Watermark
A marker in a stream stating the point up to which data is believed complete, so downstream consumers know when a window can be closed.
A watermark is a stream's statement about its own completeness: everything with an event time earlier than this mark has, as far as the system knows, already arrived. It exists because records do not arrive in the order they happened - a node reconnects, a region lags, a source replays - and something has to decide when a one-minute window is safe to compute.
The trade-off is explicit and unavoidable. Advance the watermark aggressively and windows close fast but late records are dropped or force a correction; advance it conservatively and results are more complete but arrive later. Publishing the watermark alongside the data is what lets a consumer make that decision themselves rather than inheriting the provider's, and it is a large part of what separates a dataset with real data freshness guarantees from one that simply feels current.
Real-time data, at edge speed.
A live evaluation measures EdgeOrigin against your coverage requirements: decision-ready real-time data, delivery latency into your systems, and record-level provenance.