Top two tracks in every genre

Constraints

1. 1Rank inside each genre, not across the catalogue — a jazz track competes with jazz tracks.
2. 2plays counts rows in stream_plays. A listener who plays one track three times contributes three plays.
3. 3Ties are real in this data: two indie tracks are level on 4 plays and two electronic tracks are level on 3. Both members of a tie must appear, so those genres return three rows each rather than two.
4. 4Every genre with at least one play appears.
5. 5Order by genre, then genre_rank, then track_id.

Worked example

indie has three tracks: Northern Lights on 5 plays, then Paper Boats and Glass Field on 4 each. ROW_NUMBER() would number them 1, 2, 3, and a WHERE ... <= 2 filter would keep whichever of the two 4-play tracks it happened to number 2 — a different answer on a different run. RANK() gives 1, 2, 2, so the filter keeps all three rows and the tie stays visible.

What this tests

Ranking inside a partition, and the difference between RANK and ROW_NUMBER when values tie — a choice that decides whether a tied row is reported or silently lost.