Resolving production incidents during on-call rotations, and by extension, generating most of the platform's freshest operational knowledge as a byproduct.
Mostly normal engineering work, punctuated by pages that can come at any hour. During an incident, focused entirely on resolution, not documentation — the platform drafts a knowledge article from her closed ticket afterward, which she confirms with a light edit the next morning if she remembers to.
Being paged for something that, it turns out later, someone already solved six weeks ago, because nobody thought to check first, or the earlier draft never got confirmed and never became searchable.
Lower on-call load over time, specifically from fewer repeat incidents, not just faster individual resolutions.