16:9 Export Variant and the PPT/PDF Pipeline

What the operator asked for

A variant of each pitch deck that is “more or less the same” but fits exactly into 16:9 with no scroll behavior or animations, in preparation for export to PDF and PPT — where the PPT must be near-perfect because the team works in Google Slides (imports from PPT with few glitches when the markup is elegant).

Why this is the Play-UI phase, by the book

The dididecks discipline already names this exactly (repo CLAUDE.md): every deck has two coordinated implementations —

• Scroll-UI (built 2026-07-28): responsive sections, scroll-snap, inline scripts allowed. This is where we design.
• Play-UI: standalone per-slide files at src/components/slides/{variant}/{slot}-{slug}.astro, rendered by the shell’s /play/[deck]/[variant]/[slot] route inside SlideCanvas — rigid 1920×1080, no responsive CSS, no JS in the slide — which letterboxes to any viewport and exports to PDF without hydration. The shell already ships /play/.../print.astro for the PDF path.

So “the 16:9 variant” = the Play-UI conversion of each *-p1 variant, slot-for-slot per slides.ts (recreate, don’t extract — per /api/slide-decompose discipline).

The load-bearing anticipation: content as data

Every Scroll-UI section built in this pass keeps its content in typed frontmatter arrays (stats, kpis, rows, steps, tiers, arc) plus a headline/eyebrow — not prose baked into markup. That makes three consumers possible from one source:

1. Scroll-UI section (exists)
2. Play-UI slide (this plan)
3. PPT exporter (this plan)

Refactor step (do first): lift each deck’s content arrays out of the section frontmatter into src/layouts/sections/{deck}/content.ts (typed, per-slot exports), and make sections import from it. Citations already live in per-deck citations.ts with canonical hex codes and build-time [n] indexing — the exporter reuses them as-is.

PPT pipeline (the part that must be perfect)

Approach: generate native PPT shapes with pptxgenjs — never raster. Google Slides imports native text boxes, autoshapes, and tables with high fidelity; it mangles embedded images-of-text and ignores exotic effects.

• One Node script per client: scripts/export-pptx.mjs reading content.ts + citations.ts + slides.ts per deck → one .pptx per deck (and optionally one combined).
• Layout: 13.333in × 7.5in (16:9). A small layout vocabulary mirroring the deck primitives: cover, statement (centered quote), stat-grid (2–4 cols of stat cards as rounded rectangles + text), list-cards, table-wall, sources.
• Tokens → PPT theme: export in light mode values from theme.css Tier-2 tokens (background, text, muted, accent, secondary). Gradient headline spans render as solid accent color in PPT (gradient-filled text does not survive Slides import).
• Fonts: map to Google-Slides-native faces (custom webfonts do not embed through Slides import). Heading → a Slides-safe geometric sans; body → Arial/Inter fallback. Record the mapping in DESIGN.md when it exists.
• Citations in PPT: inline [n] as superscript text runs (pptxgenjs superscript: true); each deck ends with paginated Sources slides — a PPT can’t scroll, so chunk the ordered citation list ~10–12 per 16:9 slide (Sources 1/3, 2/3 …). Hex codes stay in the speaker notes per slide for traceability.
• No animations, no transitions — by construction.
• Verification loop: import the generated PPT into Google Slides and walk it (browser-drive per the verification blueprint) before it ever goes to the team; glitches found → fix the generator, not the file.

PDF pipeline

Primary: the shell’s /play/[deck]/[variant]/print route (all slides, 1920×1080, no hydration) → headless Chromium print-to-PDF (Playwright), one PDF per deck. The Sources slides use the same pagination as PPT (no scroll in print). Fallback: pptx → PDF via LibreOffice headless if a single toolchain is preferred.

Sequencing

1. Content-module refactor — lift arrays to content.ts per deck; wire Scroll-UI sections to import them (no visual change; verify with build + smoke).
2. Play-UI exemplar — convert upward-mobility slot-by-slot (src/components/slides/um-p1/{slot}-{slug}.astro), register in the shell chooser; operator walks it.
3. PPT exporter exemplar — export-pptx.mjs for upward-mobility; import into Google Slides; iterate until clean.
4. Fan out both to the remaining seven decks (mechanical once the vocabulary is proven).
5. PDF batch via the print route.

Constraints to respect (learned this pass)

• Sources lists exceed one screen — pagination is a first-class layout, not an afterthought (Scroll-UI got SourcesPager; Play/PPT get numbered Sources slides).
• Header/chrome: Play-UI and exports have no site chrome at all — the scroll deck’s auto-hide header and c-toggle are Scroll-UI-only concerns.
• Overwhelm slides (KPI walls, money maps) must be re-typeset for fixed 1920×1080 — auto-fit grids don’t exist in PPT; the content.ts arrays make per-slot column counts an explicit exporter parameter.