I checked the TF from H1 to OpLev with new template file.
In addition, I controled PR2 pitch to center of QPDs to avoid using OpLev out ot linear range.
Figure 1 shows  the result.
TF shapes seem fine but gains are totally different from the previous measurement.
Also, TML TF becomes lower gains while P and Y have larger gain.

According to klog21408, recalubration was done after the measurement of reference, and L, P, Y gains are changed (fig2) as follows:
L: 420 -> 284
P: 120 -> 340
Y: 105 -> 215

Green graph in fig3 show the TF from H1 to P/Y scaled with the calibration factor changes.
As you can see, gain difference becomes much smaller.

Re line in fig4 shows the TF from H1 to LEN QPD HOR signals, which is L signals before decoupling.
As you can see, coherence seems much better, so LEN sensor seems fine.
Green line in fig4 shows the TF from H1 to Y with a factor of 0.07, wich is the decoupling factor between L and Y.
As you can see, the gain of H1 to Y with 0.07 is almost same: that means LEN QPD signals see just a coupling from Y.
So, it might be changed due to the change of beam spot on the mirror, which needs to check later.

One remaining mystery is why the TF from H1 to decoupled L doesn't have f^-2 stracture at high frequency.
We need more careful investigation.