Good examples. Raters have - afaik- to assume multiple possible intentions on two word searches (and general searches on one). This results in better ratings fo

Good examples. Raters have - afaik- to assume multiple possible intentions on two word searches (and general searches on one). This results in better ratings for results (top 10 generally) that partially match different search intentions. so for "pandas databases" it would match the panda python libraries, but also databases literally of pandas ... and maybe a company that is called pandas and has databases. And this is the reason why we can never have "pure" results (in most cases) for a specific search, because the algorithm and the search rating is aiming for result-topic-diversity most of the time (if not all of the time).