something else. It fires once, when the mark lands on a row — and on mount, if a

marked row is already there, which is what a reader expanding a section they had collapsed sees.

Later polls do not re-scroll, so a reader who has scrolled elsewhere stays where they are.

It also stops at a collapsed ancestor, deliberately. The viewer already expands a section the

moment its status turns to error — unless the reader collapsed that section themselves, which

`userCollapsedSections` records and honours (`useTaskRunLogSections.ts:234-247`), with the same

treatment for replicas and release files. A mark arriving into a section the reader has shut does

not reopen it. Doing so would break the precedence this decision just established, where an

explicit toggle outranks the mark, and it would override the reader at the one moment they have

most clearly said what they want on screen. Nothing is hidden by this: a collapsed section still

carries its own status, so the header turns to the error glyph and says a failure is inside. When

the reader opens it, the marked row is already unfolded and the section is already scrolled to it,

because D13's scroll fires on mount whenever a marked row is present, not only when a mark lands on

a row that is already showing.

The promise stops at the phase boundary too, deliberately. Both deploy surfaces key the viewer on the

- Collapsed ancestors (D14), same `datasetKey`: a failure arriving into a section the reader

collapsed leaves it collapsed, and expanding it afterwards shows the marked row already unfolded

and scrolled to.