Operating

Running & observability

What a run tells you while it is happening, and after.

The event timeline#

Every run produces an ordered event stream you can scrub. Dragging the ribbon drives the canvas, the log position and the metrics header together, so you can watch the run reach the moment it failed.

Token and cost accounting#

Cost is recorded per node, not per run, so an expensive workflow tells you which step is expensive. Sub-cent values are never rounded to $0.00.

Failure clustering#

Failures are grouped by signature. Eighty-three failed runs are rarely eighty-three problems — they are usually three, and the Incidents page is where that collapse happens.