A customizable categorical rating system would give each song not just a single rating, but several user-defined ratings.  This system could give a powerful method for creating situationally perfect playlists while being (hopefully) not too difficult to implement.

For example, one could rate songs like this:

All Along the Watchtower - Jimi Hendrix
Rockingness: ****
Lyrical Depth: *****
My girlfriend Likes It: ***
Overall: *****

Now, let's say I'm studying with my girlfriend.  I don't want to be distracted by good lyrics and I want my girlfriend to enjoy the music.  I would tell Amarok to make a playlist that favors songs with a low Lyrical Depth score and a high My Girlfriend Likes It score (and not tell it to worry about Rockingness).  Amarok would probably avoid All Along the Watchtower because the lyrics are too good, but the Overall score and the fact that my girlfriend doesn't hate it might make it a candidate depending on how loosely Amarok is matching values.

For another example, lets look at how I would use this system:

3rd Planet - Modest Mouse
Normal: ****
Depressed: *****
Angry:
Bedtime: ***

When I'm in a particular mood, there are some songs that are good for that mood and some that aren't.  3rd Planet is good for when I'm feeling normal, great for when I'm feeling depressed, worthless if I'm **** off at something, and okay if I just want to go to sleep.  Letting Amarok know which mood I'm in would help it make an awesome playlist for me.

I hope this request interests someone.  I'll check back here if anyone wants to discuss it.  I don't know if I could be useful in actually contributing to implementing the feature, but if anyone likes the idea and can provide direction, I could be commanded to write appropriate Python code.