Integrate reach-edu-hub into the shared shell

Why this plan exists

reach-edu-hub is the last client-site under dididecks-ai/client-sites/ that has never been wired to @dididecks/shell. It is:

• Not in the root pnpm-workspace.yaml (which lists chroma-decks, humain-vc-decks, lossless-decks), so workspace:* cannot even resolve the shell for it today.
• Shell-free — zero references to @dididecks/shell, dididecksShell, or deck-shell anywhere in its source.
• output: 'static', no auth, deliberately public (the sitemap decision is “Gating: none”).
• Rendering three bespoke scroll decks on a locally-ported PageAsDeckWrapper, with its own src/lib/scroll-decks.ts variant registry and src/lib/seo.ts, at custom URLs.

The integration follows the proven onboarding shape established by chroma-decks (the canonical template) and refined by humain-vc-decks. calmstorm-decks is the donor the shell was lifted from and is not an integration template (it does not consume the shell).

Decisions locked with the user (2026-06-29)

1. Integration depth: scroll-UI through the shell first. Defer the rigid 16:9 no-JS Play-UI per-slide-file conversion (deck-iteration-workflow Phase 2) to a later pass.
2. URL strategy: adopt the shell’s /scroll/[deck]/[variant] convention (migrate off the bespoke /story, /automation URLs).
3. Decks to create today: for reach-edu-hub, on the freshly-integrated shell.
4. Variant slugs: baseline / editorial (not v1/v2).
5. distributionTier: shared (the hub is public).
6. Old→new redirects: add them (keep any links already shared with Reach alive).
7. ModeToggle: keep reach’s own for this pass (limit theme-token risk); revisit swapping to the shell’s later.

Deck / variant model (current → shell convention)

Phase A — workspace + integration wiring

• A1. Added client-sites/reach-edu-hub to root pnpm-workspace.yaml.
• A2. Added "@dididecks/shell": "workspace:*" to reach’s package.json.
• A3. Added the dididecksShell({...}) integration to reach’s astro.config.mjs — client: "reach-edu-hub", four registry paths, distributionTier: "shared". Output flipped from static → server (see D1 finding below): under static, the shell’s injected-route getStaticPaths runs esbuild.transform inside Astro’s bundled prerender context, which has no __filename and crashes the build. server renders the injected routes on-demand — the same reason every other consumer uses server output.
• A4. Created reach’s standalone-Vercel guards: appended ignore-workspace=true (+ hoist/peer settings) to .npmrc, kept the jsr registry line, and created a local pnpm-workspace.yaml with onlyBuiltDependencies: [esbuild, sharp].
• A5. pnpm install from the monorepo root — 6 workspace projects, shell linked.

Phase B — contract files

• B1. Authored src/data/decks.ts (DECKS — story + automation).
• B2. Authored src/data/slides.ts (SLOTS for baseline, editorial, pipeline). Variant slugs are globally unique to avoid the variant-keyed collision (automation’s variant is pipeline, NOT baseline).
• B3. Created data/audits/slides.json = { "schema": 2, "ranks": {} }.

Phase C — migrate scroll pages to shell convention

• C1. Created the three pages at src/pages/scroll/{deck}/{variant}/index.astro with corrected import depths; deleted the old /story + /automation route dirs.
• C2. Converted reach’s local PageAsDeckWrapper into a thin shell re-export shim; added <DeckOverlay--Scroll-UI deckSlug variantSlug /> to each page.
• C3. Added data-slot/data-variant to all 29 T0x section components. (For TOC/discovery the manual slides.ts map is authoritative — the scanner can’t see through reach’s component imports; these annotations drive the in-scroll SlideRankPill IntersectionObserver.)
• C4. Added redirects (old URLs + bare /scroll, /scroll/story, /scroll/automation deck roots) → canonical variant routes.
• C5. Updated src/lib/scroll-decks.ts hrefs, src/lib/seo.ts keys, Header nav links + active-state detection, and the homepage showcase link.

Phase D — verify

• D1. Dev server: /scroll/story/baseline, /scroll/story/editorial, /scroll/automation/pipeline, /toc/story, /toc/story/baseline all 200; old /story + /automation 200 via redirect; /toc/story/baseline shows 9 slot rows; scroll overlay + ddd-deck-wrapper + SlideRankPill + section annotations all render. Reach’s theme.css tokens satisfy the shell chrome (it reads --ddd-* with fallbacks). Finding: output: 'static' is incompatible with the shell’s esbuild-in-prerender path → flipped to server.
• D2. pnpm --filter reach-edu-hub build clean (server output, Vercel adapter).

Phase E — adjacent session tracks (post-integration)

• New reach decks built as new variants/decks on the integrated shell (content specced when we get there).
• Shell tooling iteration: any friction this integration surfaces (slot discovery without annotations, token-contract gaps, static-vs-server tension) gets pushed back into apps/deck-shell rather than worked around in reach, and noted in [[Lift-Chroma-Decks-Generic-Code-into-Shared-Shell]].

Risks / watch-items

• output: 'static' + injected prerender=false routes. Siblings use output: 'server' because of auth; reach has none. Astro’s static output with the Vercel adapter supports per-route on-demand rendering, so static should hold. Fallback: flip to server (no DB/middleware needed, just SSR-capable).
• Theme-token contract. Shell chrome reads --color-* / --ddd-chrome-* tokens. Reach’s local PageAsDeckWrapper already uses --color-surface, --color-border, --color-text-muted, --color-text, so theme.css likely satisfies the contract — audit during D1, don’t assume.
• Slot discovery. Reach’s slides are separate T0x component files, not inline sections. They must carry data-slot / data-variant or the TOC comes up empty even with slides.ts populated.