Yesterday, we restored the original IY height to recover the DARM sensitivity.
Today, I realized that I had forgotten to restore the original IY TM P2L drive align value, which defines the IY BPC basis.
After setting the IY P2L drive alignment value back to -2.64, the sensitivity returned to the previous best level (Fig.1).

In addition, I noticed a mysterious jump in the BPC EY yaw error signal after the IFO transitioned to the DC readout state, while the other BPC error signals remained around zero (Fig.2).
This likely causes drifting in the IFO yaw alignment and eventually leads to lock loss.
Further investigation is needed.