The data file in question is mvn_sep_l2_s1-cal-svy-full_20220801_v04_r03.cdf . Filtering out times < 0 before calling cdfepoch.to_datetime() and comparing to unfiltered to_datetime() output should show the problem. We may want to do something similar in cdf_to_tplot, if this issue hasn't been resolved yet.
There's an open bug report in cdflib that with certain datasets, TT2000 times can get truncated to 1-hour resolution:
https://github.com/MAVENSDC/cdflib/issues/233
The data file in question is mvn_sep_l2_s1-cal-svy-full_20220801_v04_r03.cdf . Filtering out times < 0 before calling cdfepoch.to_datetime() and comparing to unfiltered to_datetime() output should show the problem. We may want to do something similar in cdf_to_tplot, if this issue hasn't been resolved yet.