# Impact ranking contract This is a proposed Storm Impact contract. Geographic rankings and a calibrated roof-replacement model are not implemented by these docs. Unknown source coverage and model outputs remain unknown. ## Release paths The first report uses **Missouri → counties → reviewed canvassing areas**, with weather-exposure area as its unit. The [Storm Report release contract](/stormtrust/product/platform-and-data/intelligence/report-contract/#weather-exposure-ranking) owns that mode's inputs, area calculation, coverage, and overlap rules. Parcel and roof data are not prerequisites for it. All modes still require accepted geometry and evidence appropriate to their claims. ## Later property path | Level | Scope and default ranking | Selection result | | ---------------------- | --------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------- | | State | All states intersecting the selected storm footprint; rank evaluated states by the same declared measure. | Counties in the chosen state and storm scope. | | County or equivalent | County intersections within the selected state. | ZIP/ZCTA portions within that county. | | County × ZIP/ZCTA area | Only the selected county's portion of the area. | Parcels assigned to that portion under the versioned membership rule. | | Parcel | Unique source-qualified parcel identities within the selected scope. | Evidence dossier and any linked residential roofs. | At every level, expose the entire affected scope, including unevaluated areas. Rank comparable evaluated results, and show unknown coverage separately without assigning it a zero or pretending the displayed ranking is complete. Define the affected footprint from the selected storm's evidence and versioned exposure rule. Neither the full national raster bounds nor the warning polygon alone establishes that every enclosed state or parcel experienced hail. While that footprint is unresolved, show warning scope separately from assessed impact. In the property workflow, go directly from ZIP/ZCTA area to parcels. Neighborhoods are an optional overlay or saved territory, subject to boundary provenance and overlap rules. This later hierarchy does not replace the report pilot's county-to-work-area path. ## ZIP codes and county intersections [Census defines ZCTAs](https://www.census.gov/programs-surveys/geography/guidance/geo-areas/zctas.html) as generalized areas representing ZIP-code distribution. Not every ZIP code has a ZCTA. Display **ZIP/ZCTA area** or **ZCTA** when using that dataset; retain postal ZIP addresses as a separate attribute. Do not assume geographic nesting. Use versioned boundaries and the applicable [Census relationship files](https://www.census.gov/geographies/reference-files/time-series/geo/relationship-files.2020.html) to establish intersections. Key a county portion by state GEOID, county GEOID, ZCTA, and boundary vintage. The same ZCTA can appear under multiple counties, but each row represents only its own portion. Parcel intersections describe exposure geometry. Additive counts need a single, versioned allocation rule for each modeled unit. Preserve cross-boundary flags; do not count an entire parcel or roof once in every polygon it intersects. Unassigned or ambiguous units need a visible remainder bucket, so totals remain reconcilable instead of dropping records silently. ## Unit of analysis The later property UI drills down to parcels. A parcel is not necessarily one residential building or one roof. Roof-level replacement predictions require stable roof or structure identity and a reviewed relationship to parcels. In an initial parcel-based exposure model, label counts **exposed residential parcels**, never roofs. If residential classification is unknown, report that coverage gap. When a roof model exists, aggregate unique modeled residential roofs, including explicit treatment of multi-building parcels and structures linked to multiple parcels. Missouri source selection, native identifier meanings, and data availability remain governed by [parcel source review](/stormtrust/engineering/operations/review-parcel-source/). A map search result or a county name cannot substitute for a parcel release. ## Ranking modes | Mode | Permitted interpretation | Prerequisites | | ------------------------- | ------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Weather-exposure priority | Relative investigation priority based on qualifying radar-exposed area. | Accepted county/work-area geometry and the [report ranking contract](/stormtrust/product/platform-and-data/intelligence/report-contract/#weather-exposure-ranking). | | Parcel-exposure priority | Relative investigation priority based on radar-derived exposure and declared property relevance. | Accepted boundaries, parcel/residential data, versioned exposure rules, source quality and completeness checks. | | Replacement priority | Estimated replacement need for modeled residential roofs under a specified outcome definition. | Roof identity/context, permitted outcome labels, calibrated model, validation evidence, geographic applicability, and explicit uncertainty. | A suggested first parcel-exposure measure is the count of unique residential parcels meeting a reviewed exposure rule. Define the hail/severity threshold, observation window, quality requirements, and geometry method before enabling that ranking. Do not invent weights or use an arbitrary score as a replacement probability. For a later validated replacement model, area priority can use **expected roofs needing replacement**: the sum of per-roof replacement probabilities for unique modeled roofs in that scope. This aggregate expectation does not require roofs to be independent; uncertainty intervals do require treatment of shared storm and spatial dependence. Publish evaluated-roof counts and coverage beside it. At the parcel level, show expected affected roofs when roof identities exist, plus the individual roof estimates where useful. Without a roof model, keep the parcel exposure ranking and label. Do not switch units silently between levels. Repair need, replacement need, confirmed inspection findings, insurance decisions, and sales conversion are different labels. A model for one cannot be presented as another. Use the [evaluation contract](/stormtrust/product/platform-and-data/live-revisions-and-historical-evaluation/#probability-evaluation) for storm/time-separated validation and leakage controls. ## Comparable aggregates Rank siblings by total declared impact within their scope, not by the single largest hail pixel or the highest individual probability. Total impact and impact rate answer different questions; offer rates as an explicit secondary measure with their denominator visible. Use one analysis scope (or established physical-storm revision), time window, rule/model version, data-release set, unit definition, and membership version for a comparison. Do not mix exposure scores with replacement probabilities. A partially covered state may have a known covered-scope total, but it cannot be silently compared as a statewide replacement estimate against a fully covered state. Use a stable identifier as the final tie-breaker. Exclude disallowed or failed validation inputs from computation while retaining an explanation and coverage count. A low-confidence result is not automatically zero risk. ## Ranking snapshot fields These are required contract fields, not a published API schema. | Group | Required values | | ---------- | -------------------------------------------------------------------------------------------------- | | Identity | Snapshot ID, analysis-scope ID, physical-storm ID/revision when established, scope, creation time. | | Time | Selected observation time, exposure window, source vintages, evaluation time. | | Method | Mode, measure, unit, model/rule version, thresholds, membership version. | | Data | Source releases, boundary vintage, parcel and roof/structure releases only for modes using them. | | Result | Ordered child identities, values, reasons, denominators, coverage, unknowns, tie-breaker. | | Provenance | Evidence references, validation report, quality limitations, and applicable use decisions. | ## Live updates and replay New observations can change rankings. Preserve the currently selected snapshot while the user inspects a county, parcel, or historical frame; offer an explicit refresh into the latest compatible snapshot. Show when a selected item moves in rank, and preserve the geographic path rather than jumping to the new top row. Historical views must use the selected evidence window and pinned model/data versions. A retrospective rerun with newer parcels or a newer model is a new analysis, clearly distinguished from what was available at the original time. ## Required acceptance examples A county-crossing ZCTA must not duplicate the same modeled unit in state totals. A multi-roof parcel must not become one roof by assumption. A missing county must remain unknown. An unclassified parcel must not become residential by default. A corrected MRMS frame must produce a new snapshot without changing an earlier published result. No example in this reference claims an actual Missouri ranking or a measured replacement probability.