This is the proposed first StormTrust feature release contract. It governs a weather-exposure briefing and reviewed canvassing areas before parcel matching. No production report, accepted boundary release, score, delivery schedule, or statewide source coverage is claimed here.
| Decision | First release |
|---|---|
| Audience | A supervisor and authorized canvassers in one tenant. |
| Geography | An explicitly selected Missouri pilot scope; preserve county GEOIDs and out-of-scope or unassessed areas. |
| Trigger | An eligible warning or an operator-selected radar evidence window. Neither requires a previously stored forecast. |
| Deliverable | A reviewed web report in a dashboard, an email link to its revision, and a mobile recipient view. |
| Priority path | Missouri → counties → reviewed canvassing areas. |
| Ranking | Weather-exposure priority under one approved rule and comparable evidence coverage. |
| Deferred | Parcel/roof joins, replacement probabilities, national rollout, full physical-storm tracking, PDF delivery, and automated route optimization. |
The existing hail-alert application is limited to its configured market and eligibility policy. A Missouri-wide warning intake requires an explicit change; that application is not a statewide report trigger merely because it fetches Missouri warnings. See Storm Alert.
| Input | Evidence required before use |
|---|---|
| Warning context, when used | Publisher identity, native warning/chain ID and revision, valid period, geometry, status, and retained instructions. |
| MRMS evidence | Native product, source object identities, observed/retrieved times, units, valid-data coverage, quality checks, and replayable analysis inputs. |
| County geography | Accepted geometry release and CRS, GEOIDs, source/vintage, validation, and spatial intersection method. |
| Canvassing areas | Source or reviewed drawn/grid geometry, stable ID, version, county intersections, name provenance, reviewer, and service-area fit. |
| Ranking rule | Version, hail threshold and unit, evidence window, area calculation, missing-data policy, tie-breaker, and acceptance examples. |
| Delivery | Authorized tenant audience, supervisor approval, channel configuration, and an immutable report revision. |
Census TIGER/Line ↗ is a county-boundary source candidate. Its geographic codes support joins; the files do not themselves supply demographic or roof attributes. The existing Missouri identity registry is not a substitute for an accepted polygon release. Neighborhood and residential context need their own evidence.
A proposed baseline ranks assessed counties by unique land area with qualifying MESH evidence during the selected window, descending. Within a county, rank reviewed canvassing areas by the same area measure. Show the exposed fraction of each area and the hail estimate distribution as supporting context. Do not rank a large county using only its single maximum pixel or imply that land area measures the number of roofs.
Before enabling the rule, approve the threshold, quality/completeness limits, land mask, equal-area calculation, and temporal aggregation. Form the union of qualifying spatial cells across the window; repeated frames must not count the same area repeatedly. Retain the contributing observations and aggregation version. Radar no-data cells remain unknown, not zero hail.
Compare only results meeting the same coverage policy. Show partial or unassessed counties separately with their observed coverage. Use stable area IDs to resolve ties. County-clipped work areas must not double-count overlap in parent totals: calculate the parent union independently and flag overlapping assignments. Human selection for accessibility or residential relevance is a separate, attributed decision, not a hidden score adjustment.
This method is a proposal, not a calibrated damage model. It differs from the later parcel and roof measures in the Impact ranking contract. Without accepted neighborhood or work-area geometry, do not generate neighborhood rankings. Without residential data or reviewed context, do not call a grid area a residential neighborhood.
A report may use a bounded analysis scope before physical-storm tracking exists: selected geography, warning IDs if relevant, observation window, and source revisions. Several physical storms can occur in that scope. Name it a regional hail briefing rather than falsely asserting one tracked storm.
| Record group | Required fields |
|---|---|
| Identity | Report ID, immutable revision, tenant/audience, status, author and reviewer, created/published times. |
| Analysis | Analysis-scope ID, warning references, geographic extent, observation window, physical-storm ID only when established. |
| Evidence | Source objects/revisions, products and units, observation cutoff, retrieval times, valid coverage and caveats. |
| Geography | County and work-area IDs, geometry releases, clipping/overlap policy, reviewed starting markers where provided. |
| Priority | Mode, rule revision, units, ordered results, supporting measures, unknowns, and human annotations. |
| Reproduction | Evidence snapshot or retained derived inputs, geometry/rule revisions, validation result, retention policy. |
| Distribution | Approval, recipient scope, delivery identity/status, supersession link, and recipient acknowledgment as a separate event. |
A published revision is immutable. Corrections and late radar data create a new revision. A “latest” link can resolve to a newer report, but the email's revision link must identify what was sent and disclose supersession or withdrawal.
Report evidence retention must support its promised replay period. The existing MRMS raw/derived retention policies do not automatically preserve a report's inputs indefinitely. Retain an appropriate evidence snapshot or explicitly bound reproducibility; an expired upstream object cannot be reconstructed from an attractive screenshot.
The pilot integrates Storm History for reopening its saved scope and report revision, and Storm Replay for inspecting the retained observations behind the published result. The shared acceptance contract also governs test reproduction. Broad historical search and full as-known-then reconstruction are not prerequisites for this bounded report slice.
Draft → Validated → Reviewed → Published
↘ Needs revision
Published → Superseded or WithdrawnGeneration, publication, delivery, and viewing have separate states. Delivery uses queued → accepted by provider → delivered or failed where evidence exists. A provider's acceptance is not proof the recipient read the report. Record viewing only after authorized access, and acknowledgment only after an explicit recipient action.
Use an idempotency identity containing report ID/revision, recipient, and channel. Retries reuse it; an intentional resend is a separately authorized action. Persist checkpoints so generation retries do not resend messages. Failed publication never enqueues delivery. Delivery failure does not unpublish a valid report; retry only the failed delivery.
Enforce tenant and recipient access on the dashboard, report, evidence links, and future exports. The first release uses authenticated recipients; public share links require a separate access design. A frozen warning snapshot can coexist with current official-warning information labeled with its own check time. Keep field assignment on hold when current safety information is missing or conditions remain dangerous.
| Scenario | Required evidence |
|---|---|
| Generate one real pilot briefing | Source-backed warning/window, accepted county geometry, assessed exposure, and at least one reviewed work area. |
| Review priorities | The supervisor can reproduce ranks, inspect source times, distinguish no-data from zero, and explain area choices. |
| Publish and send | Approval is recorded; one authorized recipient receives a link to the exact published revision. |
| Open on a phone | The recipient can read the summary, choose a county and work area, open its map, and acknowledge the assignment. |
| Active warning or unsafe conditions | Planning remains available; the report does not instruct immediate canvassing in dangerous conditions. |
| Missing area coverage | County evidence remains visible; no invented neighborhood or complete field-assignment claim. |
| Reopen report evidence | History finds the retained pilot entry and Replay opens its pinned evidence without changing the report or sending another message. |
| Correct or retry | A new revision retains the old evidence; retry does not duplicate delivery; withdrawn access is enforced. |
| Inspect access boundaries | Another tenant cannot retrieve the report or its linked private evidence. |
| Retain learning | A field observation references report revision and work area without automatically becoming model ground truth. |
This pilot can ship before parcel joins and before Storm Forecast ingestion. The required geographic analysis, report persistence, dashboard, permissioned delivery, mobile view, and recovery behavior still need implementation and runtime proof. Follow the journey readiness contract for the sequence and the repository's existing release procedure for deployment authority.