Resolution:
The following situation can only occur when the Policy Intervals in question are operating on the same Resource Group.

When one Policy Interval's end warning time overlaps another Policy Interval's start (or goes beyond that Policy Interval's start period): an active Policy Interval takes precedence over any remaining end warning period. However, if the end warning period is so long that it goes beyond a subsequent Policy Interval's end warning period, the Policy Interval with the longer warning period takes precedence.

The particular issue here can be seen in an example like this:

Policy Interval1 has start=noon, end=1pm, end warning period = 600 minutes
Policy Interval2 has start=2pm, end=3pm, end warning period = 5 minutes

Policy Interval1 kicks in, performs its tasks, ends, starts draining, waits for a long time. For daemons that come non-busy during that hour, they move as expected. Nothing gets abandoned (because everything is still in warning period).

Policy Interval2 kicks in, performs its tasks, ends. Its end warning period indicates that the force/abandon decision should be taken in the first cycle after 3:05. But because Policy Interval1's end warning period is so long, there's no force/abandon decision in that cycle.