Runs

Every execution — a pipeline pass, a column run, an import, an agent decision — leaves behind a Run: the receipt with status, scope, cost, and a link into the control room. A run is not a log entry you have to go dig up — every agent reply and every feed row links straight to it.

What a run carries

• Status — pending, running, succeeded, partially failed, aborted.
• Scope — how many rows/leads are affected, which playbook or table.
• Cost — credits consumed, against the cost cap the run started under.
• Link — the direct path to the detail view in the control room.

Dry run vs. live

Every run starts in one of two modes: dry run resolves templates and conditions and shows what would happen — no cost, no external effect. Live actually executes. Whether a run needs approval follows from mode × effect: a live run with external, paid, or externally visible effect (a send, a CRM write) always requires approval; a dry run never does. That’s a single rule for every kind of run, not a special case per feature — see Approvals.

Error handling

A run that stalls doesn’t sit forever on “running” — after a wait period it’s visibly marked as failed, never silently forgotten. Retry is possible per item and never repeats items that already succeeded.

For table columns and imports (see Tables), the exact same run model applies — part of the same, gradually rolling out database area.