I went back and looked and I used 10KΩ 1% resistors I ratio-matched to 0.01% for Ian's jig.
After looking at matching in situ at the high Ic currents vs using Ian's jig I've concluded that Ian's jig will be easier.
Devices matched using his jig still match well in the preamp.
I only tested maybe 4-5 pairs and the results were consistent enough I went on to something else.
Here is the latest draft schematic:
Drawing updated March 15, 2019.
A summary of the changes:
I added more jumpers to provide a +62 dB gain option. (+56 dB unbalanced output.)
(Note VR1 is only useful at lower gains when J6 is open due to the hop-off resistance of VR1.)
Added J8 to defeat the servo.
Removed R17/18 so the servo now drives the lower arms of the bias bridge.
This decreases the attenuation of the bias network from "4" to "2."
Overall servo gain now appears to be (and measures) "2."
Added R8 and R9.
The ZTX851, being a power device, has large geometry.
Cobo is about 45 pF typical. (The PNP ZTX951 is much larger at 74 pF typ.)
In comparison a 2N4401 is <10 pF.
The large Cob requires a large C4 and C5 for stability.
Slower op amps, such as the NJM2068 require an even larger C4 and C5, around 330 pF.
R8 and R9 isolate the transistor's output capacitance from the summing node allowing C4 and C5 to be reduced in size.
Banging the preamp with a square wave shows increased stability that eliminates ringing and overshoot.
Some op amps do not require R8/R9 and large C4/C5 combinations so in many cases R8 and R9 can be 0Ω.
At +62 dB gain I can't measure a difference in noise performance with R8/R9 at 0Ω or 1KΩ.
Using the NJM2068 and a particularly "bad" pair of ZTX851 I was able to get improved waveform fidelity with a 1000Ω/220pF than with 0Ω/330pF.
With a 1KΩ/220pF combination the preamp will run with just about any op amp including an NJM2068 or MC33078.
My first choice is a 5532.
I'll do a table of what combinations work best.
The R8/R9 trick was used in the Amek 9098 mic preamp.
I've also seen it used in a noise test jig that I recall posting earlier here.
That jig used an AD797 and the values were much smaller.
It's better to have spots on the board for them and not use them then need them and not have them.
I'm willing to pay for 4 extra holes.
One other thing I might add is that the bridge Vos adjustment method really only works properly (as shown above) when the bias Q-point voltage is half the collector supply. (5.5V.)
If the values of the lower arms of the bridge are lowered to reduce Ic, the differential servo gain is reduced.
I don't see that as a limitation here, but I need to think through the consequences in a different context.
A servo that introduces a common mode voltage might be better.
One thing that the Deboo offers is that it has a passive input pole and uses a single C.
A Deboo with a Vcm input might be useful later on.