I checked the 1854-0247, and the DC voltages at the emitter and collector seemed a bit off compared to the schematic (+13V, -0.3V vs +11V, -0.7V). As I had some 2n3866As in hand, I swapped it out. The original transistor tested ok in one of those budget component testers, but maybe the issue is RF specific. Well, I powered it up and...
No change.
So I decided to have a good think, and my conclusion is that there is nothing wrong with the reference output - the problem is with the user manual
.
My reasoning is this: there has been a significant revision to the reference circuit in later models like mine. See the attached drawing where I have sketched it out (note that the drawing comes from the HP 8656A schematic, as it is much clearer, however the circuit design shown in the 8656B manual is identical except for different component designators). Basically, early HP 8656Bs (and all HP 8656As) run an emitter current of about 20mA, but the later 'Bs run an emitter current of about 11mA. As the gain of a common base amplifier is inversely proportional to the dynamic emitter resistance, which is in turn inversely proportional to the emitter current (i.e. gain is proportional to emitter current), the later design has roughly 6dB less gain than the earlier models. Therefore instead of +16-19dBm, the later model output should be more like +10-13dBm, which is closer to what I am seeing.
Well, ok, there is probably some dodgy math in there, but my overall conclusion remains that the 50MHz reference is later 'Bs should be expected to have a low level by design, and this is not reflected in the available service manual.