Lesson quality rubric (draft for review)

This is NOT a template. These are principles the critic will use to review each lesson and request improvements. The idea is to give judgment, not rigid rails. Read it and tell me what to remove, change, or add.

What makes a good lesson (the north star)

A lesson teaches one competency in the way that best fits THAT content and THAT person. There is no single form: content of different natures (a procedure, an abstract concept, a physical skill, a data analysis) deserves different experiences. The system should not know specific domains; it reasons about the nature of the content, not the topic.

Principles (what the critic checks)

1. Interaction has to ADD VALUE, not be a toll

  • Interacting should make you think, decide, or attempt before seeing the answer.
  • ❌ Bad: hiding information the learner should read directly, behind a click (a list of informational steps stuffed into a reveal).
  • ✅ Good: a worked example you reveal step by step in order to reason through the next one.
  • Rule: if hiding something doesn’t force thinking, don’t hide it.

2. The right component for the content

  • Simple sequential info → show it in full (visible list of steps), not a reveal.
  • Example to reason through → progressive worked example.
  • Compare two states → before/after. Order things → drag. Data → chart/table.
  • Concept with a key term → glossary/flashcard.
  • Rule: the component is chosen for what the learner has to do, not to fill space.

3. Several screens when the content calls for it (no fixed number)

  • A simple concept can be 1 screen; one with practice, 2-3; a dense one, more.
  • Each screen is understood on its own, centered, with no scroll.
  • The content decides, never “always N screens”.

4. Always interactive, but not everything is a test

  • Every lesson has at least one real interaction.
  • Interacting ≠ evaluating. Most of it can be exploration/play; evaluate only when there is something that can genuinely be checked.
  • The last screen does not block progress (it is noted/logged, but you can continue).

5. Faithful to the source, honest about mastery

  • Everything comes from the document/pack. No inventing facts, interfaces, or scenarios the source does not have.
  • If it cannot really be evaluated (there is no oracle), the lesson is still interactive but does not certify mastery (it is support). It never fakes a test.

6. Adapted to the person

  • Adjusts support/pace based on the learner’s level and errors.
  • Designed to grow: tomorrow, adaptation for accessibility/capabilities without rebuilding anything.

7. No noise

  • No filler text, no content from another domain, no artifacts (“[Source: …]”).
  • Regular text reads as regular text (not everything in giant “lead” style).

What the critic does with this

After generating a lesson, the critic (a different perspective) checks it against these principles and leaves concrete, actionable notes (“this is info, don’t hide it behind a reveal”; “drag fits here, not a quiz”; “this is 2 screens, not 1”). The generator adjusts once. It doesn’t impose numbers or templates: judgment nudges.

Questions for you

  1. Is there a principle missing that you consider non-negotiable?
  2. Is any of them too rigid / would you remove it?
  3. Point 4 (“not everything is a test”) you left as “keep evaluation for now” — do I leave it as is for now, or do you already want to push toward less-testing?