Agreed about utility of the signature process - and of course these days built in microprocessor self-tests are the way forward. I wonder why they didn't program the signatures into the end of one of the firmware EPROMS - at least that way they would stay in sync and could easily be read out (HP's required test equipment to fix one of these things usually includes half a lab full of other HP boxes, so one more eprom programmer wouldn't hurt surely?).
Normally I prefer logical faultfinding methods, but the block in question which intermittently malfunctions includes a statemachine, serial x parallel multiplier an EPROM used as a filter and various shift registers operated under microprocessor control. Point checks with the scope show it multiplying correctly (twos-complement serial multiplication - last time I saw someone checking this with a scope was on an Elliot computer).
I did start into disassembling the firware to find out exactly what it should be doing and found the relevant code, but needed more information on the address space to make progress and next step was to get an STM32 to dump out the 82S100 address decoder and this hasn't happened quite yet.