Can I vibe code AB Tasty?

Build a lightweight A/B testing and feature flagging system for an e-commerce platform using Next.js, Cloudflare Workers, and PostgreSQL.

1. DATA MODEL:
- Experiments: id, key, status (draft/running/paused), salt, variants (json: key, weight, payload).
- FeatureFlags: id, key, description, enabled (boolean), rules (json targeting rules).
- Exposures: id, experiment_id, variant_key, user_id/anonymous_id, timestamp.
- Conversions: id, experiment_id, user_id/anonymous_id, goal_key, revenue, timestamp.

2. CORE FUNCTIONALITY:
- Edge Middleware Router: Intercept incoming requests at the Cloudflare Worker layer. Read anonymous user ID cookie or assign a new UUID. Hash (user_id + experiment.salt) using SHA-256 to assign deterministic variant buckets (0-99).
- Anti-Flicker / Server-Side Redirects: Execute URL split tests directly at the edge by rewriting HTTP requests to alternate paths before rendering HTML.
- API Service: REST API to CRUD experiments and feature flags, fetch exposure reports, and ingest conversion events.
- Stats Calculation Engine: Provide an endpoint that calculates z-score, p-value, and conversion rates between Control and Treatment variants using frequentist statistical formulas.

3. INTEGRATION:
- JavaScript SDK: 1KB client-side snippet to trigger conversion events (e.g., `track('add_to_cart', { revenue: 49.99 })`) sending payload to the ingestion API.

4. FAILURE MODES TO HANDLE:
- Missing cookies: Fallback gracefully to random transient assignment without throwing runtime errors.
- Database downtime: Edge middleware must cache active flag states in KV storage so experiment evaluation never blocks page loading.

5. OUT OF SCOPE:
- No visual drag-and-drop web editor.
- No complex multi-armed bandit automated re-weighting.

$ each button prefixes agent-specific run instructions · build your own product, never copy proprietary code, trademarks or designs