You are grading a slide deck made by an AI model from a course module. Grade it strictly and fairly; you are not told which model made it.

Files in this folder:
- source.md: the module the deck had to be built from (the ground truth).
- request.txt: the exact request the model was given.
- slides.md: the text on every slide and its speaker notes, in order (layout, diagram, image and placeholder counts in brackets).
- check.txt: output of the deck's automatic checker (layout, overflow, text size, contrast, missing notes).
- shots/: a screenshot of every slide in order (slide-01.jpg, slide-02.jpg, ...), with all builds revealed.

Do this:
1. Read source.md, request.txt, slides.md and check.txt in full.
2. Look at EVERY screenshot in shots/. Note any slide that is cut off, overlapping, unreadable, empty, broken, cramped, or whose diagram is wrong, mislabelled or does not match the text.
3. Check every number, formula, unit, date, name and technical claim in slides.md (slide text, diagram labels and notes)
   against source.md and against correct knowledge of the subject. Report each one that is wrong, and separately each claim
   that is not in source.md (invented), even if it happens to be true.
4. Score each criterion from 1 (poor) to 5 (excellent):
   - accuracy: no wrong or invented facts; formulas and worked examples correct.
   - coverage: follows the module's sections in order and keeps the key ideas, formulas and worked examples.
   - teaching: clear for the audience named in request.txt; one idea per slide; good pacing for the time it gives.
   - visual: from the screenshots: readable, well laid out, consistent, nothing cut off or overlapping.
   - diagrams: animated diagrams used where they explain better than text, and correct; image placeholders captioned with what goes there.
   - notes: speaker notes on every slide that tell the presenter what to say, not a copy of the slide.
   - requirements: everything request.txt asks for (slide count, order, quiz and summary, file name, checker passing).
5. Give an overall score from 1 to 10 for how ready this deck is to teach from tomorrow.

Return only the structured result. Quote slide text exactly in "claim". Use severity "major" for anything that would teach
students something wrong or embarrass the lecturer, "minor" otherwise.