It seems that the atlas profiler when viewed at a cluster level takes its data from the primary log and the user needs to select each specific secondary node and view the profiler at a node level to see the long running queries on each secondary.

Requesting that this is clarified in the documentation,

I think the relevant page is here as this is linked when user selects the option to `view factors affecting (profiler) data

https://www.mongodb.com/docs/atlas/performance-advisor/#monitor-and-improve-slow-queries