Following runs

Every trigger creates a run: an object with frozen input, a step-by-step timeline, output, and an outcome.

The timeline

It shows what happened, in order, in product language:

Element	Content
Steps	Each one, with its state, duration, and attempt
Input and output	What each step received and produced
Failure reason	When there was one, with the step that stopped
Child runs	When the agent called another, with links to them

The timeline is append-only: nothing is rewritten afterwards.

What the timeline does not show

It records semantic state: what the run did. It does not show the model's internal reasoning, nor "thinking…", nor tokens. What matters for diagnosis is what happened, not the monologue.

The outcomes

Outcome	Meaning
Completed	A Result step completed; the output is what it assembled
Failed	A step failed with no error path
Cancelled	Someone requested cancellation
Expired	The total deadline ran out
Not approved	An approval was rejected

While running, it can be running, waiting (a wait or a retry), or waiting for approval.

When it fails, the screen says where

A run that did not complete opens showing the step that stopped, the reason in product language, and a shortcut to open that configuration.

Cancelling

A run can be cancelled at any moment. Cancellation is cooperative: it is checked between steps, never in the middle of a call in flight.

Child runs are cancelled along with it.

Re-running

Re-running creates a new run from the same active version. The previous one's history stays intact.

• With nothing supplied, it re-runs with the same effective input.
• With different input, it replaces the original and goes through the same validations.

Re-running a production run requires a publishing role.

Durable waits

A run can legitimately sit still:

Situation	State
Waiting for someone to approve	Waiting for approval
In a Wait step	Waiting
Between retries	Waiting
Waiting for a child run	Waiting

In all those cases it releases resources and comes back on its own. A run sitting for days waiting for approval is not a stuck run.

Test runs

They appear in the same list, marked as tests. They are real runs — only with a different origin. See Testing before publishing.

Common errors

Symptom	Likely cause	What to do
The run does not leave "waiting"	A pending approval or a Wait step	Check the current step in the timeline
It failed right at the start	No active version in the environment, or an inactive service account	See Publishing and Service account
It completed, but the result came back incomplete	An error was handled by a diversion	Look at the steps with handled failures in the result
An AI step failed	The objective was not met verifiably	See AI task

Next steps

• Deciding approvals
• Execution errors
• Common problems