This is indicative of a larger issue - testing is probably as difficult as solving the halting problem i.e. code could be generated from proper tests, yet teams tend to trust their tests completely. I see high profile websites having severe usability issues or being outright broken in ways that would be immediately caught by "interns randomly click here and there" usability tests. But these version got deployed probably because testing did not show any regressions.

I tend to believe that if user complaints about new problems or regressions increase over statistical noise - there is a problem.

> yet teams tend to trust their tests completely.

Well said. This is a big problem. We see a similar problem with the use of telemetry data as well.

I noticed the same. I've wondered for years why it happened and sometimes when I'm frustrated I try to think about it. But I am not entirely sure that the degradation in search results for me happened only in the past 5 years. Maybe, but I'm not sure.

I had no idea about Hummingbird though.

In 2009 Google was amazing for diagnosing Linux issues. I would just copy the error from the console and I'd have links to the issue tracker, a work around and the version in which the bug was fixed. Today I get a link to some github project that has nothing to do with what I'm working on and was closed as being an upstream issue.

I don't have the time, money or energy to build a specific crawler, but a Linux search engine that indexed all the major distros, packages, mailing lists, forums and issue trackers would be amazing.

> Google's "Hummingbird"

I had assumed that Google search had gone downhill because it started trying to "personalize" my search results. That wasn't a great explanation though, as I don't use a Google account.

Hummingbird seems a much more likely explanation.

Oh, they still "personalize" your search results.

I do a full clear on my web browser (cookies / offline storage / history, everything) and then open YouTube in a private browsing window and it asks me which of my two Gmail accounts I want to log in with. I'd guess it's just a combo of external IP and browser fingerprint, but it's creepy.

I know they do, and I consider this a real problem. I was just saying that personalization isn't a completely satisfactory explanation for the decline in Google search result quality. It is likely to be a factor in that, though.