Sigh. I hate being handed the initiative like this; it feels like rope to hang myself with.

My recollection is that the Gestalt psychologists believed that they could gain information through a technique called introspection, but they couldn't. Their results depended fiercely on the dominant theory of the laboratory where they studied.

Unless I am mistaken, behaviorism in psychology was partly a reaction against the failures of introspection, and to some extent cognitive science (e.g. Simon and Newell's Protocol Analysis) was a reaction against excessive behaviorism.

As I understand the current scientific practice, introspection should definitely not be believed. It can be recorded as verbal behavior to be explained, but analyzing the data that you get this way is labor-intensive. So researchers usually use techniques that produce data which is easier to analyze (and get a paper out of). For example, various forms of timing tests like the IAT, or questionnaires with lots of confusion questions, either of which can be put into one of those nice Fisher's ANOVA structures.

My prejudice is that armchair-only techniques have been well explored with only moderate successes, and we should be looking for more physical, external, buildable techniques for rationality.