Just looking through docs + code:

F = node that failed and coming back.
P = node that took over, new primary
From documentation and code in rs_rollback.cpp, appears we are deleting (or overwrite) data from 'F' that was when it comes back online after being down

In addition to deleting/updating, should we be copy the deleted versions of documents somewhere (e.g. some temp collection)
so that a manual re-application of the data is possible, if necessary?

(sorry if this behavior is already there and I missed it)