fix(server): daily-rotate all deterministic mixes + diversity top-up fallback
test-go / test (push) Failing after 12s
test-go / integration (push) Failing after 6m50s

Operator feedback on the prior unification commit (7473e98d):

1. NewForYou should daily-rotate alongside Rediscover and FirstListens.
   The 'newest album first regardless of day' intent was the wrong
   call - operator wants visible day-over-day movement on every
   deterministic mix surface. Spec flipped to dailyRotate: true.

2. Diversity caps (<=2 per album / <=3 per artist) on every mix, not
   just the historically-diverse ones. The 2-per-album limit has
   helped a lot on the operator's library; extending it to NewForYou
   and FirstListens (previously album-coherent / no cap) surfaces
   more distinct albums per day. Spec flipped to diversify: true on
   all five.

3. Fallback when diversity caps strip the pool below the 100-track
   target: finishMix now calls topUpFromRaw, which appends non-capped
   tracks from the raw SQL pool (preserving original ranked order +
   skipping duplicates) until the target is hit or the pool runs out.
   On rich libraries the cap yields >= 100 and top-up never runs; on
   thin / album-heavy libraries we ship a partly-diversified 100
   instead of a strictly-diversified 40.

Net effect: every deterministic mix now rotates day-over-day, every
mix gets the same diversity treatment (with graceful degradation),
and the producer surface stays a single factory over a spec list.
This commit is contained in:
2026-06-03 13:22:23 -04:00
parent 7473e98d91
commit 6da6cb5c5a
+64 -24
View File
@@ -13,8 +13,8 @@ import (
// Discovery mixes (#419-423). One generic producer + per-mix spec.
// Replaces the five near-identical produceXxx functions that all
// fetched ranked (id, album_id, artist_id) rows, optionally
// diversified, truncated, and emitted a single playlist.
// fetched ranked (id, album_id, artist_id) rows, diversified,
// truncated, and emitted a single playlist.
//
// Day-keying is a per-mix property captured by `dailyRotate`:
//
@@ -22,18 +22,19 @@ import (
// ORDER BY md5(t.id::text || $2::text), so the Go producer keeps
// SQL order. `dailyRotate: false`.
//
// - Rediscover / FirstListens — SQL accepts only $1 user_id and
// produces deterministic ordering. Same content day-over-day
// until library state shifts. `dailyRotate: true` applies a
// daily-deterministic rotate-left of the pool BEFORE diversify+
// truncate so each day's top-100 surfaces a different slice
// while contiguous-block ordering within each slice is preserved
// (matters for FirstListens which is album-coherent).
// - Rediscover / NewForYou / FirstListens — SQL accepts only $1
// user_id and produces deterministic ordering. `dailyRotate:
// true` applies a daily-deterministic rotate-left of the pool
// BEFORE diversify+truncate so each day's top-100 surfaces a
// different slice while contiguous-block ordering within each
// slice is preserved (matters for FirstListens / NewForYou which
// are album-coherent — rotation walks the album boundary cleanly
// rather than scrambling within an album).
//
// - NewForYou — SQL produces a newest-album-first ordering whose
// intent is "see what's new". Day-over-day same content is
// correct UX: the user's "what's new" list shouldn't rotate. The
// spec carries `dailyRotate: false` deliberately.
// Diversity is `true` for every mix: per-album <= 2 / per-artist <= 3.
// On thin libraries where the cap would chop the pool below 100,
// finishMix tops up from the uncapped raw pool so the mix still
// ships a full-length playlist — see topUpFromRaw.
// discoveryMixLen caps each mix at the same depth as For-You /
// Discover so shuffle-on-play has a varied pool within a day.
@@ -139,7 +140,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
},
{
name: "New for you", variant: "new_for_you",
diversify: false, dailyRotate: false, // newest-first is the intent
diversify: true, dailyRotate: true, // operator wants daily rotation on all deterministic mixes
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, _ string) ([]discoverTrack, error) {
rows, err := q.ListNewForYouTracks(ctx, uid)
if err != nil {
@@ -171,7 +172,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
},
{
name: "First listens", variant: "first_listens",
diversify: false, dailyRotate: true, // SQL has no date arg; album-coherent so rotate (not shuffle)
diversify: true, dailyRotate: true, // SQL has no date arg; daily rotate + diversity top-up
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, _ string) ([]discoverTrack, error) {
rows, err := q.ListFirstListensTracks(ctx, uid)
if err != nil {
@@ -186,17 +187,30 @@ var discoveryMixSpecs = []discoveryMixSpec{
},
}
// finishMix caps (per-album<=2 / per-artist<=3) when diversify is
// set, truncates to discoveryMixLen, and converts to the insert
// type. Album-coherent mixes (New for you, First Listens) pass
// diversify=false so whole albums survive.
// finishMix applies diversity caps (per-album <= 2 / per-artist <= 3)
// when diversify is set, with a top-up fallback when caps strip the
// pool below discoveryMixLen: the capped result is filled out with
// non-capped tracks (preserving original SQL order) until the target
// is hit or the raw pool runs out.
//
// The fallback matters on small / album-heavy libraries — the cap
// can chop a 200-row pool down to 40, and we'd rather ship a partly-
// diversified 100 than a strictly-diversified 40. On rich libraries
// the cap yields >= 100 and the top-up path never runs.
func finishMix(rows []discoverTrack, diversify bool) []rankedCandidate {
pool := rows
var pool []discoverTrack
if diversify {
pool = capByAlbumAndArtist(pool)
}
if len(pool) > discoveryMixLen {
pool = pool[:discoveryMixLen]
capped := capByAlbumAndArtist(rows)
if len(capped) >= discoveryMixLen {
pool = capped[:discoveryMixLen]
} else {
pool = topUpFromRaw(capped, rows, discoveryMixLen)
}
} else {
pool = rows
if len(pool) > discoveryMixLen {
pool = pool[:discoveryMixLen]
}
}
if len(pool) == 0 {
return nil
@@ -208,6 +222,32 @@ func finishMix(rows []discoverTrack, diversify bool) []rankedCandidate {
return tracks
}
// topUpFromRaw appends non-capped tracks from raw (in their original
// order) onto capped, skipping any already present, until the result
// reaches target or raw is exhausted. Preserves SQL ranking semantics
// for the non-diverse fill so the topped-up tail still trends best-
// first within each album.
func topUpFromRaw(capped, raw []discoverTrack, target int) []discoverTrack {
if len(capped) >= target {
return capped[:target]
}
seen := make(map[pgtype.UUID]struct{}, len(capped))
for _, t := range capped {
seen[t.ID] = struct{}{}
}
pool := capped
for _, t := range raw {
if _, in := seen[t.ID]; in {
continue
}
pool = append(pool, t)
if len(pool) >= target {
break
}
}
return pool
}
// emit wraps the finished track list in a single builtPlaylist (the
// discovery mixes are all singletons). nil tracks → no playlist.
func emit(name, variant string, tracks []rankedCandidate) []builtPlaylist {