Skip to content

Latest commit

 

History

History
27 lines (21 loc) · 1.84 KB

methods.md

File metadata and controls

27 lines (21 loc) · 1.84 KB

Methods

The analysis pipeline is deterministic:

  1. A map click or address search creates one selected WGS84 point.
  2. Optional Nominatim reverse geocoding enriches the label; failure does not block analysis.
  3. XL modules derive city/district, demographic, housing, rent, and FUA-ready context from local structured data.
  4. L modules use a configurable radius buffer for green/open, land-use, transit, mobility, POI, and development-hint indicators.
  5. M modules generate a nearest-street fallback segment, corridor, building/tree/sun overlays, and an adaptive SVG cross-section.
  6. Overpass query modules are deterministic, cache-keyed, and stored in export provenance. When disabled or unavailable, local fallback indicators remain visible with caveats.
  7. Every source check is normalized into analysis.provenance.dataSourceRun with phase, status, counts, timing, caveats, and errors.
  8. Exports are generated from the structured AnalysisResult object.

The MVP uses deterministic fallback overlays where authoritative local geometry has not yet been preprocessed. Confidence and caveat fields expose this explicitly.

Data-source run statuses

The UI and exports use the same normalized source-run vocabulary:

  • requested: a source step was queued or started.
  • fetched: a live source or local coverage source returned usable data for the selected point.
  • cache-hit: a point cache or broad coverage cache covers the selected point.
  • empty: the source was available but returned no usable point/radius coverage.
  • missing-credentials: the source was requested but required local credentials were not available.
  • missing: the required local/preprocessed artifact was not present.
  • failed: the source request or parser failed.
  • skipped: the source was intentionally not applicable, for example an optional FUA-only layer outside FUA coverage.