active sessions per minute

Served from concurrency_minute, complete through — UTC. The newest couple of minutes are not sealed yet, so the dashed line stops short of the solid one — that distance is the pipeline's lag, shown rather than hidden. The metric covers all traffic, not just the fleet, so an unfiltered read can sit above the fleet line if anything else is writing.

filter

Phase is not a filter here on purpose: it is the session's state now, so filtering a time series by it would drop the history of every session that has since changed state — which reads as data loss rather than as a filter.

how to read the gap

• Solidis the fleet's own record. It did not infer activity — it decided the state and wrote the interval down at the moment of each transition, so it is exact by construction.
• Dashed is ClickHouse running the real five-term predicate over the events the fleet wrote. Not the load simulator's heartbeat-lease estimate, which peaks 37% high and would show a gap that says nothing about the pipeline.
• A gap means one of three things: the pipeline is wrong, events were lost on the way in, or reordering broke something. It is the one measurement here that can fail.
• The dashed line lags briefly by design — events are batched about once a second and inserted asynchronously, so the newest minute is often still filling.