Audio Research CL-30 Bias Drift

the cathode dc supply capacitor may be blown. do a visual and then remove cap(s) and test them with cap tester.
input resistor to the grid may also get blown as well. check it with continuity tester.
it's common problem with manymany tube amps.

Here's the test condition:
No tubes installed, power on and warmed up for a minute.
The bias rail before the grid circuits measures -53 volts and maintains that value. However, from left to right as viewed from the top rear, of the 4 6550 tube sockets, the two inboard sockets slowly go down in bias voltage. This is after I have replaced all of the resistors in the circuit for good measure. This makes absolutely no sense at all, and lifting the cold side of the 1uf coupling caps to the grids confirms there is no DC leakage causing any problems. The only thing I can think is the opto couplers but it still doesn't make any sense. There should be no voltage drop without a load, and even with the tubes installed, there should be minimal if any drop with a bias supply so well regulated. I have slept on on it and I woke up without any answers. A real head scratcher.
Thanks for your response.
Billy

i think it has nothing to do with coupling caps.
it's all in DC chain not in signal...

I agree. I just stated that as a measure of test procedure. In a common PP amplifier, the bias voltage can be affected by leaky coupling caps. I see this fairly often in guitar amplifiers. I will check the DC cathode circuit as you suggested.
Thanks.

Replacement of the optical isolators (LDR) made a difference. The bias has stabilized but is uneven at each pair. The difference averages out to be around 1.4 volts. I can live with that after I pair up the tubes where the current is best divided amongst the pairs.
Audio Research is a prime example of pushing the envelope at the expense of simplicity. A biasing scheme should never have a problem like this. I mean, if the design goal was to "track" the bias and adjust according to impedance swing at the output, I can see why. But, troubleshooting is another matter entirely. It's like an automatic variable voltage regulator, or power scaling but to what purpose I can only theorize. Has anyone read about this type of output stage control as it relates to the CL-30 design?
I will be glad to see this go back home with the owner never to be heard from again. I'll take a pair of class A monoblock's any day!
Thanks