A lesson you teach in your normal classroom, in your current unit, with AI present in a pedagogically meaningful way. The lesson must implement at least two different HAA extension types — you cannot pass Day 3 by running an entirely A-type or entirely C-type lesson. We want to see you transitioning between types in a single lesson.

The video must show the screen whenever AI tools are used, so the evaluator can see what the AI was producing and what you did with it. At least one moment must be an explicit mediation moment — a point where you visibly stand between the AI output and the students, reshaping, challenging, re-voicing, or refusing something. Timestamp that moment in the decision log.

An 8+ minute chunk of the lesson where you teach students how to think about AI as a cognitive extension — in your subject, for this kind of work. Three things must be present:

1. Which extension type is in use in the current activity, and what that means for what AI is and isn’t doing for them.
2. At least one subject-specific boundary condition. Not “always check the facts” — a specific type of error in your subject.
3. At least one concrete override case. An example where the teacher or a student overrode the AI — show students what saying no to the tool looks like.

The difference between Level 2 and Level 3 is whether you teach operation rules (“click here, don’t paste that”) or metacognition (“here is what AI is doing to your thinking, and here is how to notice it”). Rules are brittle and generic. Metacognition is durable and subject-specific.

At least eight entries, each timestamped (minutes:seconds into the lesson). For each AI-related decision, record:

• The action — what did you actually do?
• Extension type — A, B, or C, and whether this matched your plan or deviated.
• Pedagogical rationale — in the moment, what were you trying to do for students’ learning?
• Retrospective assessment — after the lesson, did that call work? Be honest. “It went badly” is valid and valuable.

At least one entry must record a deliberate exclusion — a moment where you decided not to use AI, with a subject-specific reason. This is the most important single entry in the log.

Entries that are conspicuously smooth — “I asked AI for X, it gave me X, I used X” — are the ones we trust least. We want entries where you weren’t sure, where the AI gave you something unexpected, where you had to decide whether to use what it produced.

After the lesson, take three pieces of student work that involved AI. Apply the Output Evaluation framework from Day 2, but now to student-produced work. For each sample:

• Identify at least one AI error or bias the student absorbed. Look for places where AI shaped the student’s thinking without either of you noticing at the time.
• Name the likely misconception that would form if left uncorrected. Trace the learning consequence.
• Write feedback addressing the student’s AI use, not just their work product. Not “this is correct/incorrect” but “this argument is two arguments stitched together, and the seam is where the AI took over. Here’s how I can tell.”

Teaching students to read their own AI-assisted work with this kind of attention is one of the things the rubric values most heavily.