Delivery Plan

Timeline overview

Days	Phase	Gate
1–2	Infrastructure + data layer	All services reachable; schema deployed; first sync completes
3–7	API surface + background jobs	All scenarios produce correct outputs
8–10	Frontend swap + parity run	Parity checks pass; staff validates workflows
11–14	Cutover	Production traffic on new system
15–21	Buffer / stabilization	Fix anything that surfaces post-cutover

Weeks 1–2 are active build. Week 3 is buffer — used only if needed.

Days 1–2: Infrastructure + data layer

• Create PlanetScale database and dev branch
• Run schema migrations via dbmate
• Create dashboard-api Worker with Hono skeleton
• Create ingest Worker with queue consumer skeleton
• Configure Hyperdrive binding to PlanetScale
• Set up R2 bucket + public domain
• Configure CF Access application (Google SSO, allowed emails)
• Create ingest-queue (Queue) + Cron Triggers
• Implement Drizzle PostgreSQL schema with all tables
• Build consolidated Shopify client module
• Implement bulk sync pipeline (start → poll → parse → write)
• Implement mart refresh with classification logic
• Implement watermark management + JSONL archival to R2
• Run first sync; verify raw_shopify_orders matches Supabase
• Verify: GET /api/health returns ok

Days 3–7: API surface + background jobs

All 14 scenarios are ported in parallel — CRUD routes and queue consumers built together since they share the same DB layer.

API routes

• Performance metrics CRUD + classification settings + summary stats
• Tag classifications (bulk upsert, unclassified)
• Tag groups CRUD
• Tag analytics (replicate RPC logic in TypeScript)
• Designer assignments CRUD + performance aggregation
• Combo logs CRUD + CSV import/export
• Combo templates CRUD + ID generation
• Order lookup
• Fulfillment monitoring summary endpoint
• CF Access middleware
• Queue producers (enqueue endpoints)

Background jobs

• Asset ingest queue consumer
• CSV export queue consumer
• Product tag sync queue consumer
• Combo suggestion generation
• Combo import queue consumer
• Wire Cron Triggers → daily sync + hourly mart refresh
• Verify: each scenario’s acceptance criteria pass

Days 8–10: Frontend swap + parity run

• Swap API base URL to dashboard-api Worker
• Remove Supabase auth code (AuthContext, login page, ProtectedRoute)
• Remove Passport login page
• Deploy to CF Pages
• Run parity checks
• Staff validates key workflows end-to-end
• Fix any discrepancies found

Days 11–14: Cutover

• DNS switch to new system
• Monitor for 48 hours
• Verify all observability alerts are operational
• Document runbooks: “validate daily sync”, “replay from R2 archive”, “interpret ingest failures”

Week 3 (buffer): Stabilization

Used only if needed. If cutover is clean, this week is for:

• Fix edge cases surfaced by real usage
• Tune batch sizes and cron schedules based on production data
• Decommission old system (after 2-week post-cutover holding period)

Success criteria

The rewrite is accepted when:

1. Operational parity: Every workflow used by Ops/Creative/Fulfillment/Design runs end-to-end
2. Security remediation: No unauthenticated endpoints; no hardcoded credentials; single auth plane
3. Reliability: Scheduled sync runs daily without manual intervention; queue retries handle transient failures
4. Performance: CRUD endpoints p99 < 200ms; daily sync < 15 minutes
5. Observability: All alerts operational; run history queryable; JSONL archives available for replay

Risk register

Risk	Likelihood	Impact	Mitigation
PlanetScale limits require more chunking than expected	Medium	Medium	Start with conservative batch sizes; tune in buffer week
Shopify bulk op API version constraints	Low	High	Pin API version; test concurrency on day 1
CF Access integration friction	Low	Medium	Test with small group on day 2; fallback to manual JWT
Data migration misses edge cases	Medium	Medium	Parity run on days 8–10 catches discrepancies
Frontend compatibility issues after auth removal	Low	Low	Deploy Pages with old API first, then swap base URL
Compressed timeline leaves no slack	Medium	Medium	Buffer week absorbs overflow; scope can be trimmed to critical scenarios first