I looked at a recent lock loss event from the CHARD blend guardian state (gps time 1223672219), 

https://ldas-jobs.ligo-wa.caltech.edu/~lockloss/?event=1223672219

I saw an interesting glitch in ETMX_L2_MASTER_OUT about a second before the lock loss (see saturations figure), and followed this up with Sheila on the lockloss tool to look at various CHARD, DHARD channels surrounding the glitch (see CHARD figure).

Sheila suspected that because the CHARD pitch/yaw gains were not occurring simultaneously (happened about 0.3 seconds apart), that the waits for the ASC-CHARD channels in the CHARD_BLEND guardian state needed to be removed. In this stage, the low pass in A (ASC-CHARD_Y_A_SWSTAT) needs to happen at the same time that the gain is added for B (ASC-CHARD_Y_B_GAIN).