Application errors
Sentry or similar should catch server/client errors, route failures, and unexpected exceptions.
- Server errors.
- Client errors.
- Source maps configured safely.
Watchtower
Monitoring and alerts
After launch, signals are needed: Stripe failures, OpenAI errors, Supabase errors, email failures, cost spikes, and user complaints.
A site does not become reliable by itself. You must hear it when it coughs.
Payment watch
Stripe webhook failures, duplicate payments, failed checkout, and disputed statuses should reach logs/dashboard.
- Webhook failures alert.
- Duplicate detection.
- Payment-to-reading link.
AI and cost guard
OpenAI errors, quality gate warnings, repair attempts, and cost spikes should be visible before a client sends an angry email.
- AI job failures.
- Quality gate warnings.
- Daily cost threshold.
checklist
What to check
- Stripe errors are visible.
- OpenAI errors are visible.
- Supabase errors are visible.
- Email failures do not stay silent.
- Manual support workflow exists.
red flags
When to stop
- Client reports an error before logs do.
- AI cost grows without reason.
- Webhook fails and nobody knows.
- Support cannot link payment and reading.
done when
Definition of done
- Alerts exist for critical errors.
- Dashboard or clear logs path exists.
- Support can restore access by email/payment.
related doors
Stripe test flowCheckout, webhook, paid status, AI, receipt, and access.Resend emailSender, receipt, recovery, and Oracle Letters.Rollback planWhat to do if something breaks after deploy.Go-live dayThe order of actions on launch day and after first payments.SupportLost links, payments, accessibility, and errors.