- The ranking algorithm I’m using isn’t great at distinguishing pages about a topic from pages which merely mention a topic in passing.
- Because the Python docs are versioned, the PageRank they deserve is spread out over several URLs and they appear less relevant than they really are.
I have plans to fix both of these problems, but they’re pretty involved and I haven’t had the time to dig into the matter yet. For the moment, it’s definitely a gamble whether the results will be any good: sometimes they’re great, and other times they’re completely useless. (There’s a reason I put links to other search engines at the bottom of the results page!)