approval-engine/deploy/README.md

Deployment gates

The StatefulSet is deployed as of 2026-09-14. The metadata receipt is docs/evidence/2026-09-14-approval-engine-deployment.json. SQLite is intentionally limited to one replica, ReadWriteOnce storage, and an OnDelete update: never start two writers against a copied database.

Image: pinned, published and deployed. Both image: references carry sha256:251941a5cb2724b57cc32cff6b693b1ab0be695bee4f56f02d51961189c0fa49 (registry tag 0.1.0-hfact-be1a388, source be1a388). Both references MUST stay identical immutable digests: migration and server share one database.

The deployed artifact carries schema v5 and predates schema v6 in this source. Schema v6 records verified entries.tenant_source; it requires a new scanned, published image, identical migration/server pins, backup and an approved rollout. The existing deployed image records: entries.principal_type and explicit approvals.human_control. Its release scan found zero HIGH/CRITICAL vulnerabilities. Disposable on-disk v4-to-v5 migration, negative service/agent binding, human binding/consumption and restart persistence passed inside the non-root container with no network and a read-only root. See the image evidence. This is artifact evidence; native identity, audit delivery and production restore remain deployment gates. Never roll the old v3 image over a v5 database; use the matching verified pre-migration backup for rollback.

The existing KeyCape consumer registration and CCR-2026-0021 audit projection were already verified by their owners. Live inspection confirmed the projection Ready, exact key name, receiver c82e0442 and existing producer ingress; custody was reused without reading or rotating its value. Added the exact namespace AND pod-label ingress to KeyCape for JWKS (deploy/keycape-ingress.yaml).

The pod requests 25m CPU/64Mi with unchanged 500m/256Mi limits: the node had 3975m of its 4000m CPU reserved, so the prior 50m request could not schedule. Initial native observation was 1m CPU/20Mi. No other workload was resized. Kubernetes token automount is disabled; the service uses no Kubernetes API.

The live service passed schema-v5 integrity, current JWKS reachability, anonymous and invalid-bearer 401, first heartbeat delivery with zero pending outbox, pod replacement with unchanged persisted outbox, verified online backup and an isolated restore opened by the actual engine. An independent 0600 backup copy was inspected locally. No live database was overwritten. The initial backup copies are test evidence, not a long-term backup schedule or disaster-recovery acceptance. Capacity and independent retention remain operational follow-up.

The real PEP adoption is still T05: CCR-2026-0019's operator group/reader, requesting identity and real human approval are not supplied by deployment.

For a future reinstall or release:

1. Register the exact approval-engine audience, caller scopes, and service clients in KeyCape. Confirm the in-cluster JWKS endpoint and configured issuer match the manifest.
2. Register approval-engine as an audit-core sender, add matching audit-core ingress, and provision approval-engine-audit through the credential owner. The Secret must contain key audit-token; never commit its value.
3. For a new release only: make image-release (build, scan, push — the scan gate fails the build on any CRITICAL or HIGH, so a failing scan blocks the push), then pin both image: references to the returned registry digest and run make deploy-dry-run. The current digest is already pinned; skip to 4 unless you are cutting a new image.
4. Take a verified backup. Roll out by deleting the sole pod, then prove readiness, restart persistence, outbox drain, heartbeat, and restore.
5. Enable a caller namespace only by applying label railiance.io/approval-engine-client=true; JWT scope checks remain the inner boundary.

The audit-core receiver-side registration, cadence findings, and accepted-count reconciliation are tracked by AUDIT-WP-0009-T04/T06/T09 and cannot be closed from this repository.