Gabriele looked at the nonstationarity of the SRCL coupling in 77666, and proposed DHARD P as a degree of freedom that could be responsible.
This morning I measured the SRCL to DARM transfer function with and without a small offset in the DHARD P error point. I chose an offset of 10 counts which is about the rms of the DHARD P error signal (spectrum attached). This changed the coupling from SRCL to DARM (with the feedforward on) by 20%. So indeed it seems that we need to reduce the RMS of the DHARD P loop to have a stationary SRCL coupling.
Excitation with offset on: 5/9/24 16:16:52 UTC - 16:20:39 UTC with offset off we took a longer stretch of data in case it is useful to look at: 16:21:47-16:34:18 UTC.