The thing is that when users are searching for something, they don’t always know how to express that well in a search criteria. The users’ goal isn’t usually to see what’s the best the search engine could do – it’s to find something.

Users get confidence from finding what they’re looking for – not from:

> I’m done. Here are the best results that I could find

Also, there’s a difference in searching when you know there’s a unique result, and when you only know what you’re looking for when you see it, like when searching for images.

When we get specific about the kind of search experience our users are looking for, that’s when we can come up with the best architecture and UI design to meet those needs – rather than just staying with the current mold out of momentum.

It sounds like what you’re saying is that technology is moving in directions to enable these kinds patterns easier. Cool.

However, windows search (without indexing) works similarly. I didn’t find myself refining the search term until I’m sure I’m not getting what I’m looking for. In my mind, I want to make sure that the search engine have done its thing to present the most optimal answer for me.

This is what we get with search engine like google. users get the impression ‘i’m done. here are the best result that i could find’. Showing incremental result will not give the same confidence to the user

Browser interaction does not necessarily rely on RPC. When using mainstream web servers, polling may be the best solution while waiting for backend results.

HTTP Push servers (Lightstreamer, Caplin Liberator, etc) or HTTP compliant middlewares (My-Channels Nirvana) exists and allow a fully asynchronous model, front to back.

Giorgio – agreed. I think pagination is an artifact of web 1.0.

I think another relevant example is twitter search. Their search results are always sorted by time, so it makes sense to display initial results and then subscribe to notifications about new ones.