Map the critical path
Start with a service catalog and the workflow it supports. A dependency graph should include declared failure domains and missing observations.
A dependency is a condition needed for the workflow to function, not just a line between systems. Record the effect of losing it, who owns it and how its recovery will be observed. Distinguish discovered relationships from authored assumptions and mark dependencies that have not been observed.
- Assign business and technical owners.
- Mark required dependencies and quorum assumptions.
- Record declared recovery and service targets.
Author the interruption
Specify the outage, duration and recovery steps. Treat repair times as assumptions unless independently measured.
Choose a specific interruption and define its start, duration and affected failure domain. Include delayed recovery and a second interruption where relevant. Repair times used by a model are assumptions unless supported by measurements from the intended operating environment.
Recovery order follows declared prerequisites
A restored service may still depend on valid identity and access.
- Test partial restoration and repeated failure.
- Compare more than one recovery order.
- Retain all model and scenario versions.
Review business continuity
Inspect service impact together with transaction correctness. Meeting a modeled target does not establish real infrastructure recovery performance.
A process can become technically reachable while the business service remains inconsistent. Check pending work, duplicated effects, required state and dependent services after restoration. Report target breaches and unresolved outcomes alongside modeled availability.
What happens if a synthetic order service returns before its queue is reconciled?
Model the outage with fixed initial state. Compare two recovery sequences and inspect pending orders, duplicate processing and the point at which the business workflow becomes usable.
A faster process restart is not necessarily faster service recovery. Review unresolved orders and state consistency before accepting the sequence.
- Check business invariants after restoration.
- Identify target breaches and single points of failure.
- Plan deployment-specific recovery qualification.
Compare recovery sequences
Use the same interruption and initial conditions when comparing restoration orders. A sequence that restarts the application first may be ineffective if its data or identity dependency is still unavailable.
Model alternative orders only within the supported scenario contract. Explain which prerequisites cause delay and which steps can proceed independently. A visual dependency graph is a review aid, not proof that every production dependency has been discovered.
Measure the right recovery milestone
| Milestone | Required observation | Common gap |
|---|---|---|
| Component available | The process or dependency is reachable. | Its state or prerequisites may still be unusable. |
| Workflow restored | The declared end-to-end path completes. | Pending work and exception paths may remain unresolved. |
| Business state verified | Assertions and reconciliation have been reviewed. | Untested failure domains still require qualification. |
- Record prerequisites and failure domains.
- Keep repair assumptions visible.
- Compare service readiness, not only component start times.
Protect state during recovery
A recovery plan needs to account for state identity, retained receipts and custody as well as backups. Restoring an older database can reintroduce state that conflicts with later erasure or custody decisions.
For the custody-bound internal environment, retain the independent custody journal and use a compatible runtime. Qualify any customer restore procedure against its own environment and targets. Preserve recovery evidence rather than overwriting it with a successful rerun.
Keep the evidence and its conditions together. A later success does not erase an earlier failure or broaden the original scope.
- Retain backups, checkpoints and failed recovery outcomes.
- Review uncertain downstream effects before another dispatch.
- Verify business state after restoration.
Prepare the outage rehearsal
Assign a business owner and technical owners for the critical dependencies. Agree the interruption, stop conditions, observations and review criteria before the rehearsal.
Use the checklist to prepare a controlled evaluation. A modeled recovery time is not a measured infrastructure guarantee. Physical-host and independent-site recovery require their own deployment-specific qualification.
Review preparation
0 / 6 prepared- Name the interruption and restoration order.
- Declare the recovery and service targets.
- Separate modeled outcomes from measured recovery evidence.