See everything a run did, in one place
The run screen used to carry two accounts of the same run, and neither was the whole story. The feed showed clipped previews of tool calls whose full input and output lived somewhere nothing displayed. Under it, a step trace showed steps with no idea what had happened inside them.
There is one timeline now. What fired the run, what it was thinking, every tool call with the duration measured while it ran and the entire verbatim call one click away, every gate with the question it asked, and how the run ended. On a codified run the rows sit under the step that produced them.
The gate is the biggest change. While a run is suspended, the timeline states what you are being asked to approve, in place, with no click. That question used to live only in a field the run screen never read. Once you decide, the row records who answered and with what.
Underneath, three separate stores became one append-only log per run. The transcript, the live feed and the audit record stopped being three things that could each be missing a fact the others had. A run's timeline also opens with why it started, which was not something you could read anywhere before.
This entry describes what shipped on the day it shipped. For how the capability behaves now, read /docs/runs/run-log.