Skip to content

METHOD

How scores are built

THREATGRID’s global level is an independent assessment of the incidents it has actually stored. It is not a CISA, DHS, or other official alert. If a sync has never succeeded and the database is empty, the level is INSUFFICIENT DATA.

Incident severity

Classification and severity are separate. A resolved minor outage can stay a service outage while its live score drops. Points come from provider impact or incident type, geographic reach, whether the case is still open, and whether a primary source is attached. Low-confidence secondary reports are capped: under 45 they cannot rise above Monitoring, and under 70 they cannot rise above Elevated.

Severity bands
Minimum scoreSeverityMeaning
72CRITICALSevere, current disruption or a confirmed major incident.
48HIGHMajor disruption to important services or a widely exploited product.
30ELEVATEDMeaningful disruption or a vulnerability that deserves prompt patching.
14MONITORINGDeveloping, minor, or not yet corroborated.
0INFORMATIONALResolved, low impact, or context only.

What is not an attack

  • Provider status pages are service outages unless the provider’s own words describe unauthorized access or a security incident.
  • Power, fire, fiber, and carrier failures are infrastructure disruptions.
  • CISA’s exploited-vulnerability catalog is a patching record, not confirmation that a named company was breached.
  • A single news article cannot produce a critical severity. A report about a classified or national-security system can reach Elevated, or High if more than one outlet describes it. It still cannot reach Critical without a primary source.
  • Three or more elevated incidents in a day, excluding patch advisories, add weight to the global level. That is a cluster, not proof of a shared cause.
  • A sentence that denies an attack (“no indication of unauthorized access”) is not treated as confirmation.

Global threat level

Open incidents, and resolved ones verified in the last 12 hours, contribute the weights below, scaled by confidence. Recently resolved items count at 40 percent. Unconfirmed items count at 60 percent. Vulnerability advisories count at 40 percent and are capped at 12 points combined, so a patch list cannot look like a crisis. Three or more affected services add 6. Two open critical incidents add 8. The score is capped at 100.

  • CRITICAL: 34 points before modifiers
  • HIGH: 18 points before modifiers
  • ELEVATED: 9 points before modifiers
  • MONITORING: 4 points before modifiers
  • INFORMATIONAL: 1 points before modifiers
Threat level bands
MinimumLevelMeaning
61CRITICALSeveral severe, current incidents with broad reach.
37HIGHAt least one major ongoing incident, or several elevated ones.
19ELEVATEDMeaningful incidents are active, without a severe global picture.
8GUARDEDLimited activity worth watching. Not an emergency.
0NORMALNo significant active incidents in the scoring window.

DEFCON index

The stability index is separate from the global threat level. Level 5 is quiet. Level 1 requires verified critical disruptions in three sectors from primary sources. News alone cannot raise it. If no source has synced and the database is empty, it reads Insufficient Data. History is written on sync when the level changes or 12 hours have passed.

Maps

A coordinate is stored when a provider names a known place, or when a grid story names a state as the affected area. A street address is never invented. The marker is the affected area, not an attack origin.

What makes the feed

Cyber reports still have to name an attack, an outage, or a vulnerability. Power, banking, and carrier stories also have to be large: several states, a named storm system, or language like nationwide or hundreds of thousands. Separately, the utility tally files a state only at 100,000 customers without power, or 20,000 customers concentrated in three counties or fewer. When the count falls back under that bar, the record is marked resolved on the next successful check. One failed source does not clear the others.

AI

Briefings are generated only when someone asks, from stored excerpts, and cached. The model is instructed not to invent facts and not to declare an emergency. The global level never comes from the model.