Reliability: liveness+readiness+startup probes; resource requests/limits; PDB; multi-AZ; HPA

Rollout: immutable image by digest; canary/blue-green; --atomic Helm; tested rollback

Security: Workload Identity (no secrets); Key Vault CSI; signed images; NetworkPolicy; non-root

Networking: Private Endpoints for data; NSGs; ingress + TLS; WAF if internet-facing

Observability: golden-signal dashboards; SLOs + burn-rate alerts; structured logs; tracing

Data: backups + tested restore; expand/contract migrations; connection-pool sizing

Cost: right-sized requests; Spot where tolerant; tags for showback; budget alert

Ops: runbook; on-call + escalation; capacity/load test; chaos/failure drill

A production-readiness review is the staff-level synthesis of everything: reliability (correct probes, requests/limits, PodDisruptionBudgets, multi-AZ spread, autoscaling), safe rollout (immutable digests, progressive delivery, atomic deploys, a tested rollback), security (secret-less Workload Identity, Key Vault integration, signed images enforced at admission, NetworkPolicies, non-root containers), networking (Private Endpoints for data, ingress TLS, WAF when exposed), observability (golden-signal dashboards, SLOs with burn-rate alerts, logs, tracing), data (backups with tested restores, expand/contract migrations), cost (right-sizing, Spot, tagging, budgets), and operational readiness (runbook, on-call, load and failure testing). The signal interviewers want: you treat "going live" as a deliberate, checklist-driven gate — and ideally codify it as an automated, self-service readiness check so the safe path is the default.