Original validation report · July 2026
USDA address-checker reliability report
A transparent regression test of the independent location-screening pipeline across 50 public U.S. addresses.
What was tested
The suite submits public, non-sensitive street addresses through the same production workflow a visitor uses. It checks geocoding, coordinate handling, USDA map-layer response, result classification, source attribution, and failure behavior. Cases are intentionally distributed across multiple states and expected map outcomes.
| Test group | Cases |
|---|---|
| result | 30 |
| no_match | 18 |
| upstream_error | 2 |
Pass criteria
- The address resolves through the configured public geocoder.
- The resulting coordinates are accepted by the current USDA geographic layer.
- The returned screen matches the case expectation.
- The response includes a timestamp, source-layer identity, and official-verification route.
- No case is represented as an underwriting or approval decision.
Reproducibility
The automated suite is maintained in the project as tools/run-checker-regression.mjs. A scheduled workflow reruns it against production and retains the machine-readable report. Tests are also run before material releases.
Download the machine-readable results · Try the checker
Known limitations
- The suite is a bounded sample, not every U.S. address or boundary condition.
- Geocoders can place new, rural, unit, route, and ambiguous addresses incorrectly.
- USDA layers and implementation can change after a test run.
- A geographic screen does not assess income, borrower, property, appraisal, lender-overlay, or funding requirements.
- Near-boundary properties require direct confirmation in USDA’s official system.
Governance
A failed scheduled run is retained and escalated; results are not silently removed. The report is refreshed after material checker changes. Corrections can be submitted through the corrections process.