 audit — it means you either didn’t look, or your work had no efficiency pressure at all, which is almost never true for real teachers.

Your unit plan from Phases 1–2, updated based on the audit. The revision log is not a diff; it is a reasoning document. For each change:

• Cite the audit finding that drove the change (risk name from Task 1 or purpose drift item from Task 3).
• State what changed — the specific text, activity, tool, or sequence that is now different.
• Explain the expected effect on student development — not just “less risky” but what the student will now experience that they wouldn’t have.

At least one revision must change tool or platform, not just use behaviour. This is the hardest Day 5 requirement. If the risk is that a platform’s training data is extractive, the mitigation cannot be “use it more carefully” — it has to be “use a different one, or none.”