Transparent by designEvery score can be questioned.
Hatua separates observed climate signals, prototype exposure profiles, decision rules and AI-assisted language. Nothing is hidden behind a black box.
Open Mission ControlImpact methodologyFrom hazard to consequence
HazardRainfall deficit 50% · soil moisture 30% · evapotranspiration stress 20%
ExposurePastoral population 40% · livestock 35% · critical water points 25%
VulnerabilityWater distance 35% · vegetation 30% · food/market stress 20% · capacity gap 15%
Impact risk45% hazard + 30% exposure + 25% vulnerability
Action suitability40% urgency + 25% feasibility + 20% coverage + 15% timing fit
All inputs are normalized to 0–100. Thresholds: Low below 35, Watch 35–54, Alert 55–74, Emergency 75 and above.
AI guardrailsAI drafts. Humans decide.
- Retrieve: deterministic rules select eligible actions.
- Constrain: the model can use only approved catalog IDs.
- Validate: every response must pass a strict schema.
- Fallback: timeouts or invalid output use a reliable template.
- Approve: an officer releases the plan and alert.
Evidence register
Sources and attribution
Retrieved 29 July 2026
Responsible use
Prototype limitations
- Hatua does not replace ICPAC, national meteorological services, NDMA or county emergency protocols.
- Sub-county exposure and vulnerability profiles are demonstration estimates, not validated operational classifications.
- Open-Meteo data is model output and must be interpreted with local observations.
- Swahili messages are drafts and require validation by a local language specialist.
- Impact totals represent completed prototype actions, not independently evaluated real-world outcomes.