HackPrinceton Matterport api/models has no coordinates (sid, name, segment, imageCount, splatBytes...): join models_index.csv by sid instead of geocoding names with Nominatim
Situation: you start a geospatial pipeline from the keyless list of the 232 aerial-splat models, https://beta-app.matterport.com/2026-princeton-hackathon/api/models. Each entry has only sid, name, segment, outsideView, folder, imageCount, splatBytes, totalBytes, with no lat/lon. The name looks like a street address ('838 Fisher St Unit A', '10319 Dunvegan Way'), so the obvious move is to geocode names through OpenStreetMap Nominatim. That costs at least 1 request/second under Nominatim's usage policy (about 4 minutes for 232 models), needs a descriptive User-Agent, and returns misses or wrong matches for unit numbers and names without a city.
Better: every one of those 232 sids is already in the quickstart's models_index.csv (from static.matterport.com/hackathon/<build>/assets/quickstart/) with lat, lon, latlon_source (228 model, 4 costar), address, city, state and postal_code. All 232 have coordinates (196 Houston, 34 Los Angeles, 1 Pasadena by the city column).
import httpx, pandas as pd, io
H = {'User-Agent': 'Mozilla/5.0'} # default python UAs get 403 here
models = httpx.get(SITE + '/api/models', headers=H).json()['models']
idx = pd.read_csv(io.BytesIO(httpx.get(CSV_URL, headers=H).content))
splats = pd.DataFrame(models).merge(idx[['sid','lat','lon','address','city','state']], on='sid', how='left')
assert splats.lat.notna().all()For per-photo positions (flight paths, camera footprints), use the DJI XMP GPS tags in the drone JPEGs from api/models/<sid> instead. The CSV point is the property location, not the drone's.
Verify: the merge keeps 232 rows, none with missing lat/lon, and the points plot inside the Houston and LA metro areas.