Some thoughts I had on this, as I tend to agree with the first two comments that the complexity of the filter interface is excessive.
I think it comes mainly down finding the category structure rather confusing and hard to control. I understand that each category represents a set of related suboptions, but the logic that controls that is not always easy to understand. For example, "Significance" includes "Minor edits" and "Non-minor edits". I can see those as mutually exclusive, and I realize that checking both ORs them together, but so does not checking them both. Why am I being given 2 ways to express the same thing? This does not make much sense. I tend to think of checkboxes as switches, and a design that results in leaving all switches off having the same meaning as leaving them all on is just bizarre.
It becomes more complex in larger categories where it is less obvious how the coverage of the individual options interacts. For example, on "Watchlisted pages", I can select "On Watchlist" and "Not on Watchlist". Logically, ORing those together, that should cover everything, but there's a third option "New Watchlist changes". So if I don't select that, then I would see pages on my watchlist, but not if they are new changes? That's either a contradiction or at least a peculiar combination to support, and moreover makes it even less clear what the design principle is. The options within a category are ORed together, but how are they derived----they are clearly not meant to be covering (partitioning the space), since they include negative versions, but they also do not sum to 0 (cover in both positive and negative senses). Having some options subsumed by others results in unclear combinations (A OR not-A, but not (B subset A)?).
Basically, an AND/OR structure to the across/within categories does makes some sense and is easy to reason about abstractly, but it does not always seem clear how this works with the way the options within categories are formed, using positive and negatives that sometimes includes overlap.
I imagine lots of thought was put into the logical design and that other interface designs were considered. Just to be sure, though, were there reasons why a simple list of selectable, non-overlapping, positive-only options (perhaps only stylistically separated into categories) was not sufficient?
Other, more specific issues I have noted:
- . I have checked all filters in the "user registration and experience" category, so I can highlight some of them differently. This works, but in the list of active filters they are all shown greyed out, and any that I did not change the highlighting for are declared as having no effect. However, having selected some to highlight, had I not selected the remaining ones those would be excluded. So, contrary to the mouseover text, they actually are having an effect.
After some investigation, digging through the help text, I did find out that highlighting and filtering are independent. So I guess I could leave them all unchecked, and just modify highlighting to generate the same effect. This, again, comes down to the confusing ability to specify the same thing in two ways. It also points to an interface design problem----the filtering and highlighting are in the same row, prefixed by the filter checkbox, visually implying the checkbox controls both features. Perhaps putting the highlight drop-down prior to the checkbox would make it more obvious that these are independent.
- Bots are shown despite selecting "Human (not bot)" and not selecting "Bot". For example, "Bawolff bot" edits continue to be shown. That doesn't gibe with my understanding of how this works at all, so either I remain confused, or this is a bug.
- Under the "Translations" category I originally selected "Not translations", and not any of the other options, hoping to exclude translations. But translations still showed up! I understand (now) that this option is referring to the separate sub-pages for translation units, but those unit-pages are an (obscure, low-level) artefact of how translations are implemented. Certainly it did not match my expectation of what it meant to show or not show "translated pages".
Note that despite the above comments I do find these filters very useful, and I will continue using them on Recent Changes. It's just that my effective use of them has been based more on repeated, slightly frustrating trial-and-error than any kind of well reasoned methodology or clear abstract model given by the interface.