How NearbyIndex scores nearby infrastructure

What the current model answers

For an address or coordinate, the current model answers one narrow question: which daily amenities are close in straight-line distance? It does not decide whether the complete walking experience is safe, practical, or comfortable.

The eight categories

Every score is built from eight category sub-scores. Each category looks at points of interest within a search radius around the location, weights closer amenities more than far ones, and produces a 0–100 sub-score. The categories are:

• Groceries — supermarkets and grocery stores
• Restaurants — restaurants, cafés, bars
• Education — schools, universities, libraries, kindergartens
• Parks — parks, playgrounds, and gardens
• Transit — public transport stops and stations
• Healthcare — hospitals, clinics, pharmacies
• Shopping — general retail beyond groceries
• Entertainment — cinemas, museums, theatres, nightlife

The total score is a weighted blend of all eight categories. The current model uses straight-line distance. It does not use walking routes or travel time.

Where the data comes from

• Points of interest — Overture Maps Foundation (open data). Categories are normalised into one internal taxonomy. Source coverage still differs across cities and countries.
• City list and populations — GeoNames (CC BY 4.0).
• Neighborhood boundaries — Overture Maps division_area dataset, used to label areas inside a city page's breakdown table.

How city pages are computed

The current process infers a rectangular urban area from the middle 90% of local point-of-interest coordinates. It then scores an unweighted grid across that rectangle.

Each point has equal influence. The rectangle is not an official municipal boundary and can include nearby places. It can also give sparse edge areas more influence than their population suggests.

City pages publish the median, the 10th-to-90th percentile range, and shares above 60 and 80. These thresholds describe amenity access. They do not prove car-free living or complete walkability.

Amenity Access Profile v1 formulas

Every reviewed city uses one versioned evidence profile. The city page, comparison table, and public JSON file use the same code and the same values.

• Median access — the 50th percentile analyzed point score.
• High-access coverage — the percentage of analyzed points that score at least 60.
• Consistency — 100 minus the difference between the 90th and 10th percentile scores.
• Category balance — 100 minus the difference between the highest and lowest category scores.
• Named-area coverage — the percentage of grid points matched to a real source boundary.
• Estimated scored area — the rectangular grid extent reconstructed from its cell centres.
• Mapped place density — scored-category place records divided by the estimated scored area.

Generated compass labels do not count as real area labels. A city page with less than 40% named-area coverage does not publish a ranked area table.

The complete reviewed dataset is available as versioned JSON.

What full walkability needs next

The current place data is enough for amenity access. It is not enough for a full walkability score. That model needs four global inputs before NearbyIndex can make a stronger claim:

1. A routable pedestrian graph for real walking distance, crossings, access rules, and barriers. The planned source is the Overture transportation theme.
2. A population grid for resident-weighted city results. The planned source is the European Commission GHS-POP dataset.
3. Elevation data for slope cost. The planned global source is the Copernicus DEM.
4. GTFS schedules for service frequency and operating periods. The Mobility Database is the planned global feed catalog.

Sidewalk condition, crossing quality, traffic injury risk, lighting, and accessibility remain local-data problems. NearbyIndex will not infer those factors where reliable source data does not exist.

How fresh the data is

City statistics are precomputed and stored with a calculation timestamp. A point request can use a recent cached result. Overture releases new source data regularly, but an update is not immediate.

What the score does not capture

• Neighborhood safety or crime rates.
• School quality (we count schools, we don't rank them).
• Sidewalk quality, hilliness, weather, or how nice the walk feels.
• Travel time — only straight-line distance is used in the MVP.
• Official city boundaries or population-weighted city averages.
• Property values, rents, or commute distance to a workplace.
• Local events, opening hours, or seasonal availability.

Treat the score as a starting filter, not a verdict. Two equal scores describe similar model output. They do not prove that two places offer the same living experience.