Disney Decision Desk

Diátaxis: reference

Ask-first Disney decision support backed by curated Walt Disney World & Disneyland tips extracted from expert YouTube channels.

Live Site

• Production: https://disney.bound.tips/
• Health: https://disney.bound.tips/api/health

Commands

# Pipeline
npm run fetch            # RSS + transcript fetch (Disney YouTube channels)
npm run extract          # Gemini 2.5 Flash Lite tip extraction
npm run clean:tips       # Apply deterministic quality gates and regenerate public feed/sitemap/health
npm run embed            # Generate Gemini embeddings for semantic search
npm run backfill         # Retry videos with missing transcripts
npm run check-staleness  # Validate freshness / dist sync
npm run verify-live      # Check live site matches local data
npm run pilot -- readiness # Decision-plan pilot launch readiness
npm run pilot -- summary   # Decision-plan pilot validation ledger
npm run pipeline         # fetch -> extract -> check-staleness
npm run pipeline:deploy  # fetch -> backfill -> extract -> dedupe -> embed -> build -> check-staleness -> verify-live

# Frontend
npm run dev
npm run build
npm run preview

# Tests
npm test

Environment Variables

Stored in .env.local:

Required:

• GEMINI_API_KEY – Google AI Studio API key (tip extraction + embeddings)

Optional:

• GEMINI_MODEL (default: gemini-2.5-flash-lite)
• RESEND_API_KEY – Resend API key (email subscriptions and planning-request notifications)
• RESEND_AUDIENCE_ID – Resend audience ID
• PLAN_REQUEST_RECIPIENT – email address for new manual planning requests; required before outreach
• RESEND_FROM_EMAIL – optional sender for planning-request notification emails
• PLAN_PAYMENT_URL – optional payment link returned after a planning request
• SITE_URL (default: https://disney.bound.tips)
• PORT (default: 3000)
• WARP_PROXY_HOST (default: 127.0.0.1)
• WARP_PROXY_PORT (default: 1080)
• DENO_PATH (default: ~/.deno/bin/deno)
• TRANSCRIPT_STRICT_PREFLIGHT (default: true)
• TRANSCRIPT_USE_DENO_RUNTIME (default: true)
• TRANSCRIPT_TIMEOUT_MS (default: 30000)

Architecture

YouTube RSS → yt-dlp (+ WARP proxy) → data/pipeline/videos.json
           → Gemini extraction → data/public/tips.json
           → Gemini embeddings → data/public/embeddings.json
           → Vite build → dist/
           → Express server (port 3000) → Traefik → disney.bound.tips

Pipeline (scripts/): Runs via systemd timer. Fetches videos from the curated Disney YouTube source roster in shared/types.ts, extracts structured tips with Gemini, generates embeddings for semantic search.

Server (server/index.ts): Express app serving static files from dist/, plus API endpoints:

• POST /api/ask – Ask-first sourced preview for a natural-language Disney planning decision
• POST /api/search – Semantic search (Gemini embeddings) with text fallback
• POST /api/embed-query – Legacy Gemini query-vector endpoint; main UI uses /api/search and /api/ask
• POST /api/subscribe – Email subscription via Resend
• POST /api/planning-request – Captures the $39 manual decision-plan request to data/private/planning-requests.jsonl
• GET /api/health – Health check (tip count, embeddings, semantic search status)

Frontend (src/, index.html, plan.html, tips.html, design.md): Vite static site with an Ask-first decision desk, a paid human-reviewed decision-plan page with sample proof and intake form, and a searchable research archive with category/park filters and semantic/text search.

Key files:

scripts/
  fetch-videos.ts, extract-tips.ts, embed-tips.ts, dedupe-tips.ts,
  backfill-transcripts.ts, check-staleness.ts, verify-live.ts, prerender.ts
  lib/transcript.ts, lib/state.ts

server/index.ts                  Express server
shared/decisionPreview.ts        Ask-first sourced preview contract + decision area classifier
shared/planningRequest.ts        Manual decision-plan request validation + email formatting
shared/tipSearch.ts              Shared category/park/text search helpers
design.md                        Theme-park planning notebook style reference

data/
  pipeline/videos.json           Raw video metadata + transcripts
  pipeline/processed-videos.json Ledger of processed videos
  public/tips.json               Extracted structured tips (~4,000)
  public/embeddings.json         Gemini embeddings loaded server-side for semantic search
  private/planning-requests.jsonl Ignored local/private lead capture for the decision-plan pilot
  public/feed.xml                RSS 2.0 feed
  public/health.json, sitemap.xml, robots.txt

src/main.ts, src/styles.css      Frontend application
shared/types.ts                  Shared types (pipeline + frontend)

docs/planning-pilot.md           Manual decision-plan pilot operation and validation ledger
docs/first-customer-outreach.md  First five customer outreach scripts

Deploy

Container uses bind-mounted dist/ and data/public/, so pipeline rebuilds go live immediately without a container restart. The planning request flow also needs writable data/private/ mounted at /app/data/private so lead captures persist across container rebuilds.

npm run pipeline:deploy    # Full pipeline with live verification
npm run build              # Just rebuild static files
npm run verify-live        # Check live site matches local data
npm run pilot -- readiness # Check decision-plan pilot env/validation status

Troubleshooting

• Verify yt-dlp: yt-dlp --version
• Verify proxy: nc -z 127.0.0.1 1080
• Verify Deno path: ls ~/.deno/bin/deno
• Check live health: curl https://disney.bound.tips/api/health
• Check container mounts: docker inspect web-disney --format '{{json .Mounts}}'
• Use degraded mode: TRANSCRIPT_STRICT_PREFLIGHT=false