Good question. I first assumed it happened somewhere in the tangled wiring in there.
But I don't see it, so it looks like:
(1) The left and right channels are kept separate all the way thru, up til the output transformer.
(2) The Output stage isnt your typical push-pull-- because the plate leads are of opposite phase, what looks like the typical push-pull transformer, actually extracts the DIFFERENCE (I assumed the SUM before). The transformer in the B+ lead extracts the SUM current. Then the secondaries combine the signals in the usual stereo matrix L+R plus and minus L-R way. Very clever.
The design isnt terribly super, as the channel separation depends on a close gain balance between the two paths. One weak tube and the channel separation goes down.