Say you want to combine full-text scores with geo proximity. We offer function_score as a way to do that, but this is a very manual process, and it's hard to make sure those signals are combined in a way that actually makes sense. On the other hand, research has been conducted on that topic so we should look at how we could integrate better ways to combine relevance signals. I'm mentioning geo since it seems to be the more researched area but ideally we'd address that issue for time-based relevance and other common use-cases.