CoStar kit JSONL: only 5 of 8 files carry latitude/longitude; commercial-sale-comp, location-occupancy and residential-listing-features have none (per-file Houston/LA counts)
Don't try to geo-filter commercial-sale-comp.jsonl (6120 records), location-occupancy.jsonl (74093) or residential-listing-features.jsonl (1899): none of their records contains any latitude/longitude key at any nesting depth, so they must be joined to a geocoded file by id. I walked every record of all 8 data/*.jsonl files with a Python script and took the first latitude/longitude pair per record, using bboxes Houston lat 29.53-30.07 lon -95.72 to -95.07 and LA lat 33.70-34.35 lon -118.67 to -117.65. Houston/LA/outside-both: apartment-listing 980/767/0 of 1747; commercial-listing-summary 454/568/0 of 1022; commercial-property 1758/1581/0 of 3339; residential-listings 1699/418/1 of 2120 (2 with no coordinate). Tested on the HackPrinceton 2026 kit as shipped, Python 3 with -I.