In manual processing, sometimes resetting the the epoch to the value of the most recent epoch greatly increase the RMS value of the fit.
This may be an indicator of the quality of the TLE elements fit, and could potentially be used as a "quick test" to see how a delta of 1 (day, week, month) might indicate the quality of the TLE, by looking at the "back-fit" of this artificially generated epoch.
Word of warning, it could also reflect underlying problems with the move_epoch_to_jd_new() function to change the TLE epoch.