Quantum operators should not meet their incident playbooks for the first time during a real outage. Exercises convert rare failures into rehearsed behavior. They also reveal hidden dependencies, weak handoffs, and ambiguous authority before the platform is under stress.
NIST SP 800-84 provides guidance for test, training, and exercise programs for IT plans and capabilities [R193]. Quantum platforms can adapt that structure to calibration failures, evidence gaps, provider outages, corrupted telemetry, and invalid scientific claims.
View diagram source
flowchart LR
Plan[Exercise plan] --> Scenario[Scenario]
Scenario --> Conduct[Conduct exercise]
Conduct --> Observe[Observe decisions]
Observe --> Evaluate[Evaluate gaps]
Evaluate --> Improve[Improve runbooks]
Improve --> PlanExercise types
| Type | Purpose | Quantum example |
|---|---|---|
| tabletop | clarify decisions and authority | provider calibration data becomes unavailable |
| simulation | test workflows without live disruption | replay queue outage in staging broker |
| game day | intentionally inject controlled failure | block evidence writes for low-risk test jobs |
| red-team style | probe misuse and abuse cases | attempt policy bypass for restricted workload |
| science review drill | test claim governance | challenge benchmark result before publication |
View diagram source
mindmap
root((Exercise types))
Tabletop
Simulation
Game day
Red-team style
Science review drillQuantum resilience scenarios
View diagram source
flowchart TB
Scenario[Scenario catalog] --> Provider[Provider outage]
Scenario --> Calibration[Calibration rollback]
Scenario --> Evidence[Evidence corruption]
Scenario --> Compiler[Compiler regression]
Scenario --> Decoder[Decoder backlog]
Scenario --> Cost[Cost runaway]
Scenario --> Claim[External claim challenge]The best scenarios are specific. “QPU down” is too vague. “Backend A reports available status, but all chemistry workloads show a sudden readout-bias shift after a compiler target refresh” is useful.
Exercise design packet
Each exercise should have a design packet.
View diagram source
flowchart LR
Objectives[Objectives] --> Injects[Injects]
Injects --> Participants[Participants]
Participants --> Constraints[Constraints]
Constraints --> Metrics[Metrics]
Metrics --> AfterAction[After-action report]Minimum packet fields:
- objective
- scenario narrative
- systems in scope
- excluded systems
- participant roles
- inject timeline
- expected decisions
- stop conditions
- evidence to collect
- after-action owner
Metrics for drills
| Metric | Why it matters |
|---|---|
| time to detection | shows whether monitoring sees the fault |
| time to classification | shows whether operators understand the failure domain |
| time to containment | shows whether blast radius is bounded |
| evidence completeness | shows whether decisions can be reconstructed |
| runbook ambiguity count | shows documentation debt |
| rollback success rate | shows whether recovery is executable |
View diagram source
sequenceDiagram
participant Inject as Inject controller
participant Ops as Operators
participant Monitor as Monitoring
participant Owner as Service owner
participant Review as After-action review
Inject->>Monitor: trigger controlled fault
Monitor->>Ops: alert
Ops->>Owner: classify and escalate
Owner-->>Ops: approve containment
Ops->>Review: record timeline and evidenceSafety boundaries
Quantum game days should begin in sandbox environments. Production drills require explicit risk acceptance. Physical-lab injects need additional controls for cryogenic, laser, vacuum, RF, and electrical safety.
View diagram source
stateDiagram-v2
[*] --> Sandbox
Sandbox --> Staging: passed
Staging --> ProductionLimited: risk accepted
ProductionLimited --> ProductionBroad: repeated success
ProductionLimited --> Sandbox: unsafe findingAfter-action discipline
An exercise is incomplete until lessons are converted into changes.
View diagram source
flowchart LR
Finding[Finding] --> Owner[Owner]
Owner --> Fix[Fix]
Fix --> Verify[Verification]
Verify --> Close[Closure]
Close --> Retest[Retest later]Avoid blame. The point is to expose system weaknesses, not individual heroics.
Chapter close
Resilience training turns quantum operations into muscle memory. The platform will fail in unfamiliar ways; operators should not respond with unfamiliar procedures.