The same is largely true of DDG as well.
So ... what do you use when you want to do an "exact search" on the web?
Edit: In reply to a couple of comments ... I don't have a recent example to hand, and recent attempts have failed to reproduce the exact behaviour. Also, of course, Google tries to customise search results, so my example of a screw-up might behave perfectly for you. But I will keep trying to get an reproducable example, and would welcome examples from anyone else.
Edit 2: Example: I have open in another tab a specific web page, and from it I have copy/pasted "So a germ of an idea is enough" into Google, which fails to find it. Another phrase from that page is found, so the page has been indexed. Indeed, if I use the "site:" qualifier, it does find the reference. This is not exactly the behaviour I recently saw and was asking about, but it's close enough to be of interest.
Here is a recent HN discussion about it. I rant about it there. I won't repeat myself here. You can read plenty of other people say the same thing.
Consider this search: "I'm not a mathematician or a puzzler"
That finds this page: https://www.solipsys.co.uk/new/SpeakingAtThe2017MathsJam.htm...
This search fails: "So a germ of an idea is enough"
That fails, even though it's taken from the same page.
This search succeeds: site:www.solipsys.co.uk "So a germ of an idea is enough"
This is the example the triggered the question, but the question itself is about experiences I feel I've been having more often over the past 6 or 12 months, and that's better illustrated by the example given in [0].
No. Because to a first approximation, it is not possible to make it work all the time in practice.
Results returned are best effort. Best effort is largely limited by latency.
At the scale of the entire web, parts of the index will be absolutely inaccessible due to partitioning and practically inaccessible because they are sitting on disk (or simply don’t exist because they wouldn’t fit into memory) or are so far away that they can’t reliably be returned within the limits of acceptable latency — e.g. on a server in Japan for a London query (and anyway how likely is relevance of Japanese results?).
I mean maybe there are results for the “site” qualifier in memory on the server your query hits and maybe their aren’t. Either way, Google’s going to return results in 250ms if that’s the performance target. Either way Google’s not going to spin for twenty minutes or seconds it would take to crawl the site, process the results, and build the inverted index.
Everything is engineered. The engineering is guided by statistics. But so is performance. Sometimes better information is close by in hot memory. Sometimes it isn’t. Sometimes bandwidth is saturated. Sometimes your cell service is slow.
That’s probably why HN uses Algolia even though it costs money.