Experimenting a bit further with the previous setup (and realized that the ~10 Hz was simply the sum from the mixer, aliased due to sample rate, and not a frequency discrepancy between the two input signals!).
The addition from the previous experiment, is that I've added a 500 kHz low-pass filter on the output, so that I can remove the sum and retain just the DC value. As before, I then adjusted the phase from -180 to +180 manually turning the dial, and captured that to a 'scope. This time I'm using 1 MHz signals because I forgot to set them to 10 MHz when I captured the 'scope trace. Everything is quite wideband in the circuit anyway, so that doesn't change anything material.
The output is pretty rock-solid, easy to measure with a multimeter or view with 'scope.
I *think* this could be useful to observe phase noise. This is for no real reason, I'm just kicking tyres with this experiment.
Currently I'm just using two outputs from a single 'sig-gen, but might try to split that across two tomorrow just to see what happens. Or I could try to use a scrap AliExpress OCXO for one of the inputs, I have a few and been meaning to do something with them.