Great post! Just a few thoughts though.

Correct me if i’m wrong but with ffill(), bfill() or any other resampling method you are augmenting your data with sometimes misleading synthetic data in cases you need to assure no data were reported on a specific time range.

What I usually do is capping the minutes, seconds, micro, nano seconds of every value then applying a groupby followed by an aggregation function: mean() quantile() etc