Observed locations and residents
Café locations come from an OpenStreetMap snapshot dated 7 October 2026: 80 mapped cafés inside Leuven plus 18 nearby competitors. This is not a complete business census. Missing, closed or misclassified venues can change the picture.
Statbel’s 2025 population and matching boundaries locate 104,761 residents in 134 sectors. Another 145 residents have no mapped sector and are excluded from spatial modelling.
How the insights are produced
Café summaries count other mapped venues within 750 metres and identify the nearest alternative. Distances are straight-line. They describe location context, not measured sales, visits, popularity or quality. Branch names include an address or neighbourhood to distinguish locations.
Green residential areas have at least the median resident density across Leuven sectors and no more than three mapped cafés within 750 metres of the sector’s representative point. Areas without tested points are not highlighted. This is a transparent screening rule, not a validated definition of an underserved area.
The shortlist applies that same density requirement and a maximum of three cafés near each actual test point, then ranks eligible points with the Huff scenario. Selected points are at least 800 metres apart.
Optional model assumptions
The established huff library, version 1.9.13, precomputes nine scenarios across 351 points on a 400 m grid. Existing cafés have equal assumed appeal. A candidate has appeal 0.5, 1 or 2; distance sensitivity is 1, 2 or 3. Sector residents are represented at one interior point, with a 100 m distance floor. These assumptions have not been calibrated to observed visits.
Choice shares are conditional resident-weighted model probabilities, with no non-purchase option. They are not customer counts or profit forecasts. Unregistered students, commuters, visitors, foot traffic, route barriers, capacity, rents and planning permissions are not measured.
Data checks and the next practical test
Population totals reconcile, model probabilities sum to one and increased candidate appeal never lowers its model share. Closely spaced venues are not automatically merged: Onan’s own site describes two locations across from each other. Its mapped records have different addresses and are retained.
Check candidate streets and café coverage in person, then compare pedestrian counts before recommending premises. Independent public-data research, developed with AI assistance; no commercial results are claimed.