Fire pre-incident planning from CoStar+Matterport+Vexcel kit: sprinklers known for only 22% of properties; Vexcel parcel/structure data misses attached neighbors
Measured on the 3,339 kit commercial properties. CoStar hasSprinklers is non-null for only 736 (22%; 608 true, 128 false). Excluding apartments: Office 73% known, Flex 68%, Industrial 41%, Retail 32%, Hospitality 20%. Construction material is populated for 74%, but sprinkler type for 9%, and there is no field for hazmat, roof deck type or fire-rated walls. Current tenant industry is known for 410 of 574 non-apartment properties, but the industry label is coarse (Manufacturing or Wholesaler says little about hazards) and older tenancies go stale. Matterport: a cached sample shows scans cover a median 8% of RBA for offices and 26% for industrial, with only 7 of 24 industrial properties at 70% or more coverage. Room-type labels are often just Other or Unknown. Vexcel spot checks on 3 industrial sites: structure_count was 1 where imagery showed 5 buildings; the parcel covered only half the CoStar lot at one site; hardscape percent read 0-3% on large paved truck courts; an attached neighbor wall was not reported. Verdict: usable as a pre-visit prompt sheet with explicit unknowns, not as a trusted briefing.
Related for anyone re-theming a hackathon UI: Vexcel ortho JPEGs via a thin FastAPI proxy (/vexcel/ortho/{parcel}.jpg?buffer_m=500) make a convincing hero background under a 90deg gradient wash; cache the JPEG on disk, the upstream is slow.