This appendix provides worksheets for resilience drills and energy-aware operations.
View diagram source
flowchart LR
Worksheet[Worksheet] --> Run[Run exercise]
Run --> Review[Review]
Review --> Backlog[Create backlog items]Chaos drill worksheet
chaos_drill:
id: qdrill-YYYYMMDD-NNN
hypothesis: "If the preferred target is unavailable, jobs fail safe and no invalid claim is produced."
owner: platform@example.com
scope:
tenants: [research]
targets: [simulator, qpu-a]
max_jobs_affected: 5
injection:
method: provider_adapter_timeout
start_time: <timestamp>
duration_minutes: 20
abort_conditions:
- safety_alarm
- unbounded cost growth
- production tenant affected
expected_behavior:
- broker marks target degraded
- policy blocks hardware-only claims
- fallback routes simulator-eligible jobs
metrics:
- time_to_detect
- time_to_safe_state
- invalid_jobs_admitted
- fallback_success_rate
review:
scheduled_at: <timestamp>
reviewer: sre@example.comView diagram source
flowchart TB
Hypothesis[Hypothesis] --> Injection[Injection]
Injection --> Abort[Abort conditions]
Injection --> Metrics[Metrics]
Metrics --> Review[Review]
Abort --> ReviewCalibration staleness drill
calibration_staleness_drill:
fixture:
calibration_age_minutes: 180
workload_sensitivity: high
expected_policy_decision: deny
expected_reason_code: calibration_stale
evidence_required:
- blocked_request_record
- policy_decision_record
- alert_recordResult-ingest failure drill
result_ingest_failure_drill:
injected_fault: metadata_store_unavailable
expected_behavior:
- raw result retained in staging
- publication gate blocks claim promotion
- retry queue created
- alert emittedView diagram source
flowchart LR
Result[Result] --> Ingest{Ingest available?}
Ingest -- yes --> Warehouse[Warehouse]
Ingest -- no --> Staging[Staging]
Staging --> Retry[Retry queue]
Retry --> WarehouseSustainability worksheet
sustainability_record:
workload_id: qexp-YYYYMMDD-NNN
useful_work_unit: valid_shot
submitted_jobs: 12
accepted_jobs: 8
valid_shots: 640000
invalid_run_count: 2
classical_compute:
cpu_hours: 12.4
gpu_hours: 3.1
storage:
raw_gb: 18
retained_gb: 9
facility_or_provider_estimate:
energy_kwh: null
carbon_kgco2e: null
notes:
- provider-specific facility data unavailable
- report energy proxy separately from measured energyView diagram source
flowchart TB
Sustainability[Sustainability record] --> Useful[Useful work unit]
Sustainability --> Compute[Classical compute]
Sustainability --> Storage[Storage]
Sustainability --> Waste[Invalid runs]
Sustainability --> Estimate[Energy or carbon estimate]Invalid-work reduction plan
invalid_work_reduction:
observation: invalid_run_rate_above_threshold
suspected_causes:
- weak simulator validation
- stale target capability records
- missing admission policy
actions:
- add simulator smoke test
- refresh capability catalog hourly
- block unsigned QPU jobs
success_metric: invalid_run_rate
target: "< 2 percent over 30 days"Energy-aware scheduling review
energy_aware_scheduling_review:
workload_class: batch_exploratory
latency_sensitivity: low
calibration_sensitivity: medium
allowed_windows:
- 2026-04-19T08:00:00Z/2026-04-19T12:00:00Z
optimization_objectives:
- minimize_queue_delay
- minimize_cost
- avoid_stale_calibration
not_allowed:
- changing claim semantics
- using unapproved fallback targetView diagram source
flowchart LR
Workload[Batch workload] --> Objectives[Objectives]
Objectives --> Window[Allowed window]
Window --> Check{Claim semantics preserved?}
Check -- yes --> Schedule[Schedule]
Check -- no --> Review[Human review]