On this page

Perfect Deep Dive Page Contract

Perfect Deep Dive Page Contract (v1)

Status: Active — drafted 2026-06-11 per the 2026-05-09 deep-dive handoff, before any C24 code edits. Golden specimen: c24. Validation specimens: sc48, webtv-cortez. Premium-blocked: avegant-glyph (needs archive→R2_MASTER curation first). Governing issues: #68 (C24 normalization), #72 (Epic W4), #67 (asset routing, parallel/subordinate).

A deep-dive page is Perfect when page story, schema data, assets, provenance, local dev behavior, mobile behavior, HUD behavior, source trail, and agentic consumption all agree. “Ready state” (structurally hydrated, body text dumped at the end) is explicitly NOT done.

The Four Gates

Every page passes the gates in order. A gate failure stops promotion; do not patch downstream symptoms.

  1. Source truth — evidence exists in the asset pipeline and NLM registry, not just in prose.
  2. Schema hydration — structured frontmatter carries the data; hydration runs through hydrate_content.py, never manual MDX-only edits that the hydrator would overwrite.
  3. Detail-page composition — one authored narrative, components activated, no empty states.
  4. Visual / congruence verification — rendered page agrees with schema and source on desktop and mobile.

Exit Criteria

A. Identity & machine readability

  • title is the real product name (not a bare codename slug).
  • description present, 140–300 chars, states role + product + quantified outcome.
  • Frontmatter validates against the projects schema; no banned fields (forensic_data is z.never()).
  • employer, category, industry, tools, date, endDate, production, productionScale populated.
  • draft: false and the page is intentionally listed (or intentionally unlisted, recorded here).

B. Story

  • Exactly one authored narrative body. No duplicate or machine-stitched report blocks, no dumped text after the structured content.
  • Each crisis is told as Trigger → Intervention → Result with quantified impact.
  • No unresolved citation artifacts (e.g. [64-66]) in visitor-facing prose; provenance moves to the source trail (§D).
  • Numbers section exists and is deduplicated (one canonical figure per claim).

C. Structured data

  • cast populated and rendering (DossierCast).
  • forensic_summary (trigger/intervention/result) populated.
  • forensic_metrics and/or metrics populated from source truth as available; timeline, bom, scars wired as W1 data lands (absence is acceptable, empty-but-promised is not).
  • cyberspace.stickies populated; every sticky has id + title (HUD labels resolve via label → title → id).
  • Deep Data HUD opens on the first tab that has data; tabs without data do not render.

D. Assets, provenance & source trail

  • Every referenced asset (hero, gallery, audio, 3D) returns 200 locally (served from R2_MIRROR via /assets/...) and in production (R2).
  • Assets entered the page only via the pipeline: local evidence store → R2_MASTER (curation) → R2_MIRROR (processed) → R2. Never point page content at a raw archive directly.
  • Asset paths use /assets/r2/{slug}/... or /assets/{slug}/... (both tolerated until #67 canonicalizes; new hydration emits /assets/r2/).
  • A visible source trail exists on the page: the evidence artifacts (ECOs, DCDs, inspection reports, photos) are named so a reader can see what the claims rest on.
  • audio_url / notebook_url / nlm_url set when the artifact exists.

E. Behavior

  • Fixed HUD is legible over long-form content (glass background once scrolled past the intro; never transparent over body text).
  • Scrolly dot nav has real accessible labels (no “Scroll to undefined”).
  • No inert controls on any viewport — mobile menu either works or is not rendered.
  • No debug logging (console.log) in production components.
  • No hard runtime errors in the browser console on load or scroll-through.

F. Agentic consumption

  • Core narrative renders server-side (visible to crawlers/LLMs without JS).
  • title + description land in the page <head>.
  • Structured frontmatter is the machine-readable source for the headless surface (Epic #31).

Validation Procedure

  1. npm run build passes; page present in dist.
  2. Desktop screenshot: intro, mid-scroll (HUD over text), dossier open.
  3. Mobile screenshot: hero, nav surface (no dead buttons).
  4. Console check: zero errors, zero debug logs.
  5. Criteria boxes above all check; record the pass in the page’s PR.

Ready-State Cohort (audited 2026-06-11: 22 pages)

320-slot-optical-carousel, avegant-glyph, backsplash, bazooka, c24, cinema-one, d-command, d-control, dispensers, dv700, extension-switches, kplayer-6000, kserver-1500, kserver-5000, ksystem-120, m700, makeline, motorola-mp3, portion-cup, sc48, webtv-cortez, webtv-galaxy

All 22 slugs verified to exist as project directories. Scale-out order after C24: validation specimens first (sc48, webtv-cortez), then the cohort by asset readiness; avegant-glyph only after its curation pass.