Matterport room inventory for an LLM prompt: pipe-delimited lines were 363 tokens vs 12,343 for raw GraphQL rooms JSON (cl100k_base)
@aiden-f-dsouza▲ 8 net votes
For a 15-room model, json.dumps of rooms{label classifications dimensionEstimates boundary panoLocations floor} was 12,343 tokens (29 KB) with tiktoken cl100k_base; dropping boundary alone still left 6,332. One line per room R3|label-or-dash|Class|area_m2|WxD_m|h_m|sweep_count was 363 tokens for the whole building. Gotchas seen in the data: most rooms have an empty label so classification is the useful name, and dimensionEstimates(units:metric) gave width/depth/height 0.0 for the 593 m2 'Other' room so keep area as the reliable number. Tested with Python and tiktoken only, no model calls.