I am rather fond of the coupled-line hybrid idea: it can be built in
a way that everything stays ratiometric, so the coupling ratio is very
nearly constant over temperature, and of course the directionality
lets you observe things you can't just from monitoring voltage at a
point. It's possible to build one with low coupling without too much
trouble; -60dB coupling isn't out of the question, for sure. I'm
imagining a design I could make reliably with simple machine tools
that would work well for the OP's application: 100 watts at about
1GHz as I recall in the through line, and coupling on the order of
-60dB to get to about -10dBm coupled power and have negligible effect
on the through line. There's a free fields solver software package
that will accurately predict the coupling, and with the right design
and normal machine shop tolerances the coupling and impedance should
be accurate to a fraction of a dB and better than a percent,
respectively. Perhaps I can run some examples to see if I'm off-base
on that, but that's what my mental calculations tell me at the moment.


Actually, the exact coupling ratio probably isn't important in this
application, because it could be "calibrated out". Stability would be a
bigger concern, and it's certainly possible to design a coupler that is
very temperature stable by choosing the right dimensions so that things
change in the right ratios.