It looks like the timeseries agg fuzzer ignores any generated indexes and creates its own relatively simple, ascending indexes only. This may not be true for the update fuzzer. 

I think that represents a potentially large gap in testing and it at the very least confusing.