Event Time vs Processing Time

Event time is the moment an event actually occurred in the world. Processing time is the moment a system touched the record describing it. They are never the same number, and the gap between them is where every honest latency figure comes from. A provider that timestamps on arrival is reporting its own processing time and calling it real-time data, which quietly hides the delay the record accumulated before it was stamped.

The distinction has teeth in analysis, not just in marketing. Aggregating by processing time means a network hiccup rewrites your history: events that happened in one minute land in the bucket of the minute they arrived. Aggregating by event time keeps the series faithful to the world, but forces the system to decide how long to wait for stragglers, which is what a watermark is for. Any dataset used for backtesting has to be indexed by event time, or the test is measuring your own infrastructure rather than the market.

EdgeOrigin stamps event time at the edge node that heard the event, against a network-wide reference clock, and exposes the serving node's time separately - so latency is a subtraction the client can perform rather than a figure taken on faith.