Quality
Processing latency
Also called pipeline delay.
Processing latency is time spent inside the collector and warehouse before a hit is queryable. It is the pipeline's share of lag.
How it is measured
Measure from the moment the collector accepts a request to the moment a query returns it. Queues, batch writes and index builds all add to it.
In Tarsier, hits go to the collector and then to the analytics store, and the store may make them queryable after a short delay.
Worked example
A news site's collector accepts a hit in 40 ms, writes it in a batch every 10 seconds, and the query layer sees it 25 seconds after that. Total processing latency is about 35 seconds.
An editor's complaint that the dashboard is behind turns out to be that 35 seconds.
How it differs
Processing latency is the inside-the-pipeline delay. Lag is the total delay a person sees. Data freshness is how current a report is. A hit is the unit being delayed.
Common errors
Blaming the browser for pipeline delay. Ignoring batch windows. Expecting zero. Forgetting retries. Measuring once.
In practice
Time a test hit end to end. Ask for tighter batching only if 35 seconds matters.