System

warpsys: — records dropped by the lossy recording pipelines (adapters, endpoints, client) when their bounded channel was full.

Like Issues, this is chart only: the key is warpsys:records-dropped:{pipeline}:{tier}:{stamp}, always time-bucketed, with no lifetime total.

Nothing here is a job failure. A non-zero value means a recording pipeline saturated and threw diagnostics away rather than blocking the work it was recording — deliberate behaviour, but it does mean the adapter, endpoint or client surfaces are under-reporting for that window. The Dashboard's Dropped (24h) tile reads the same data, and a throttled operational event fires on the same signal.

If this is regularly non-zero the fix is upstream: record less (per-adapter FailuresOnly), or route the firehose to OTel instead of the database.