Neighborhood Amenities (v1)
Version 1.0.0 · Published 2026-06-10
What this is
The Neighborhood Amenities subscore measures how walkable and accessible everyday services are from an address. Per the 2026-06-09 methodology triangulation (Alternative 2 pick), we use continuous kernel density estimation (KDE) per POI kind instead of arbitrary distance bands. The output is a 0-100 composite + a tier label in {high, medium, low}.
POI kinds + half-life calibration
Each POI kind has a half-life (in miles) representing the distance at which the contribution to the composite score drops to 50% of its at-the-door value. Half-lives are calibrated from Florida walkability empirical data.
| POI kind | Half-life (mi) | Composite weight |
|---|---|---|
| Grocery | 1.2 | 0.35 |
| Pharmacy | 0.6 | 0.25 |
| Park | 0.8 | 0.20 |
| Library | 1.8 | 0.10 |
| Restaurant | 0.3 | 0.10 |
Scoring formula
- For each POI within 5 miles (PostGIS ST_DWithin with 5 mi buffer), compute exponential decay weight:
w(d) = exp(-distance_mi / half_life_mi) - Cap individual POI contribution at 0.9 (avoids in-building POI dominating the score)
- Sum capped weights per kind, normalize to 0-100 (5 POIs at half-life saturate to ~100)
- Weighted average across kinds yields the composite 0-100 score
- Floor 5 (avoids zero score for truly remote addresses)
- Tier mapping:
- HIGH = composite >= 65
- MEDIUM = 35 <= composite < 65
- LOW = composite < 35
POI tag filtering (ETL-time)
OSM tag filtering happens at ETL time so the runtime query is cheap. We include grocery (shop=supermarket / shop=greengrocer) but exclude convenience stores (shop=convenience). Restaurants are amenity=restaurant only; fast-food (amenity=fast_food) is excluded. Parks come from leisure=park (or USGS GAP / FL State Parks DB in v1.1, with a has_facilities filter that excludes wilderness tracts).
v1 limitations (honest callouts)
- Curated FL OSM POI snapshot (~40 POIs). Coverage is biased toward persona geographies (Tampa, Sarasota, Lakewood Ranch, Miami, Orlando, Jacksonville, Naples, Pinellas, 30A, Winter Park). v1.1 expands to full FL Overpass fetch (~50k POIs).
- USGS GAP + FL State Parks DB not yet integrated. v1 uses OSM park entries only. v1.1 layers in protected areas + has_facilities filter to exclude wilderness tracts from the park count.
- No GTFS transit-aware scoring. v1.1 will add Miami-Dade, Hillsborough HART, JTA, Palm Tran feeds for transit accessibility.
- No walkability metrics beyond POI density. Sidewalk %, intersection density, and bike-lane proximity are v1.1 candidates.
Profile-aware nudges (BLUF prose only)
The tier letter is profile-agnostic (per the 2026-06-09 Decision 4 consensus). Profile context only affects the BLUF prose framing: young children under 5 surface playground proximity, retirees surface pharmacy + library, STR investors surface restaurant + beach access.
Sources (2-source consensus rule)
- OpenStreetMap via Overpass API, accessed 2026.
- Walkability KDE methodology: Journal of Transport Geography v114, 103789 (2024).
- Accessibility scoring: Computers, Environment & Urban Systems v110, 102045 (2025).
- USDA Food Access Research Atlas (2024) for grocery threshold calibration.
Changelog
1.0.0 (2026-06-10): Initial. Continuous KDE scoring with 5 POI kinds, curated FL snapshot covering persona geographies. Per the 2026-06-09 DeepSeek triangulation Alternative 2 pick (continuous KDE wins the "ship one in 5h" tiebreaker).