fix(recommendation): make the candidate draw reproducible, not accidentally so

Four arms of the candidate query ended in a bare `ORDER BY random()` with no
seed: similar_artists, likes_overlap, coplay_artists and random_fill.

Such an arm returns a STABLE set only while its LIMIT exceeds the rows
eligible for it — at that point it returns all of them and the order stops
mattering, because scoreAndSortCandidates sorts by track id before drawing
jitter. Below that threshold it returns a random SUBSET, and two builds on
the same day draw different ones.

So daily determinism held BY ACCIDENT, and only for libraries smaller than
the limits. Any real library is larger, which means same-day rebuilds have
been producing different mixes since those arms were written — invisible,
because a mix that changes after a refresh looks like a feature rather than
a broken promise.

Found by breaking it: cutting RandomFill to 10 while tuning Songs-like
turned TestBuildSystemPlaylists_DailyNonceDeterminism red. That test seeds
~20 tracks against a default RandomFill of 30, so its determinism came from
the limit exceeding the library, not from the code being right. It is now a
real guard.

The arms order by md5(id || $12) instead. The CALLER decides what that
means, which is the point: system mixes pass a per-(user, day) seed and get
the determinism they promise, radio passes a fresh value per request and
keeps varying, which is what a radio should do. Same shape the browse
queries in this file already use (`md5(id::text || current_date::text)`) —
existing idiom, not a new one.

This also unblocks the trim that #3881 wanted and could not have. Shrinking
a randomly-ordered arm was what broke membership; a seeded one takes a
smaller but REPRODUCIBLE slice. Songs-like's seed-independent share drops
from 29% to 12%, which was the original intent before determinism forced it
back to 20%.

TestSongsLikeLimits_DoNotShrinkTheUnseededRandomArms is DELETED rather than
kept passing. It existed to stop anyone trimming those arms while the
ordering was broken; the ordering is fixed, so the constraint is gone and a
guard enforcing it would now forbid correct code.

Was filed as blocked on tooling. It was not: `make generate-go` runs sqlc as
a pinned Go tool and is the same path CI takes.

One thing worth knowing for next time: three files in internal/db/dbq are
owned by root, left by `make generate` running sqlc in Docker. sqlc errored
on the first it could not write. They are untouched by this change and the
regeneration of recommendation.sql.go completed, but `make generate` will
keep failing until they are chowned.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01SQ31KQpYbStyK5y58UmPLH

internal/db/dbq/recommendation.sql.go

@@ -829,7 +829,7 @@ similar_artists AS (
     JOIN seed_artist sa ON asim.artist_a_id = sa.artist_id
     WHERE asim.source = 'listenbrainz'
       AND t.id NOT IN (SELECT id FROM excluded_ids)
-    ORDER BY asim.score DESC, random()
+    ORDER BY asim.score DESC, md5(t.id::text || $12::text)
     LIMIT $6
 ),
 tag_overlap AS (
@@ -857,7 +857,7 @@ likes_overlap AS (
           WHERE t.id = gl.track_id
             AND trim(g_overlap.g) IN (SELECT tag FROM seed_tags)
       )
-    ORDER BY random()
+    ORDER BY md5(gl.track_id::text || $12::text)
     LIMIT $8
 ),
 taste_overlap AS (
@@ -884,7 +884,7 @@ coplay_artists AS (
     WHERE asim.source = 'user_cooccurrence'
       AND t.id NOT IN (SELECT id FROM excluded_ids)
       AND t.id <> $2
-    ORDER BY asim.score DESC, random()
+    ORDER BY asim.score DESC, md5(t.id::text || $12::text)
     LIMIT $11
 ),
 random_fill AS (
@@ -900,7 +900,7 @@ random_fill AS (
           UNION SELECT track_id FROM taste_overlap
           UNION SELECT track_id FROM coplay_artists
       )
-    ORDER BY random()
+    ORDER BY md5(t.id::text || $12::text)
     LIMIT $9
 )
 SELECT
@@ -938,17 +938,18 @@ GROUP BY t.id, t.title, t.album_id, t.artist_id, t.duration_ms, t.file_path,
 `
 
 type LoadRadioCandidatesV2Params struct {
-	UserID  pgtype.UUID
-	ID      pgtype.UUID
-	Column3 interface{}
-	Column4 []pgtype.UUID
-	Limit   int32
-	Limit_2 int32
-	Limit_3 int32
-	Limit_4 int32
-	Limit_5 int32
-	Limit_6 int32
-	Limit_7 int32
+	UserID   pgtype.UUID
+	ID       pgtype.UUID
+	Column3  interface{}
+	Column4  []pgtype.UUID
+	Limit    int32
+	Limit_2  int32
+	Limit_3  int32
+	Limit_4  int32
+	Limit_5  int32
+	Limit_6  int32
+	Limit_7  int32
+	Column12 string
 }
 
 type LoadRadioCandidatesV2Row struct {
@@ -971,8 +972,22 @@ type LoadRadioCandidatesV2Row struct {
 //	enter the pool even when the similarity/random arms miss them; scored
 //	in Go via TasteMatch, so sim_score here is 0 pool-inclusion),
 //	$11 coplay_artists K (#1533 — tracks by artists co-played across the
-//	instance with the seed's artist; source='user_cooccurrence').
+//	instance with the seed's artist; source='user_cooccurrence'),
+//	$12 order_seed (text) — see below.
 //
+// $12 REPLACES `ORDER BY random()` IN FOUR ARMS (#3889). Those arms returned
+// a stable set only while their LIMIT exceeded the rows eligible for them: at
+// that point they returned all of them and the order stopped mattering,
+// because the caller sorts by track id before scoring. Below that threshold
+// they returned a random SUBSET, and two builds on the same day drew
+// different ones — so "daily determinism" held by accident, and only for
+// libraries smaller than the limits.
+//
+// md5(id || seed) keeps the intent — an arbitrary spread that changes when
+// the seed does — while making it reproducible for a given seed. The CALLER
+// decides what that means: system mixes pass a per-(user, day) string and get
+// the determinism they promise; radio passes a fresh value per request and
+// keeps varying, which is what a radio should do.
 // Returns same shape as LoadRadioCandidates plus similarity_score column.
 func (q *Queries) LoadRadioCandidatesV2(ctx context.Context, arg LoadRadioCandidatesV2Params) ([]LoadRadioCandidatesV2Row, error) {
 	rows, err := q.db.Query(ctx, loadRadioCandidatesV2,
@@ -987,6 +1002,7 @@ func (q *Queries) LoadRadioCandidatesV2(ctx context.Context, arg LoadRadioCandid
 		arg.Limit_5,
 		arg.Limit_6,
 		arg.Limit_7,
+		arg.Column12,
 	)
 	if err != nil {
 		return nil, err

internal/db/queries/recommendation.sql

@@ -45,7 +45,22 @@ WHERE t.id <> $2
 --   enter the pool even when the similarity/random arms miss them; scored
 --   in Go via TasteMatch, so sim_score here is 0 pool-inclusion),
 --   $11 coplay_artists K (#1533 — tracks by artists co-played across the
---   instance with the seed's artist; source='user_cooccurrence').
+--   instance with the seed's artist; source='user_cooccurrence'),
+--   $12 order_seed (text) — see below.
+--
+-- $12 REPLACES `ORDER BY random()` IN FOUR ARMS (#3889). Those arms returned
+-- a stable set only while their LIMIT exceeded the rows eligible for them: at
+-- that point they returned all of them and the order stopped mattering,
+-- because the caller sorts by track id before scoring. Below that threshold
+-- they returned a random SUBSET, and two builds on the same day drew
+-- different ones — so "daily determinism" held by accident, and only for
+-- libraries smaller than the limits.
+--
+-- md5(id || seed) keeps the intent — an arbitrary spread that changes when
+-- the seed does — while making it reproducible for a given seed. The CALLER
+-- decides what that means: system mixes pass a per-(user, day) string and get
+-- the determinism they promise; radio passes a fresh value per request and
+-- keeps varying, which is what a radio should do.
 -- Returns same shape as LoadRadioCandidates plus similarity_score column.
 
 WITH
@@ -87,7 +102,7 @@ similar_artists AS (
     JOIN seed_artist sa ON asim.artist_a_id = sa.artist_id
     WHERE asim.source = 'listenbrainz'
       AND t.id NOT IN (SELECT id FROM excluded_ids)
-    ORDER BY asim.score DESC, random()
+    ORDER BY asim.score DESC, md5(t.id::text || $12::text)
     LIMIT $6
 ),
 tag_overlap AS (
@@ -115,7 +130,7 @@ likes_overlap AS (
           WHERE t.id = gl.track_id
             AND trim(g_overlap.g) IN (SELECT tag FROM seed_tags)
       )
-    ORDER BY random()
+    ORDER BY md5(gl.track_id::text || $12::text)
     LIMIT $8
 ),
 taste_overlap AS (
@@ -142,7 +157,7 @@ coplay_artists AS (
     WHERE asim.source = 'user_cooccurrence'
       AND t.id NOT IN (SELECT id FROM excluded_ids)
       AND t.id <> $2
-    ORDER BY asim.score DESC, random()
+    ORDER BY asim.score DESC, md5(t.id::text || $12::text)
     LIMIT $11
 ),
 random_fill AS (
@@ -158,7 +173,7 @@ random_fill AS (
           UNION SELECT track_id FROM taste_overlap
           UNION SELECT track_id FROM coplay_artists
       )
-    ORDER BY random()
+    ORDER BY md5(t.id::text || $12::text)
     LIMIT $9
 )
 SELECT