Production Readiness Layer Pack

This pack turns “real full stack is more than frontend plus backend” into a control-level assurance model. It is a lane pack inside Valdris, not the whole SDLC.

Thirteen layers

Layer	Enterprise question	Catalog controls
Frontend	Are critical client journeys accessible, correct, and within performance budgets?	accessibility, behavior, performance
Backend/API logic	Are contracts compatible and timeout/retry/idempotency/failure paths observable?	contract, failure, trace
Database/storage	Are migrations, integrity boundaries, retention, backup, and restore proven?	migration, integrity, recovery
Auth/permissions/RLS	Does every operation enforce server-side and tenant-safe authorization across the identity lifecycle?	authz, tenant, lifecycle
Hosting/deployment	Is promotion attributable, target-environment health measured, and rollback exercised?	promotion, health, rollback
Cloud/compute	Is topology reproducible, least-privilege, drift-aware, regional, owned, and cost-bounded?	IaC, boundary, cost
CI/CD/version control	Do required checks, supply-chain scans, provenance, and approvals protect the release?	gates, supply chain, provenance
Security	Are current threats, secrets, vulnerabilities, and trust boundaries governed?	threat, secrets, vulnerability
Rate limiting	Are limits, overload behavior, retry storms, and usage meters controlled?	policy, failure, metering
Caching/CDN	Are identity-aware keys, invalidation, freshness, purge, and tenant isolation correct?	key, invalidation, isolation
Load balancing/scaling	Are capacity, progressive delivery, abort, routing, and failure modes tested?	capacity, progressive, failover
Error tracking/logs/observability	Are flows correlated, SLO-owned, alertable, and privacy-safe?	telemetry, SLO, redaction
Availability/recovery/DR	Are workload-specific objectives, restore/failover, and incident roles exercised?	objectives, restore, incident

The machine catalog is controls/production-layers.v2.json. Its dependency fields form an acyclic assurance prerequisite graph.

Per-run artifact

Every production-impacting run writes production/layer-assessment.json using uash.production-readiness.v2.

A required layer must name its owner and every catalog control. Every passing control contains typed evidence. A non-applicable layer uses applicability: "not-applicable", status: "skipped", and a precise reason. Intake may temporarily mark a layer potentially-affected in run/route.json, but that state must resolve before the v2 finish-line artifact is written.

Typed evidence is validated rather than trusted as prose:

• repository-contained, hash-bound artifacts;
• zero-exit commands with output digests;
• metrics whose operators and thresholds are recomputed;
• scoped approvals granted by a named human;
• digest-bound provider reports.

Enterprise and regulated artifact evidence also binds commit and environment. See docs/ENTERPRISE_CONTROL_MODEL_V2.md for the complete contract.

Validate

npm run production:gate

Legacy uash.production-readiness.v1 history remains readable, but the commission generator emits v2.

Skip and finish rules

• No canonical layer may be silently omitted.
• A required layer or control cannot be skipped.
• failed, pending, blocked, required, and needs_approval block completion.
• Approval authorizes a risk decision; it does not turn missing or failing technical evidence into a pass.
• The finish line requires every applicable control to pass and every non-applicable layer to carry a workload-specific reason.