The unlock at 04:55:31 UTC, as long as the ones at 03:40:02 and 02:39:17 are most likely due to a pathological behavior of the ML that we still need to understand. They are the same as the one reported in #67675.

We can observe EOM and PZT corrections increasing a few minutes before the unlock (fig.1) and the error signal oscillating at 60kHz just before the unlock (fig 2.)