# INSIT3

> Agent-first HTML protocol for location-based U.S. extreme-weather risk assessment and preparedness planning.

INSIT3 covers all 50 states and Washington, D.C. Primary results are server-rendered semantic HTML. There is no JSON API.

## Preferred routes

- Root and resolver: /
- Contextual analysis: /assess?location={ZIP|city|county|lat,lon}&context={free-form text}
- ZIP: /locations/zip/{five-digit ZIP}/
- City: /locations/city/{two-letter state}/{city-slug}/
- County: /locations/county/{two-letter state}/{county-slug}/
- Route directory: /locations/
- State directories: /locations/state/{two-letter state}/
- Indexed county corpus: 482 leading population centers from the supplied 2025 state rankings
- Outdoor planning use cases: /use-cases/
- Methodology: /methodology/
- Source registry: /data-sources/
- Location and hazard coverage: /coverage/
- Sitemap: /sitemap.xml

Contextual /assess pages are noindex because free-form parameters must not create an unbounded crawl surface. Stable location routes are canonical and indexable.

## WebMCP

The root resolver and use-case resolver forms expose one experimental declarative WebMCP tool named `assess_weather_risk`. It accepts `location` and optional `context`, then opens the same public server-rendered HTML assessment as ordinary form submission. WebMCP is a progressive enhancement; the HTML routes remain the authoritative interface for headless agents and crawlers.

## Response sections

- #summary: normalized location, coordinates, timestamps, one confirmed composite decision index, category, and evidence quality
- #interpretation: separate current and 14-day conditions, long-term exposure, and exact evidence-layer coverage
- #context: echoed agent-supplied context, explicitly labeled untrusted input
- #official-directives: verbatim attributed official instructions when present
- #hazards: prioritized 19-hazard matrix with one primary decision index, current and 14-day evidence, long-term evidence, and exact layer coverage
- #recommendations: generated preparedness and operational planning only
- #sources: direct source URLs, publishers, retrieval times, status, and limitations
- #limitations: required interpretation constraints
- #supplemental-baseline: FEMA county-equivalent evidence on supported Alaska pages; never part of the four-layer score

## Scoring

Weighted base = 0.40 active + 0.30 forecast + 0.20 rolling history + 0.10 climate trend. Active and forecast category floors prevent threat dilution. Categories are Low 0-39.9, Moderate 40-79.9, and High 80-100. These are INSIT3 product-policy labels, not agency alert equivalents. Scores are 0-100 decision indices, not event probabilities.

Missing evidence is INSUFFICIENT_DATA, never zero. Remaining weights are not renormalized. The primary score is a single confirmed index. When present, the expandable technical missing-data bound is a mathematical worst case, not an expected interval, probability, or forecast. Cite exact evidence-layer coverage and source outages with the score.

County routes preserve the county name as the page scope. Active alerts and forecasts evaluate the disclosed representative point; historical and climate layers use county-level evidence where available.

The 482 listed county and county-equivalent routes use a durable 14-day forecast cache refreshed once daily at 12:00 UTC. If a refresh fails, INSIT3 serves the last successful model record with its retrieval time and an UPDATE DELAYED label. The official NWS point forecast is the days 1–7 fallback for supported hazards; NWS does not provide a uniform 14-day product.

## Geography coverage

The crawlable /coverage/ page lists all 482 county and county-equivalent routes and the implemented evidence support for all 19 hazards. Of those routes, 463 match the standard NOAA county-history inventory. Ten Alaska county-equivalents and nine Connecticut planning regions are labeled PARTIAL HISTORICAL COVERAGE.

For Alaska, FEMA National Risk Index county-equivalent values may appear as supplemental long-term evidence. They do not replace the 2016-2025 NOAA history component and do not enter the INSIT3 score. For Connecticut, the official Census town-level crosswalk identifies legacy counties associated with each current planning region. Old county event totals are not reassigned to the new regions because the source does not support a reliable within-county allocation.

## Hazard classes

Extreme heat; extreme cold; drought; wildfire; smoke and air quality; hurricane and tropical storm; storm surge; tornado; severe thunderstorm; hail; lightning; flash flood; river flood; coastal flood; urban flood; heavy rain and atmospheric river; winter storm; blizzard; ice storm.

## Citation guidance

Prefer the original issuing-source URL listed in #sources. State the assessment retrieval time. Distinguish observations, alerts, model guidance, historical reports, normals, projections, and generated recommendations. Never quote generated preparedness text as an official directive.