Hi Nigel
Why would you offset the input nodes (+&-) by 2 diode drops if you are using global feedback for DC bias? If your output stage is an equal follower between the rails, you shouldn't have much offset with a simple differential, unless the input stage isn't balanced correctly because of R3 value. VR3 is 1.2V as per Vbe of Q5. Q1 and Q2 should have equal 2.5mA through each as set by Q3 5mA CCS. 1.2V/2.5mA = 480R....or 470R for a common value. For offset control, you could use a small trim pot as emitter degeneration on Q's 1 & 2, wiper to collector of Q3. This would let you slightly mis-balance the differential so as to achieve closer 0VDC output. Personally, I don't like that method, but that doesn't mean it doesn't work.

Perhaps a trim pot in series with R3, say 20% of R3? However, there are better,
less simple ways to control offset. Also, I would think you'd get better linearity from using a discrete buffered VAS instead of a Darlington, JMO.
My two cents worth .....(even less than two pence

)
