Production Service Readiness

Production Service Readiness

Read Production Service Readiness: templates guidance with evidence limits, practical checks and editable source material.

Use this record for one declared workflow slice and release before shadow, canary, bounded execution, or customer handoff. It assembles target-specific evidence against the Guide's existing controls; it is not a certification, a generic infrastructure checklist, or a substitute for the exact release record and approval path.

Do not calculate a readiness score. One unresolved high-consequence boundary can hold a release even when every other row is strong.

Assessment identity

Field Value
Workflow, segment, and environment
Release ID, version, and digest
Requested release gate or rollout Design / sandbox / shadow / canary / bounded segment / full segment / handoff
Actor mode and maximum effect
Assessment owner and date
Technical, operational, risk, and receiving-service reviewers
Evidence cutoff and review due

Status contract

Use only the following statuses. Status describes the maturity of evidence for the declared slice; it does not authorize release.

Status Meaning
required The dimension applies, but its evidence has not yet been assessed.
not applicable The dimension does not apply to this slice; record the owner-approved rationale and the trigger that would make it apply.
unresolved The dimension applies and required evidence, ownership, or a design decision is missing, stale, or conflicting.
designed A bounded design and owner exist, but target-specific behavior has not been exercised.
tested Current target-specific evidence exercises the declared design under representative and negative cases; production operation is not implied.
operational Current production evidence shows the declared control is owned, monitored, and recoverable for the named segment and review window.

Architecture prose, intended configuration, generated output, and model judgment cannot by themselves advance a row beyond designed. tested requires executable target-system evidence. operational additionally requires current production telemetry, exercised ownership, and recovery evidence.

Readiness matrix

Start every row as required. Change a row to not applicable only with a named owner, target-specific rationale, and an applicability trigger. Keep customer or target-system evidence separate from Guide control references.

Dimension Status Accountable owner Target-system evidence and revision Gap and next proof Guide controls
Identity and authorization required IAM-001, IAM-002, IAM-003, SEC-005
Source integration, reconciliation, and environment constraints required CTX-001, CTX-005, CTX-006, CTX-008, CTX-009, DEL-002, SEC-006, SEC-007
Durable state and data recovery required STA-001, STA-002, STA-003, OPS-005
Async work, concurrency, retries, and idempotency required REL-001, REL-002, REL-004, OPS-005
Rate, cost, and capacity limits required REL-002, TOL-005, CST-001, CST-002
Telemetry, alerts, and service objectives required OPS-001, OPS-004, OPS-006
Failure, degraded operation, and rollback required DEL-002, OPS-002, OPS-003, CTX-009
Evaluation and human-AI deployment qualification required EVA-001, EVA-002, EVA-003, EVA-005, EVA-006, HUM-003
Operator workflow, adoption, and review capacity required ADP-001, FDE-003, VAL-002, OPS-006
Scaling assumptions and limits required REL-004, SEC-005, CST-002
Service ownership and evidence lifecycle required ADP-002, DEL-001, OPS-007

Dimension prompts

Identity and authorization

  • Name interactive and unattended identities, tenant binding, current scopes, policy revision, and the action-boundary recheck.
  • Record denial, stale-policy, deprovisioning, and cross-tenant tests where applicable.

Source integration, reconciliation, and environment constraints

  • Name each source owner, interface, extraction cursor, schema and semantic revision, freshness objective, deletion and late-arrival behavior, transformation build, quarantine, and independent reconciliation.
  • Exercise incomplete extracts, unmatched keys, corrected history, backfills, schema drift, restricted-network promotion, artifact provenance, credential rotation, and rollback where applicable.
  • A completed request, file load, or batch proves transport only. Advance the row to tested only when target-system evidence reconciles the declared business population and failure cases.
  • Use Enterprise Integration and Scale Reality to map local or teaching components to target responsibilities without mandating one infrastructure stack.

Durable state and data recovery

  • Classify workflow and business state; record source of truth, retention, deletion, recovery point, replay, and migration behavior.
  • A database is required only when the workflow needs durable or queryable state. Record not applicable for genuinely stateless work with an owner-approved rationale.

Async work, concurrency, retries, and idempotency

  • Record timeout, cancellation, retry, duplicate-delivery, partial-failure, and conflict semantics for the selected route.
  • A queue or durable workflow engine is required only when work outlives a request, must survive interruption, or needs controlled fan-out. Synchronous bounded work may be not applicable with evidence.

Rate, cost, and capacity limits

  • Bind per-run and service-level limits for requests, steps, retries, parallelism, result size, latency, spend, and human review.
  • Record terminal behavior, backpressure, escalation, and the named capacity owner.

Telemetry, alerts, and service objectives

  • Bind trace identity, context, policy, state, tools, effects, readback, outcome, cost, and stop reason to the exact release.
  • Define service objectives and exercise alert routes; dashboards or intended alerts alone remain designed.

Failure, degraded operation, and rollback

  • For every severe failure, record detection, containment, safe fallback, readback, recovery, owner, regression, and rollback trigger.
  • Show that new work, writes, identities, egress, and affected capabilities can be stopped without hiding pending effects.

Evaluation and human-AI deployment qualification

  • Bind the tested workflow claim, eligible population, cases, environment, components, resource budgets, trial semantics, uncertainty, contamination controls, and independent decision to the exact release.
  • When the claim depends on human review or routing, bind the frozen oversight policy, disjoint development and qualification evidence, sampling unit and dependence, reliability lower bound, autonomous coverage, review burden, reviewer-effectiveness basis, latency, capacity, total cost, risk constraints, and requalification triggers.
  • Use saved terminal trajectories only when intervention cannot alter subsequent execution. Otherwise require a representative policy-in-loop rerun or faithful simulation.
  • A benchmark, point estimate, assumed reviewer, or unexercised queue cannot advance this row to tested.

Operator workflow, adoption, and review capacity

  • Exercise the complete operating path: initiating actor, applicable scope and trigger, bounded orchestration and state, governed context, selected mechanisms, evaluation, abstention and escalation, controlled effect, source-of-truth readback, and durable result destination.
  • Exercise the final persistent surface with representative operators: find the work, inspect evidence and uncertainty, correct, reject, escalate, interrupt, and resume without losing state or audit history.
  • Reconcile eligible, exposed, completed, independently accepted, and business-effect events to the frozen adoption contract. Sample non-use and abandonment by reason rather than treating logins or runs as adoption.
  • Name access, training, support, domain-review, and escalation owners. Exercise expected reviewer/support load, queue and wait limits, overflow, degraded operation, and the manual fallback for the proposed cohort.
  • Advance this row to tested only when target users and the receiving team have exercised the path. Documentation, a demo, or delivery-team-led usage remains designed.

Scaling assumptions and limits

  • State the expected workload, burst, tenant count, payload size, dependency ceilings, hot keys, contention points, and reviewer capacity.
  • Caching and horizontal scaling are required only when measured load, latency, availability, or isolation demands them. When used, bind cache keys and invalidation to tenant and source revisions, and test scale-out state and duplicate safety.

Service ownership and evidence lifecycle

  • Name the receiving service owner, on-call and escalation path, change approvers, support route, evidence retention, re-review cadence, and retirement owner.
  • Link the exact release, evaluation, rollout, rollback, handoff, and training evidence; documents alone cannot prove exercised operating capability.

Decision and handoff

Field Value
Decision Hold / reject / approve declared rollout / accept handoff with dated conditions
Blocking dimensions and why
Time-bounded remediation and owners
Accepted residual risks, approving principals, and expiry
Rollout scope and stop triggers
Verified rollback or restoration evidence
Evidence required for the next gate

Bind the decision to the declared release version and digest. A changed behavioral, data, policy, runtime, capability, evaluator, or operating dependency requires impact review and may invalidate prior evidence. Use the release gates, change management, and the applicable target-software or model/agent release record for promotion.