Criora Severity

Every hazard event on Criora carries a rank, a number from 0 to 100 that compares events across hazards: a flood with an earthquake, a storm with a fire. The map’s list of what matters now, searches and answers to AI assistants are all ordered by it. This page says what the rank weighs, how it was checked and where it can mislead.

What the rank weighs

Three kinds of evidence, each counted only where the event has it:

• The official level. The level the issuing authority gave the event, read on one common scale for every issuer. Each issuer’s levels are placed on it by what the issuer says they mean: a CAP severity, a MeteoAlarm or GDACS colour, a PAGER alert, a hurricane category, a volcano’s alert level. A warning that calls its hazard only possible, as a watch does, ranks below a warning of the same level, and one that calls it unlikely adds nothing.
• The people exposed. The residents inside the event’s footprint, counted on GHS-POP 2025, the population grid of the European Commission’s Joint Research Centre. The footprint is the area the issuer drew where it draws one, and otherwise an area drawn from the event itself, such as the ground an earthquake shakes hard enough to damage buildings or the outline of a fire’s satellite detections. Each event says which in footprint and gives the count in people.
• The reading. The event’s intensity against the others of its type: a magnitude among magnitudes, a fire’s radiative power among fires’.

A fire has no official level, and an event whose footprint nobody lives in has no people: what an event lacks adds nothing. The official level is always returned beside the rank, never replaced by it.

How far it can be relied on

How much each kind of evidence counts was not set by hand. It was fitted to real past situations, moments when several events were active at once, ordered by what happened afterwards: deaths, people displaced and emergencies declared.

• Ordered by Criora Severity, 78.5% of pairs of events come out in the order their outcomes put them.
• On events held out of the fit, 77%.
• The ranking Criora used before ordered 30%.

These situations are part of Criora’s test suite, so a change that orders them worse cannot be released.

Limits

• A reading’s place within its own type says little across hazards, since a strong reading where nobody lives is common. It helps order events of one type.
• The people counted are residents. Visitors, commuters and workers are not in the grid.
• An earthquake’s footprint is estimated from its magnitude and depth, not from the shaking measured on the ground.
• Every change to the method is a new methodology version with its entry in the changelog.