Both iterated data.topArtists (capped at 50 by play count), while genres.ts deliberately iterates the full artistPlayCounts map "to capture the long tail of less-played genres". Artist enrichment already fetches info for every unique artist a listener has, not just their top 50, so mood/obscurity were leaving that already-fetched data on the table — for any listener with more than 50 distinct artists (most listeners with real history), both scores were computed on an arbitrarily truncated, less representative subset for no reason.