I would like to make sure that Flume works properly and does not discard events which meet my selection criteria.
Probably you can log the complete events into a file using file_roll sink and then you can apply the filtration logic (whichever you used in the interceptor) on that file to get the list of filtered events. Then this can be matched and verified with the Flume filtered events. If you are dealing with huge number of events then you could do this for a sampling duration.
Note: To test the above solution, you can create two different channels for same source so that one can be used for logging and other for the filtration