The 2-week metrics review found Discover beating the manual baseline on
skip rate — which for a discovery surface means it plays it safe. On a
single-user server it's effectively dormant + crude-random, and the
per-user taste profile (taste_profile_tags, #796) went unused (#1254 gap).
Add a fourth Discover candidate bucket, ListTasteUnheardTracksForDiscover:
unheard / non-liked / non-quarantined tracks ranked by summed taste-tag
weight over tracks.genre (split like the radio tag_overlap arm), md5
tiebreak. Its picks are stamped pick_kind = 'taste_unheard' (migration
0041 widens the CHECK on playlist_tracks + play_events, rule #36).
Rebalance the slot allocation 40/30/30 → taste_unheard 35 / dormant 30 /
cross_user 20 / random 15, and lead the interleave with taste_unheard so
a track shared with another bucket keeps the taste stamp and the
targeted-novelty arm stays measurable. Metrics label "Taste-matched" +
order entry added to the single server-side pickKindLabels map, so web
and Android surface the new breakdown row with no client change.
Cold start (empty taste_profile_tags) yields an empty taste bucket that
redistributes to the survivors, so Discover still fills.
Scribe #1488. Companion review outcomes: Songs-like starvation already
fixed (#1255); For You v2 ratified as-is (fresh-injection cost disproven).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The cross-user bucket query combined SELECT DISTINCT with ORDER BY md5(...),
which Postgres rejects at plan time (SQLSTATE 42P10). buildDiscoverCandidates
returned that error on first bucket failure, so the whole Discover playlist
was skipped every nightly run — even though the random bucket pool was
healthy. Switched to GROUP BY so the md5 ordering expression no longer needs
to appear in the select list, and hardened the function so future single-
bucket failures degrade gracefully via slot redistribution instead of taking
out the whole playlist.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds 'discover' as an accepted system_variant on the playlists
table — alongside 'for_you' and 'songs_like_artist' — and three
sqlc bucket queries that back the new playlist's daily candidate
selection.
The three buckets surface tracks the user hasn't played and
hasn't liked:
- Dormant artists: artists this user has played < 10 times total.
Surfaces the long tail of their library.
- Cross-user likes: tracks any OTHER user has liked but this one
hasn't engaged with. Empty on single-user servers; the Go-side
allocator redistributes the deficit across the other two
buckets.
- Random unheard: pure random sample as the safety net and the
cold-start fallback.
All three share exclusion filters: not-played by this user, not
liked by this user, not in lidarr_quarantine for this user. All
order by md5(track_id || dateStr) for daily determinism — same
pattern as the existing tieBreakHash logic in system.go.
LIMIT values are generous (80/60/200) so the per-album (<=2) /
per-artist (<=3) caps and slot redistribution in T2 have headroom.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
bvandeusen