I've spot-checked a few of the published results. I am pretty sure that all the FARs in the GWTC papers are in units of inverse julian years, e.g., https://git.ligo.org/publications/O3/o3b-cbc-catalog/-/blob/master/data/data_process?ref_type=heads#L30 And like @tito-canton says, it seems that PyCBC uses julian years (again a spot check): https://pycbc.org/pycbc/latest/html/_modules/pycbc/events/coinc.html#LiveCoincTimeslideBackgroundEstimator.ifar I was trying to find the scripts that were used to produce FARs for GWTC-2 and GWTC-3 for GstLAL (was it Becca that produced these?) but I don't have them immediately available. I think I did this for GWTC-1 and if my memory serves I used julian years for that in alignment with the GWTC-1 paper. Again, I haven't been able to quickly locate the conversion scripts so I'm not 100% sure, but it seems that things are generally pretty consistently julian years.
So I vote for julian years.