I3INSIT3
METHOD SPEC / V0.3
TRANSPARENT MODEL / PRODUCT POLICY

Risk methodology.

INSIT3 separates observations, forecasts, historical reports, climate signals, confidence, and missing evidence. It does not present one data class as another.

01 / HAZARD SCORE

Weighted base with threat preservation

B = 0.40A + 0.30F + 0.20H + 0.10T
S = max(B, category_floor(A), category_floor(F))

Each component is normalized to a 0–100 decision index. It is not an event probability. The category floor prevents a current or forecast threat from being diluted by quieter historical or climate components. This is INSIT3 product policy, not an agency standard.

02 / CATEGORIES

Decision bands

ScoreLabel
0–39.9Low
40–79.9Moderate
80–100High

These labels are not equivalent to NWS alert labels. Official Common Alerting Protocol severity, certainty, urgency, and instructions remain visible as issued.

03 / MISSING DATA

Missing is never zero

Unavailable, stale, geographically uncovered, out-of-season, or scientifically unsupported evidence produces INSUFFICIENT_DATA. Remaining weights are not renormalized. The primary interface publishes one confirmed score and exact evidence coverage. A technical worst-case bound is retained inside expandable component evidence; it is not an expected range, probability, or forecast.

04 / TIME HORIZONS

Four non-interchangeable layers

ACTIVE / 40%
Current official alerts and observations valid for the evaluated point.
FORECAST / 30%
Forecasts valid within 14 days. Days 8–14 are clearly labeled probabilistic or model guidance where no deterministic official product exists.
HISTORY / 20%
Latest ten complete calendar years, currently 2016–2025, using NOAA Storm Events county reports. Reports are not unique storms and are affected by reporting practices.
CLIMATE / 10%
The latest complete ten-year county mean compared with NOAA’s 1991–2020 normal series. Event hazards without defensible projections remain unscored.
05 / CONFIDENCE

Evidence quality, not event probability

The 0–100 evidence quality index records authority, coverage, validity, published uncertainty, and agreement. It does not mean “chance the event occurs.” Exact layer availability is reported separately. Every component retains its source URL, retrieval time, valid time where available, raw evidence class, transformation, and calibration version.

06 / GEOGRAPHY COVERAGE

Current boundaries without invented history

The canonical directory uses current county and county-equivalent names. NOAA Storm Events history is joined only where the reporting geography supports a defensible match. Alaska boroughs and census areas retain an unscored history component and may show FEMA National Risk Index county-equivalent values as supplemental evidence. Connecticut planning regions retain the official Census town crosswalk to legacy counties, but old county totals are not reassigned because that would imply unsupported within-county precision.

OPEN LOCATION AND HAZARD COVERAGE →

07 / SAFETY BOUNDARY

Official instructions stay official

Evacuation and shelter instructions are reproduced verbatim only when provided by the issuing authority, with attribution and timestamps. Generated material is labeled “Preparedness and operational planning” and never invents, paraphrases, overrides, or contradicts an official directive.

08 / PRIMARY REFERENCES

Standards and documentation