ProtectedabandonedThe run whose abandoned packages the live region has already announced.
ProtectedfollowingThe run the loop is following, so a new one starts with a clean slate.
ProtectedlastThe HTTP status of the last failed drive, so the panel can be specific.
ProtectedmissesConsecutive failed polls. A host hiccup is not a failed update.
ProtectedoutcomeWhat the last failure means and what to do next — see runState.pollOutcome.
ProtectedpollOne alert for a run of failed polls, not one per poll.
ProtectedpollingProtectedrollingProtectedwatchingTrue once we have fallen back to reading progress instead of driving it.
The live region's sentence: progress while running, the outcome once it ends, nothing for a failure (the alert says it).
Undoing a run is asked for, finished or failed, and says whether the database goes back too.
ProtectedfollowA removal only takes things away (it resolves with
--no-update-with-dependencies), so it has no Downloading step. Shown and
then skipped, it read as something that had gone wrong (ClaudiusH,
2026-10-13).
Which extensions this run is for: the names, or the first few and a count.
ProtectedwatchRead-only progress, for when we cannot drive.
Keeps polling on the same backoff so a transient failure recovers into driving again by itself.
What is happening, while it happens.
This panel is the product. An update that works but shows a spinner is the thing being replaced — the complaint that started all of this was not "it failed", it was "the status has been showing running for a while and I don't know what is going on". So every piece of state the server knows is on screen: which phase, which item of how many, every line of the log, how long since anything moved, and which driver is turning the handle.
It also never advertises progress it has not been told about. When a poll comes back saying another driver holds the run, that is shown as exactly that rather than as a stalled bar.