I turned off the LSC RFPD interface chassis (D1102060) for IMC length diode using the switch at the back panel, disconnected the length diode cable, connected a DB9 break out board at the front panel without connecting anything else (but the back panel was still connected to the ADC), switched on/off the chassis (again using the switch at the back) and measured how the +-18V supply for the diodes behave.

To my disappointment, every time I did this test, for about 1ms to 1.5ms there was a window where the negative power was -18V but the positive was basically non-existent (below 5V). It seems as if the negative voltage goes live immediately. See the first picture, the scale is 5V and 1ms/div.

This is the unit that we've been using since the start of IMC installation.

I did the same test for WFS interface (D1101865), and it was only marginally better than the length interface (second picture, same scale as the first one).

This is the unit that replaced the broken unit, but since this one was put into service WFS was connected/disconnected hot.

Switching off was much better for both of these units, basically both of the rails came down slowly at the same time (the last picture, note the slower 4ms/div scale).

During these tests, the only things powered by the chassis were the DC interface cards in the chassis and of course the power protection board itself. 

I don't know if the units are broken, we'll test them further though the change Rich Abbott made to diode boxes will hopefully make this behavior harmless.