Looking back through data I have been able to find when the 3Hz bump in B1/DARM started.

Figure 1 shows the week it has happened. On December  17 there is a change in the DARM spectrum, with noise in the 10Hz-25Hz noise band getting much lower, but a bump with variable height appearing at 3Hz.

Figure 2 shows a couple of spectra before the change (red and cyan) a couple after (blue and green) and the current situation (purple).

Looking in the logbook the origin of the change is clear. There was an ISC shift on subtraction filters https://logbook.virgo-gw.eu/virgo/?r=65832 . The last activity were changes in the CARM2DARM subtraction filter, and in figure 4 of that entry a bump in DARM at 3Hz clearly appears when the subtraction is turned on. This explains why the bump appears in the OMC spectrum, the OMC length has to follow CARM. 

This is probably the origin of the more numerous short data segment for Virgo, with B1 saturation chopping data usable by data analysis into short segments, many of them to short to usable by offline pipeline. This have been reported at the Virgo week this week, https://tds.virgo-gw.eu/?r=24945, and apparently the problem has started in December. So most likely it is the same problem.

Looking at the starting point of the Detchar investigation, the problem did indeed start in the middle of December, https://logbook.virgo-gw.eu/virgo/?r=66288

Figure 1. Looking at the CARM correction coherence with DARM and h(t), it is present around 20Hz for DARM but not for h(t). My guess is that the h(t) reconstruction uses precisely measured response of the individual mirrors (NE and WE), and knows that some of the CARM correction is actually applied in the DARM degree of freedom, because the strength of the NE coil-magnet pairs and the WE coil-magnet pairs is not exactly the same. 

The WE coil-magnet pair strength has changed due to the mirror replacement, so the CARM to DARM coupling has increased as the driving matrix has not been update. But the results with Hrec show that instead of doing a CARM to DARM subtraction, a more finely adjusted driving matrix should be able to avoid any coupling of CARM to DARM.

I disabled the CARM2DARM subtraction from 6:56 UTC, the driving matrix will be studied offline.

2025-06-20 06h53m46 UTC AcConstChCmSet> CARM_DARM_GAIN, val 0 rampTime 10 - loaded and request to apply received - sender "ISC_shell_bersanet_farmn5_20250618_154343"

Two plots regarding the impact of CARM2DARM subtraction on the bump at 3 Hz on DARM.

B1 PD1 saturation rate in Science mode (1-day bins), from September 1st until this morning. The two vertical dashed lines correspond roughly to when the  CARM2DARM subtraction filter was in production.

Same plot, updated with one week more of data.