So back to the drawing board. I went back to measure the digital signals on the switching chips, this time not with the multimeter as I had done the first time, but with the scope ...
they are ok and match what I measure in my functional device. I also figured out that in the 20V range, I can get a signal through to the track and hold. Then I get a decent signal on TP 103 - see the scope image and a reading that is stable albeit off by a factor of two. Note how the Track and Hold signal is not as symmetrical as in the functional device.
I then checked the Lm339 again, too. In contrast to the manual, it shows -26.5V on pin 12. Its supposed to have -5.5V at zero input. However, here it seems the defective 8840A/AF and the functional 8840A differ in construction. The zener CR303 is connected to -15V in the functional 8840A and to -30V in the defective 8840A/AF. So this could be by design and I just don't have the right manual - or the zener is broken. The CR303 in the defective 8840A/AF has the same marking 74 6 610 as CR309/11 which are both 3.9V Zeners that clamp the output of the track and hold. So I guess its by design.
That would be odd, though, as the LM339 is rated for symmetric supplies with +/- 18V max. Now +5/-26.5 doesn't exceed that differential, but still.
The PCB of the defective AF states 8840A/AF-3001 Rev C, the functional A is Rev E - both are dated 1983, though.