My preference is to think deeply about it alone, and either take the results of that to the team or even come up with a very rough mockup/prototype and/or some working code for the bits that I initially have no idea how to implement (again, in a time-boxed manner)- and only then have a proper bike-shedding session.

Otherwise I find so much time is wasted on tiny details that don't matter, and the overall picture still fails to emerge (aka a definitional bike-shedding session).