Method
Four tiers
Risk here is a function of two things: how directly the component's output affects a person's care or safety, and how much human judgement sits between the output and the consequence. A highly accurate model with no human review can be higher risk than a rough one that only informs a planning meeting.
| Tier | Definition | Controls applied |
|---|---|---|
| Tier 1 — Critical | Output could contribute to a decision affecting an individual's immediate survival or clinical outcome. | Mandatory human confirmation before any effect; BODH benchmarking before deployment; under-triage rate published; independent clinical review of model changes; kill-switch tested quarterly. |
| Tier 2 — High | Output influences commitment of scarce resources between competing recipients. | Human authorisation with mandatory reason capture; equity analysis across population segments; explicit and visible priority rules; full audit trail. |
| Tier 3 — Moderate | Output shapes planning, posture and pre-positioning but commits nothing. | Confidence intervals published; backtested accuracy disclosed; human review before operational adoption. |
| Tier 4 — Low | Output is descriptive, aggregate or administrative. | Standard logging and periodic accuracy review. |
Applied
Component classification
| Component | Reference | Tier |
|---|---|---|
| Clinical triage support | BAEMS BM-01 | Tier 1 — Critical |
| Air ambulance demand forecasting | AA-05 / MOD-5.2 | Tier 2 — High |
| Resource allocation optimisation | MOD-3.2 | Tier 2 — High |
| Aid verification risk tiering | MOD-4.2 | Tier 2 — High |
| Need forecasting | MOD-5.2 | Tier 3 — Moderate |
| Disaster progression simulation | MOD-5.3 | Tier 3 — Moderate |
| Trend and cluster analysis | MOD-5.1 | Tier 3 — Moderate |
| ML routing optimisation (2027) | Roadmap | Tier 2 — High |
| Reporting and briefing generation | MOD-1.4 | Tier 4 — Low |
| Documentation and translation support | MOD-8.3 | Tier 4 — Low |