Tracks people come back to

Constraints

1. 1A pair qualifies when that listener played that track on 2 or more distinct days.
2. 2listening_days counts distinct days; plays counts rows. They differ whenever a listener repeated a track within one day.
3. 3COUNT(*) >= 2 is the wrong test — it accepts two plays four minutes apart.
4. 4The day comes from date(played_at); the time of day is irrelevant.
5. 5The filter tests the group, so it belongs in HAVING.
6. 6Order by listening_days descending, then user_id, then track_id.

Worked example

u2 played Static Bloom twice on 6 January and again on 11 February: 3 plays across 2 days, so the pair qualifies with listening_days 2 and plays 3. u8 played Quiet Riot Hour twice on 20 March, four minutes apart, and never again. Filter on COUNT(*) >= 2 and u8 joins the list, reporting a habit that was really one evening. Counting distinct days correctly leaves them out.

What this tests

Choosing the counting unit — COUNT(*) against COUNT(DISTINCT date(...)) — and placing the test in HAVING because it is a property of the group, not the row.