Okay - U3 was a puzzle to me. I think I know how it works. So a bit of a circuit description:
The inputs to U3 are Div_C and COM_Sense (two "grounds"). The output of U3 is the output voltage (+ the voltage drop of the buffer transistor).
How does it work?
Div_C is the "wiper" of the divider.
Div_D is the OUT_SENSE line. It has the output voltage on it.
Div_B is the reference voltage: -10V for positive voltages and +10V for negative voltages (opposite polarity).
The ratios of the divider are calculated so that the voltage at Div_C nulls, when the correct output voltage and an opposite polarity reference voltage are applied to the divider. U3 - a uA714 (OP07) is responsible for nulling Div_C by creating the proper output voltage. C6 1uF and C7 0.1uF (the two big metalized film caps connected to Div_C) are there to probably ensure the output reacts fast to dynamic loads (given that the entire system is littered with RC pairs for stability).
As an example, for a selected +10V output, the "Wiper" is centered, and when U3 settles the 10V output (on Div_D) and -10V (on Div_B) cancel each other out.
Also, since U3 tries zero Div_C - it does not really matter if OUT needs to be a source or a sink (the NPN or PNP buffers are active). Of course as long as the current is within compliance (+-60mA or so).
The ability to have the 10V at one polarity and the 19.999V at the opposite polarity is the reason V and V2 are floating relative to each other. Note - Since both are regulated to 33V (minus the BE forward bias voltage of the MJE340) - I don't really understand why the RMS voltage of the V and V2 windings are different. The power supplies consist of a current source (2x 2N2907) driving a 33V Zener (1N4752) buffered by an MJE340 power transistor. V2 power supply is a bit different in that Q7 detects the V2 current flow (mainly U3) and lights the Overload LED.
Also - as per the previous posts, DIV_E, DIV_F, DIV_G and the DIV_H (COM) connect to the polarity lever and set the polarity by flipping the anode/cathode of the LM399 between COM and U1. It uses a crossover configuration switch (2x2 matrix). This then in turn flips the 10V reference a negative 10V (for positive output voltages) or to a positive 10V (for negative output voltages).
Since the 10V and the main divider are the most important in terms of TC - the resistors used are part of matched sets - hence the weird markings on the resistors. (HighVoltage has a red set of resistors from these guys:
http://prpinc.com/products/leaded-metal-film-resistors/pr-series-leaded-ultra-precise-metal-film-resistors). I don't know who makes the other style resistors. I also don't understand why the values are always different (and thus requiring a different R32-R37). Perhaps they are trying to match the Tempco of the lever resistors?
I also have no idea who makes the lever divider resistors. They look like PWW resistors, edge sealed with epoxy. The switches are Cherry but I cannot find them in any catalog.
COM level set: Q3 and Q4 using two diodes D18 and D19 respectively (I think they are ~3V Zeners) - at 3V below TP4 (for negative voltages) and at 2.8V above -V2 for positive voltages. TP1_10V is the trigger that causes Q3 and Q4 to switch the voltage.
Don't know what the difference between COM and Div_A is (and if the common ground is in the divider - which would make sense). Div_a and Div_H measure like a short...