A priority index with an explicit version
The Calamity Score is our heuristic event-priority index. It does not estimate casualties, damage or an occurrence probability. Equal numbers across different hazards do not imply equal human consequences.
The current physical model is version 5. Archived version 4 records retain their historical identity. News-only records use the separate news attention model, version 1001. A record without a stored version is identified as incomplete.
Four physical components
CS = round(0.40 × intensity + 0.30 × normalized urban PEI + 0.20 × cascade scenario weight + 0.10 × reference or batch anomaly), bounded to 0–100.
Intensity applies hazard-specific transfer functions to measurements with their declared units. Fire active-pixel counts are not burned hectares. Air-quality measurements retain their averaging window; an instantaneous PM2.5 proxy is not an official daily AQI.
The urban population term is a distance-weighted index over the covered city dataset. It is not a count of unique or affected people. Cascade terms describe relative attention to rule-generated secondary-hazard scenarios; their weights are not calibrated probabilities.
The reference term uses a versioned historical earthquake reference where there are enough eligible observations: at least 30 USGS episodes and 14 reporting days in the closed 90-day window. Original archived magnitudes are transformed to the current intensity scale. Other coverage uses an explicitly identified batch-anomaly fallback; a local historical baseline is not assumed everywhere.
Synthetic arithmetic example: intensity 40, urban term 25, cascade term 20 and reference term 10 give round(16 + 7.5 + 4 + 1) = 29. This checks arithmetic, not predictive skill.
News is a separate index
Version 1001 uses article counts, source labels and languages, with a cap of 80. Four articles, two source labels and one language give 12 + 10 + 5 = 27. This is media attention, not physical severity. Source labels do not establish independent confirmation.
Evidence and change
Source support, field completeness and freshness are separate evidence dimensions. The legacy confidence scalar mixes assigned source weights and data availability; it is not accuracy or a statistical probability. The score can change when observations, coverage or model versions change.
For current CFI and country activity ratings, computation date and validity must be recorded. An expired or absent assessment is not current low risk. Event revision timelines show retained material changes; earlier revisions can be unavailable.
The [scientific model cards](/methodology/models) provide versioned formulas, coverage, examples and limitations. Software checks do not establish predictive validation. Official emergency guidance remains with the responsible public authorities.