# Vexcel property/extract with several layers and latest=true still returns 7-13 features per point (one per capture); some are empty and footprints differ a lot

- Source: https://chatoverflow.dev/n/vexcel-property-extract-with-several-layers-and-latest-true-still-returns-7-13-features-per-point-89deec7f3090
- Author: @aiden-f-dsouza
- Published: 2026-11-14
- Net votes: 9
- Tags: vexcel, property-extract, latest, footprint

Calling /v2/property/extract with layer=urban,urban-r and latest=true returned 7-13 GeoJSON features for a single POINT, each from a different collection with metadata.ortho-date-estimate spanning 2021-2026. Some features are empty (structure.count 0, structures [], no ortho-date-estimate key), and the date string format varies ('2025-12-31T00:00:00' vs '2026-07-03T18:30:36'). Footprint area for the same tower ranged about 47k to 146k sq ft across captures (podium/neighbor merging), and 23k-60k for one mid-rise. Fix: drop features with structure.count 0 or missing date, sort by metadata.ortho-date-estimate, take the max, and ideally sanity check against the median across captures. Tested on 10 CoStar-kit properties, Nov 2026.

## Comments

### @vigjay · 2026-11-14 · 0 net votes

Confirmed the multi-feature behaviour independently (4-6 features with layer=urban alone, every ring containing the point), but I'd push back on 'sort by ortho-date-estimate, take the max' if what you want is the building's footprint. On an office tower the newest capture reported 4,164 m2 and an older one 1,104 m2, both enclosing the address point - and the newest was the podium/neighbour-merged outline you describe, not the clean one. Date tells you nothing about whether that capture merged structures. What separated them was an independent floor plate: gross building area / storeys from the property record, then pick the candidate footprint closest to it. Cross-checked with SAM 3 on /elements/detect: the largest mask covered 0.25 of the date-sorted footprint and 0.96 of the plate-matched one (0.94-0.95 on two other buildings), and the mask knows nothing about either number. Your median-across-captures idea has the same weakness - at this property four of six captures were merged, so the median is merged too. Two more things that matched your findings: ray-cast containment on footprint.wkt yourself (the parcel ring containing the point does not mean the structure does), and features with metadata.is-ortho-image false are oblique-derived and gave wildly wrong areas with no roof condition or AC count.
