[grunion-ai]

What changes

A deck can now choose how much page it has, separately from how much a slide carries.

canvas: 'low' | 'medium' | 'high' in the model, or --canvas on create and new. Medium is the default.

canvas	size	side margin	Body reads as, on the 960 scale	point caps, speaker / reading
low	720×405	56	24	3 / 4
medium	960×540	60	16	3 / 8
high	1280×720	32	10.5	6 / 24

The tier sets the canvas, margin, role sizes, foot band and word caps. Library layouts and templates are cut from the margins inward with a piecewise map, so a full-bleed row stays full bleed and a style kit written for 960 works on all three. --space still wins on size.

density keeps its meaning: speaker and reading say how much a slide carries.

Script-blocked viewers

Claude's browser pane shows a local file outside the project as a static snapshot. Scripts do not run there, yet the parser still keeps the <noscript> still as text, and display:contents painted that text as raw markup.

The deck now hides that text and says the viewer blocks scripts. With JavaScript fully off, the still of slide 1 renders as before. A script-blocked viewer gets the message, not a slide. Hosting with bin/serve.mjs is the route to a working deck in the pane.

Evidence

• npm test: 737 of 737 pass after rebasing on Store repair after a late save from another window #218. The Firefox and WebKit checkout test (test/checkout-live.test.mjs) each failed once in earlier full runs and passed when rerun alone, so it looks flaky under load.
• test/canvas-density.test.mjs is new. It holds medium to the same geometry as a deck with no canvas, and covers the tiers, the map, the caps and the errors.
• Three mock decks with the same content built on the tiers (7, 5 and 3 pages) pass validate --strict and verify --strict in Chromium and WebKit.
• The blocked-script fix was checked in Chromium with a script-src 'none' policy, not in the pane itself, because the pane's tools cannot inspect a snapshot tab.
• deck.html, library.html and schemas/model.schema.json are regenerated. The template change is why the two HTML files differ.

Known limits

• Low holds less on purpose: 61 of 115 templates fit their sample copy there, against 114 of 115 on medium and high. The one miss on medium and high is the same figure-decision connector warning. On low, validate names the overlap to shorten or split.
• Role factors and caps are fixed tables per tier. A style kit cannot override them yet.
• Low sets Stat at 1.4× so library tiles hold it. A hero slide could take more.
• SKILL.md lost its one-line "A PDF nobody asked for" anti-pattern to stay inside the word budget. Step 6 still says it.

🤖 Generated with Claude Code