Service recovery and evidence exercise
An interruption leaves a civilian team working locally while reports arrive from several partners. Check how long a backup can run, what was actually recovered, and which reports come from independent origins.
How long can the local service run, and when did recovery occur?
In the backup exercise, stored energy is 12 kWh DC, the reserve is 2 kWh DC and conversion efficiency is 80%. The inverter can provide up to 4 kW. The load is 3 kW for four hours. The fixture assumes no extra auxiliary draw.
Usable AC energy = (12 − 2) × 0.80 = 8 kWh AC. Time to first shortfall = 8 ÷ 3 = about 2 hours 40 minutes.
In a separate recovery exercise, the team verifies restored service 220 minutes after the interruption. The recovery target is 240 minutes. The newest consistent recovered record predates the interruption by 80 minutes, exceeding the 60-minute target for data loss.
What the results mean. Backup runtime, time to service restoration and age of recovered data answer different questions. Verify the load and usable capacity in a local drill. Test that essential workflows and records are available when the shared service is withdrawn.
How many independent observations support the reports?
The evidence fixture includes 5 reports, some derived from the same original observation. Following the declared source links identifies 2 traceable origin clusters and 1 report with unresolved origin.
A reviewer can compare the two traceable origins and ask for the missing chain of custody. The number of origins does not establish that the reports are accurate, that they describe the same event, or who caused the interruption.
What changes under the three arrangements?
In this comparison, each arrangement sees the same five reports at the same time and has the same staff and review budget. The named procedure changes how the work is organized.
| Measure | Fragmented arrangements | Structured non-AI cooperation | MAR: AI-assisted cooperation |
|---|---|---|---|
| How teams handle information | Each team checks copies it receives and contacts the originating agency if it finds a discrepancy. | Teams share an approved origin register and designated human reviewers resolve mismatches. | AI can draft links and flag possible duplicates. People verify links against source records and keep disputed items open. |
| Known traceable origin clusters | 2 origin clusters | 2 origin clusters | 2 origin clusters |
| Reports with unknown origin | 1 reports | 1 reports | 1 reports |
In a separate timed fixture with 100 fictional records, manual review uses 2 staff-hours and AI drafting, review and correction use 1.5. The audited error rates are 1% and 3%, respectively. A team can use those two measures to ask whether the time difference is worth the added errors. The fixture supplies those measurements and establishes no expected AI effect.
Reproduce the review time and error comparison Download fictional inputs
What would a fair test record?
Write down the allowed records and disclosure rules, verified source origins, time of acknowledgment, false and missed matches, local fallback test results, and human decision gates. Compare both cooperating arrangements against a fragmented baseline, then compare AI assistance directly against structured human review. Preserve a manual path when the shared channel or AI tool is unavailable.
These arithmetic and evidence exercises cannot establish diplomatic intent or show whether an intervention would prevent escalation. Read the AI resilience testing and monitoring guide