It is absolutely *NOT* a
R-2R ladder. What it is, is a binary weighted conductance resistor network, (+ R6, 510 ohms in parallel to set a 'base' current) sourcing current from the OPAMP's +V to its -in, which acts a a virtual earth, at a potential controlled by the +in which is biassed by the R9 pot, which thus sets the resistance to current factor. Q3 passes >99% of the balancing feedback current to the virtual earth through the load, acting as a controlled current source with a compliance voltage approx 1V below the virtual earth potential.
As all the transistors are PNP and the whole circuit 'hangs' of the positive rail, to avoid confusion measure with respect to OPAMP V+ (bridge rectifier + out)!
U4 pin 6 (OUT) behaviour is exactly as expected. Irrespective of the load at the M01 terminals, and the digital inputs, it should track the voltage at the R6 pot wiper, except one Vbe drop higher, till the OPAMP rails, approx 2V below its + supply. Therefore R6 should be ineffective for approx 1/3 of its wiper range, with the effective end being wiper towards the 7905 regulator output.
The circuit's min full scale current is approx 13mA, and its max 43.5mA, adjusted by R9. To convert this to 0-10V at the M01 terminals it requires a load across those terminals of between 230 ohms and approx 750 ohms. I suspect it was originally designed as a 4-20mA loop sender, with R8 adjustable to trim to 4mA. To test it, simply connect a DMM on a 200mA range across the M01 terminals.
Odds are its missing a load resistor across the VFD's input terminals. If the VFD has been replaced this is a prime candidate. Check the specs on the old VFD, if it had a 4-20mA loop input option, all is explained.
Edit: corrected direction of Q3 Vbe drop offset.