Network and security architecture
Trust boundaries
| Boundary | Primary controls |
|---|---|
| Internet to static application | HTTPS, Cloudflare edge delivery, baseline DDoS mitigation, and configured edge rules |
| Browser to managed backend | TLS, publishable client identity, signed user session, request validation, grants, and row-level policies |
| Managed gateway to data | Database roles, object permissions, row-level policies, constraints, and restricted trusted operations |
| Serverless workflow to external provider | Server-side credentials, provider authentication, input minimisation, timeouts, callback verification, and auditable status |
| Application to monitoring | Configured diagnostic context, source-map handling, masking, sampling, retention, and access controls |
Segmentation model
The static delivery plane is separated from the stateful backend. Ordinary browsers do not receive database passwords or privileged service credentials. Application data access is mediated by the managed gateway and database authorization. Trusted serverless operations are an elevated boundary. Selected operations revalidate the caller, tenant, role, target, and input; complete operation coverage is not available.
Allowed communication paths
| Source | Destination | Protocol and identity | Permitted purpose | Default failure behavior |
|---|---|---|---|---|
| User browser | Cloudflare edge | HTTPS with certificate validation | Load the approved static release | Refuse insecure transport and expose no backend credential |
| Browser application | Supabase gateway | TLS, publishable client context, and signed user session | Authentication, policy-controlled data, object, realtime, and approved function requests | Deny protected access without a valid session or policy result |
| Trusted function | PostgreSQL or object storage | Provider-internal service context and scoped server credential | Privileged transaction, scheduled work, or object processing | Abort on failed actor, tenant, role, input, or ownership validation |
| Trusted function | Approved provider | TLS and provider-specific credential | Messaging, payment, quote, credit, call, or reporting workflow | Record an explicit provider failure or indeterminate state; do not assume completion |
| Approved provider | Trusted callback | HTTPS plus signature, shared secret, or equivalent authenticity proof | Deliver a correlated event or final provider state | Reject unverifiable or replayed requests before business mutation |
| Browser and backend | Monitoring | TLS with project-scoped ingestion context | Errors, traces, release, and selected diagnostic telemetry | Application correctness does not depend on monitoring availability |
Network threat treatment
| Threat | Architectural treatment |
|---|---|
| Direct database exposure | Application traffic enters through managed APIs and policies; no application-owned public database listener is part of the topology |
| Browser tampering | Browser state and route visibility are non-authoritative; protected operations re-evaluate identity and authorization at the backend |
| Cross-tenant access | Signed identity, tenant membership, hierarchy, ownership, and row policy reduce every ordinary request to the permitted record set |
| Credential disclosure | Public clients receive only publishable configuration; provider and privileged credentials remain in trusted configuration |
| Callback forgery | Provider callbacks are treated as untrusted until authenticity, correlation, freshness, and idempotency checks pass |
| Data exfiltration through integrations | Destinations are approved by purpose; payloads are minimized and server-mediated where credentials or sensitive business state are involved |
| Denial of service and abuse | Provider baselines address network availability; application rate limits and business-flow controls address account-level abuse |
Limitations
This document intentionally omits domains, IP addresses, account identifiers, provider endpoints, internal object names, and secrets. It does not prove current DNS, WAF, network restriction, TLS-policy, or egress-allowlist configuration; those require account-level evidence.