but of course it can't really take things like road surface quality into account.

I see this and I think "why the heck not??". It would require some level of crowd sourcing, but that'd be a useful thing to suggest to the team. If I work there I may try and do that :-}.

Your point about the recommendations is interesting, though; I've had multiple modes in my own biking life of whether I preferred direct path with traffic, or mellow bikepath without. That might be an area for personal preference learning. Hmmm.

(Can you tell I'm considering working at Google? It gives a much more personal investment in these kind of issues :-})