How recently a dataset was updated relative to the world it describes - closely related to latency, but measured as staleness rather than delay.
Freshness describes how current a dataset is: given a record, how long ago was the underlying observation made, and how far behind reality the dataset is as a whole. Where latency measures the delay on a single update, freshness is the standing property of the dataset - a dataset can have low latency on the records it does send yet still be stale if it only refreshes a source once a day.
Teams track freshness with an explicit update cadence and, often, a watermark that marks the point up to which the data is known to be complete. A dataset that quietly goes stale is worse than one that is openly delayed, because a consumer will keep trading or training on numbers they assume are current. Publishing freshness alongside the data - not just the data - is part of what makes a dataset trustworthy.
A live evaluation measures EdgeOrigin against your coverage requirements: decision-ready real-time data, delivery latency into your systems, and record-level provenance.