I see some improvements possible with split-personalities like Bernstein. The missing “composers” filter criteria feels relatively simple to work out. We’ll take a look.
Artists like Bernstein are tricky to pick apart automatically. There are a lot of people that appear to meet the criteria of being both conductor/performer + composer as a result of one out-of-character recording or some inaccurate metadata, so there is a balance to be struck when defining data-driven rules. In any system like that false positives or negatives are possible.
This part of the automation was designed to fail in the less damaging direction–hiding some options on the page of people like Bernstein. If we turn the knob the other way, we’d risk snowing the Beethoven or Mozart pages with thousands of albums that have those composers erroneously tagged as a performing artist.
The design space for data problems like this one does not look much like the end product–we are tweaking knobs pretty far away from the interface to produce the best emergent behavior possible for as many people as possible.
Like I said, I think we can do better, and we’ll look at this case–just wanted to give a little bit of insight here.