How a Workbook runs (execution model)

How a Workbook runs (the execution model)

Understanding the run model makes everything predictable. A Workbook is a DAG — nodes are steps, edges define order and data flow.

Levels and parallelism

Before running, the orchestrator computes topological levels: nodes with no dependency on each other sit in the same level and execute in parallel; the next level waits for the previous. So a Login → {APM Observe, AI Analyze} fan-out runs the two children at the same time, not one after the other.

Node status — what each means

Status	Meaning
success	The node ran and its check passed.
failed	A semantic failure — e.g. an assertion missed, an eval below threshold. Not retried.
error	An uncaught/infra error (timeout, network, exception). Eligible for retry.
skipped	A branch deselected it, or all its parents were skipped.
waiting	Queued, or paused on an Approval.

Retries

Every node has a Retry & timeout section in its config panel: Retries, Delay between retries and a Hard timeout. Retries apply only to error status (transient/infra) — never to a semantic failed (retrying a missed assertion is pointless) nor to success.

Outputs and variables

Each node writes outputs that become {{variables}} for downstream nodes. The orchestrator also injects system variables — {{_executionId}}, {{_workflowId}}, {{_nodeId}} and your organization's id {{_orgId}} — available to every node.

The run timeline

A run streams live: each node emits start/progress/complete events; the timeline shows status, outputs and AI narration as it happens, and the whole run is saved as a replayable report.