Model-assisted drafts for lesson plans: guardrails we use

Generative tools can accelerate outlines and quiz variations, yet they hallucinate API names and policy citations. Our mentors require human verification on every code snippet before it ships to learners, and we ban pasted model output in reviewer-facing text without source checks.

We log which lessons used assistance for transparency, and we do not grade students on speed of generation—we grade on understanding demonstrated in mentor reviews. When models suggest refactors, students must narrate trade-offs in their own words.

If your organisation bans external LLMs, our workflows still work: the guardrails are about provenance, not the specific vendor. Consult your IT policy before connecting classroom exercises to cloud APIs.