Short, practical things that work on Monday morning.
Polished, confident prose is exactly what a language model produces when it is wrong. Give students marking codes and make them annotate the answer.
Students can ace the retrieval starter and still not see what it was for. The bridge between lessons has to be said out loud.
Boredom and anxiety are the same design fault: one task, one difficulty, thirty different skill levels. Let them choose where to step in.
Trips, simulations and practicals do not teach by themselves. The learning happens in the structured debrief, which is the part that gets cut.
If your top level is the only respectable place to land, most of your class fails by design. Anchor the middle as the genuine target.
Productive failure needs a precisely calibrated task: solvable in part with what they know, unsolvable without what you are about to teach. Let AI calibrate it.
When accounts contradict, students pick a winner or give up. Historians treat the contradiction itself as evidence. Teach that move with a two-column matrix.
Students who explain a concept in their own words first learn more from whatever the AI says next. Order is the whole trick.
Breaks are not the problem; the ragged five minutes after them is. Decide the way back in before you let them out.
'Materials manager' is a title, not a role. Interdependence appears when the group genuinely fails if any one member's job goes undone.
Collect the work, pick one answer, put it under the visualiser, and let the class do the marking. One artefact, thirty people thinking.
You write for an hour; they read for a minute, if at all. Three minutes of live marking does what the margins never did.
