Looks good to me.
I follow your description, but with some things I have a field of probabilities and don't want to make too few/many assumptions.
If you're getting some data, but not what you'd fully expect then I'd entertain the event normalization piece (data not being normalized for "reasons"), but it's a pretty small edge case so hopefully you'll be able to see real data for expected events.
If we were looking at it side by side we could probably figure it out in short order, so hopefully you see the events from your test which I'd say would make my theory be plausible, otherwise if you need it done in a crunch Support should be able to help demystify it pretty quickly.
Happy to keep discussing, just sometimes time is the more finite resource.