I suspect what you're noticing is that a reboot returns the events to normal when shutdown disables the data source, then startup re-enables the data source and it considers it a different event.
It may warrant some thinking about a reboot in particular, but in general if one had a state detector of false on a point, which was the state, then disabled the point or source for an indeterminate amount of time, and upon first poll is still in the state, should the event be raised? Currently the answer is yes.
Or, similarly, does it make sense to say the event is still occurring if the data source is disabled, or is that not known?
There is almost certainly a workaround but it could take some tinkering. It would probably involve a modification to the custom template to check if some points timestamp is appropriate and some handling to record the startup time or shutdown times such that they are usable.
Another surprising possibility would be to terminate Mango more aggressively. In Windows, killing the command prompt and in Unix doing a -9 will end the process without shutting down the data sources (which can easily corrupt the last few samples in the database) and then the events will be deemed duplicates at startup and mostly not raise.