The command controller on the PFI is different from the ones on the test stands ( a "VCCM600S" vs. a Xilinx KC705). One important difference is that with the same firmware image, the polarity of the power control bits is 1=off, 0=on. (some LVDS pairs got flipped, maybe?). This affect both the power and reset parts of the POW MPS command.

Either the FPGA itself or the FPGA control code needs to convert the polarity from the logical 1=on, 0=off to the correct physical polarity. Updating the FPGA is probably the right thing, if we can figure out how to distinguish the two models.

Changing the host-side python library code would be more flexible – there may be other differences. I propose adding "pfi" as a hostname, in addition to the existing "fpga" and "localhost", which are used to distinguish between a real system and the simulator.