A plot or widget in every notebook (three figures are already written but commented out)

[project-delphi]

The problem

Twelve notebooks about the shape of data, and almost nothing is ever drawn. matplotlib appears in exactly three places and all three are commented out:

• scripts/content.py:639 — §04, "Optional, and worth doing once", showing the reshape-vs-transpose scramble
• scripts/content.py:1317 — §09, original / sobel / blurred side by side
• scripts/content.py:1656 — §11, the PCA scatter

The §04 one is the clearest loss. The entire section argues that reshape and transpose give identical shapes and completely different data — and the one thing that makes that undeniable in half a second, seeing the scrambled image, is behind a comment marker that most students will not uncomment.

The ask

One visual per notebook. A plot where a plot is the point; an ipywidgets slider where interaction is the point. Not decoration — each one should replace a paragraph of explanation.

• 00 setup-and-data — thumbnails of the datasets being loaded, confirming visually that the download worked (see Add CLAUDE.md, fix the checker's numbering, pin line endings to LF #3)
• 01 what-a-tensor-is — a slice and a fiber highlighted on the same image
• 02 thinking-in-n-dimensions — this one is a discussion block with no code by design; a diagram may fit better than a plot, or skip it
• 03 indexing-and-broadcasting — slider over the tumour feature index; the zero-variance pixels made visible
• 04 reshape-and-transpose — uncomment the existing figure (content.py:639) and make it non-optional
• 05 video-pipeline-design — shape diagram per pipeline stage
• 06 contraction-with-einsum — which axis gets summed, drawn
• 07 inverses-and-pseudoinverse — residuals of the least-squares fit
• 08 recursion-with-matrices — the airline forecast against the actual series; eigenvector directions under repeated application
• 09 convolution-and-deconvolution — uncomment the existing figure (content.py:1317), plus a slider over kernel size
• 10 tucker-decomposition — slider over rank k (pairs with PCA is not a factorization — remove it from the map of factorizations #4)
• 11 wrap-up-and-take-homes — uncomment the existing PCA scatter (content.py:1656); scree plot showing the standardisation trap

Take these as a starting list, not a specification. If a section is better served by no plot, say so and tick it off.

Constraints — this one touches every file, so read carefully

• All twelve edits are in scripts/content.py. The .ipynb files are generated by scripts/gen_notebooks.py; hand-editing them is caught by CI (which reruns both generators and fails if the tree changes) but only after you push.
• No committed outputs. scripts/check_links.py fails on any cell with outputs or an execution count. A widget that renders only when executed is fine; a widget whose state gets serialised into the .ipynb on save is not. Check git diff --stat before committing — a stray figure or widget blob shows up as a suspiciously large diff on a file you only added ten lines to.
• ipywidgets in Colab: works, but some widget types need google.colab.output.enable_custom_widget_manager() first. Put that in the cell with a comment rather than assuming the student's runtime is configured. Students are on Colab, not local Jupyter — test there.
• Nothing here changes the built site. _quarto.yml serves notebooks/*.ipynb as raw resources:, never rendered — Quarto does not execute them. So CI checks validity and nothing more. The only real verification is opening a couple in Colab and clicking the sliders.
• Bilingual prose on every new markdown cell — 🇪🇸 blockquote alongside the English, matching the existing pattern.

uv run --with pyyaml,nbformat python scripts/gen_notebooks.py
uv run --with pyyaml,nbformat python scripts/check_links.py --notebooks-only

Please split this up

Twelve sections in one PR is unreviewable and will sit open. Several PRs, grouped by two or three sections each. The three already-written-but-commented-out figures (§04, §09, §11) are a good first PR on their own — small, obviously correct, and it unblocks the pattern for the rest.

[Laverde97]

We want widgets with output to engage readers