Thanks for shipping this, Max. Ran a stress test today: fed it a 9,800-character brief for a 10th-grade ELA lesson on Bradbury's "A Sound of Thunder" through a classical rhetoric framework (stasis theory / FDQP), with Minecraft as a staging ground for butterfly-effect analysis. The regeneration workflow produced a usable draft. A few observations:
First, one thing that shouldn't ship: the "Enhance with AI" panel offers a button labeled "Differentiate activities for various learning styles." Learning styles theory — the claim that matching instruction to visual/auditory/kinesthetic "learner types" improves outcomes — has been empirically debunked for nearly two decades (Pashler et al. 2008, Kirschner 2017, Willingham throughout). It's routinely classed as a "neuromyth." Putting it in the UI of a lesson-planning tool marketed to teachers embeds a discredited framework as a one-click affordance and signals endorsement, whether or not that's intended. Worth removing before more teachers build lessons around it.
What the tool did well: it preserved custom framework terminology (FDQP, stasis theory, the deliberation spectrum I use with Shakespeare), named anchor texts, retrieval-practice scaffolding, and differentiation moves from a dense brief. Carrying that much specificity through is harder than it looks. On a second regeneration pass with corrected picker inputs, standards alignment was accurate, and the 85-minute block was honored with proper phase distribution.
Where it still needs work:
1) Expanded length didn't actually expand on the first pass. The initial draft came back at 60 minutes with denser bullets per section, despite an explicit 85-minute request in the brief. Length only changed on regeneration.
2) Mismatched inputs go unflagged. I initially selected RL.9-10.6 (world literature) in the picker despite naming Bradbury as the primary anchor in the brief. The tool used my selection; the error was mine. But a consistency check — "your selected standard fits the content you've described poorly, did you mean RI.9-10.6?" — would move this from generator toward assistant.
3) Platform reassertion persists. Even with explicit instructions against tech-demo framing, "Book and Quill," "Camera and Portfolio," spawn-egg inventories, and a Minecraft 101 course suggestion kept finding their way into both drafts.
The framing — "You bring the teaching expertise" — is good. The tool earns that claim once the learning-styles button is retired, and when teachers audit standards, re-budget time, and strip platform promotion.