# Deployment Models Choose the deployment shape that matches where you are in rollout. ## Local Setup - Run services/dev-protection, services/dev-eval, services/api-gateway, and apps/developer locally. - Set a local base URL override only when your SDK calls should target http://localhost:4000 instead of the production API origin. - Use local providers or a staging deployment while you shape policy. ## Split-Plane Baseline - Use services/dev-protection for owner-facing control-plane traffic. - Use services/dev-eval for runtime evaluations behind the same gateway contract. - This is now the recommended baseline production shape. ## Split-Plane Rollout - Scale services/dev-eval independently from services/dev-protection when runtime volume grows. - Preserve the same SDK contract so application code does not change. - Runtime hardening now includes policy-head caching, stale-policy fallback, explicit idempotency handling, and a dedicated runtime load gate. - Google Cloud Run is a valid runtime target for backend-contract services as long as runtime API-key auth and forwarded IP handling stay explicit. ## Go-Live Checks - Runtime API keys scoped and rotated. - Decision logging visible in analytics. - Quotas and billing state verified. - Failure paths tested for 403, 409, 429, 503, and upstream timeouts. - tests/load/developer_runtime_load_gate.sh passing against the target environment.