Paying for real world data only helps an agent when the data pipeline can survive connection gaps Fresh activity from @vangrid_io today points toward agents paying for data to do real work The streaming layer shows a harder problem underneath that idea Vangrid can keep an open connection for one geographic zone and push new ground truth events as they arrive Each event carries its own timestamp, geometry, observation type, confidence, node count and provenance hash When nothing new arrives, the server sends a heartbeat every 30 seconds But the stream does not keep a replay buffer Drop the connection and observations arriving during that gap do not come back automatically The client has to reconnect, then use the spatial query endpoint to backfill whatever happened while the stream went dark That creates a different integration pattern for robots and physical agents Stream the world while connected Detect the gap Query the missing window Merge the missing observations Then continue from the live feed For a dashboard, a short disconnect can mean a blank patch on a chart For a machine moving through the physical world, the same gap can hide a vehicle, blocked path or changed environment Real time data alone cannot solve that The system consuming it also needs a way to rebuild state after connectivity breaks Vangrid puts that responsibility clearly at the API boundary Live events come through the stream Historical recovery comes through spatial queries Two paths around the same physical state That distinction gets more important once software agents start using ground truth to do real work instead of only reading it