I wasn't thinking commercial, either.

An expert system is what you originally described -- an interface that uses a binary tree or other decision pattern to diagnose and offer solutions for a problem. Like Windows' diagnostic help thingy -- you choose from a category of problems and it starts with the simplest solution asking "did this work?" or something like "did this produce <X> or <Y>?". It then drills down to more complicated ideas until the expert system is exhausted. The lowest level is "please ask someone" -- in this case, "post your query to PM".

Yes, I know M$.

But I was thinking about a way of reducing the expected results of a search, nothing about trial & error. If we could acurately define the structure of our questions, we might even finish not having a question at all!

Sometimes, by just well structuring our doubts, we find the answer, don't you think so?

I was thinking of a way of helping us structure the question by reducing it's scope by choosing a sequential number of options. Each option should have a number of 'possible' tags associated to it.

It might be called an expert system, but I feel this name as something too pretentious for this subject :) . I was just thinking of a way of eliminating 'labels' attached to certain words found in any question (by intersections, for example). All the labels that are left should define a complete environment to the question (see here: http://thingamy.com/tagwork/# 4 an example). Then, the search would limit to that situation. Is it so difficult to build?

Everybody could help by attaching or eliminating such labels to/from every answer (how does http://del.icio.us/ do it?). The available labels might be supervised by experts, and every page should limit to the labels that are involved in such subject.

Every answer for the specified questions might be 'taged' for a later revision to get the required labels.

So, then, every question, might have to pass through a sort of filter that would reduce the implied labels to the ones that are declared related to the question. (4 example: applying an intersection pattern to the total possible labels)

It should work in a similar way as when doing authomatic translations of a text. The translated word should be interpreted as a member of an environment that is limited by the sense of the subject.

Resuming: one comes with a number of possible words about a question. After passing some decision making filters, the words get increased by the site,but also reduced in number by the filters until the expected answer is reached. Or at least, until the desired subject is well defined.

.{\('v')/}   C H E E R   U P !
 _`(___)' ___a_l_b_e_r_t_o_________

Wherever I lay my KNOPPIX disk, a new FREE LINUX nation could be established.