Milestone 127: recommendation quality — provenance, tiered mixes, For You v2, tuning lab #107

Merged
bvandeusen merged 11 commits from dev into main 2026-07-03 10:13:50 -04:00
59 changed files with 5508 additions and 504 deletions
+12
View File
@@ -25,6 +25,7 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/lidarrrequests"
"git.fabledsword.com/bvandeusen/minstrel/internal/logging"
"git.fabledsword.com/bvandeusen/minstrel/internal/playlists"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
"git.fabledsword.com/bvandeusen/minstrel/internal/scrobble"
"git.fabledsword.com/bvandeusen/minstrel/internal/scrobble/listenbrainz"
"git.fabledsword.com/bvandeusen/minstrel/internal/server"
@@ -146,6 +147,16 @@ func run() error {
coverEnricher := coverart.NewEnricher(pool, logger.With("component", "coverart"), coverSettings)
coverEnricher.DataDir = cfg.Storage.DataDir
// Recommendation tuning lab (#1250): seeds shipped defaults on first
// boot and pushes the daily_mix weights + taste config into package
// playlists — must precede the scheduler so the first builds score
// with the operator's tuned values, not the pre-push literals.
recSettings, err := recsettings.New(ctx, pool, logger.With("component", "recsettings"))
if err != nil {
logger.Error("recommendation settings service init failed", "err", err)
os.Exit(1)
}
// One unified scan chain: library walk → MBID backfill → cover enrich.
// Boot-time scan and manual-trigger scans share this path; results land
// in scan_runs for the admin overview.
@@ -294,6 +305,7 @@ func run() error {
}, cfg.Events, cfg.Recommendation, cfg.Storage.DataDir, cfg.Branding, coverEnricher, coverSettings, scanner, scanCfg)
srv.Bus = bus
srv.PlaylistScheduler = playlistScheduler
srv.RecSettings = recSettings
srv.StreamSecret = cfg.StreamSecret
httpServer := &http.Server{
Addr: cfg.Server.Address,
+202
View File
@@ -0,0 +1,202 @@
// Admin recommendation-trends endpoint (#1251): weekly per-surface
// outcome series with tuning-audit markers — the verify half of the
// tune→verify loop the tuning lab (#1250) opens. Aggregated across
// all users because the knobs are global; rows carry rates only.
package api
import (
"encoding/json"
"net/http"
"sort"
"strconv"
"time"
"git.fabledsword.com/bvandeusen/minstrel/internal/apierror"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
)
const (
trendsDefaultWeeks = 12
trendsMaxWeeks = 52
// trendsAuditFetchCap bounds the audit fetch; markers older than
// the window are dropped in Go. Far above any real knob-turn count
// inside a year.
trendsAuditFetchCap = 500
)
// trendPoint is one week of one surface family's outcomes.
type trendPoint struct {
WeekStart string `json:"week_start"` // ISO date (Monday)
Plays int64 `json:"plays"`
Skips int64 `json:"skips"`
SkipRate float64 `json:"skip_rate"`
AvgCompletion float64 `json:"avg_completion"`
// TasteHitRate is the share of plays whose track's artist carries a
// positive weight in the player's current taste profile — a drifted
// but retroactive read on whether the surface feeds taste-fitting
// tracks.
TasteHitRate float64 `json:"taste_hit_rate"`
// completionN carries the completion sample count through same-week
// merges so avg_completion stays sample-weighted; not serialized.
completionN int64
}
// trendSeries is one surface family's weekly series.
type trendSeries struct {
Key string `json:"key"`
Label string `json:"label"`
Intent string `json:"intent"` // go_to | discovery | direct; "" for the manual baseline
Plays int64 `json:"plays"` // window total, for sorting/volume-gating
Points []trendPoint `json:"points"`
}
// trendMarker is one tuning-audit event annotated on the timeline.
type trendMarker struct {
ChangedAt string `json:"changed_at"`
Scope string `json:"scope"`
Action string `json:"action"`
Changes json.RawMessage `json:"changes"`
}
type trendsResp struct {
Weeks int `json:"weeks"`
Series []trendSeries `json:"series"`
Markers []trendMarker `json:"markers"`
}
// handleGetRecommendationTrends implements
// GET /api/admin/recommendation-trends?weeks=N (default 12, cap 52).
func (h *handlers) handleGetRecommendationTrends(w http.ResponseWriter, r *http.Request) {
weeks := trendsDefaultWeeks
if v := r.URL.Query().Get("weeks"); v != "" {
n, err := strconv.Atoi(v)
if err != nil || n < 1 {
writeErr(w, apierror.BadRequest("bad_request", "invalid weeks"))
return
}
if n > trendsMaxWeeks {
n = trendsMaxWeeks
}
weeks = n
}
q := dbq.New(h.pool)
rows, err := q.RecommendationWeeklyTrends(r.Context(), int32(weeks))
if err != nil {
h.logger.Error("api: recommendation trends", "err", err)
writeErr(w, apierror.InternalMsg("lookup failed", err))
return
}
audits, err := q.ListTuningAudit(r.Context(), trendsAuditFetchCap)
if err != nil {
h.logger.Error("api: recommendation trends audit", "err", err)
writeErr(w, apierror.InternalMsg("lookup failed", err))
return
}
writeJSON(w, http.StatusOK, buildTrendsResponse(weeks, time.Now().UTC(), rows, audits))
}
// buildTrendsResponse folds weekly rows into per-family series and
// windows the audit markers. Split from the handler for pure-unit
// testability.
func buildTrendsResponse(
weeks int, now time.Time,
rows []dbq.RecommendationWeeklyTrendsRow,
audits []dbq.RecommendationTuningAudit,
) trendsResp {
type accum struct {
fam recFamily
plays int64
points []trendPoint
}
families := map[string]*accum{}
for _, row := range rows {
fam := recFamily{key: "manual", label: "Manual library plays"}
if row.Source != nil && *row.Source != "" {
fam = bucketRecSource(*row.Source)
}
acc, ok := families[fam.key]
if !ok {
acc = &accum{fam: fam}
families[fam.key] = acc
}
acc.plays += row.Plays
p := trendPoint{
WeekStart: row.WeekStart.Time.Format("2006-01-02"),
Plays: row.Plays,
Skips: row.Skips,
AvgCompletion: row.AvgCompletion,
completionN: row.CompletionN,
}
if row.Plays > 0 {
p.SkipRate = float64(row.Skips) / float64(row.Plays)
p.TasteHitRate = float64(row.TasteHits) / float64(row.Plays)
}
// Same family can arrive as several raw sources (radio:<uuid>);
// merge same-week points play-weighted.
if n := len(acc.points); n > 0 && acc.points[n-1].WeekStart == p.WeekStart {
acc.points[n-1] = mergeTrendPoints(acc.points[n-1], p)
} else {
acc.points = append(acc.points, p)
}
}
resp := trendsResp{Weeks: weeks, Series: []trendSeries{}, Markers: []trendMarker{}}
for _, acc := range families {
resp.Series = append(resp.Series, trendSeries{
Key: acc.fam.key,
Label: acc.fam.label,
Intent: acc.fam.intent,
Plays: acc.plays,
Points: acc.points,
})
}
sort.Slice(resp.Series, func(i, j int) bool {
if resp.Series[i].Plays != resp.Series[j].Plays {
return resp.Series[i].Plays > resp.Series[j].Plays
}
return resp.Series[i].Key < resp.Series[j].Key
})
cutoff := now.Add(-time.Duration(weeks) * 7 * 24 * time.Hour)
for _, a := range audits {
if a.ChangedAt.Time.Before(cutoff) {
continue
}
resp.Markers = append(resp.Markers, trendMarker{
ChangedAt: a.ChangedAt.Time.UTC().Format(time.RFC3339),
Scope: a.Scope,
Action: a.Action,
Changes: json.RawMessage(a.Changes),
})
}
// ListTuningAudit returns newest-first; the timeline reads better
// oldest-first.
sort.Slice(resp.Markers, func(i, j int) bool {
return resp.Markers[i].ChangedAt < resp.Markers[j].ChangedAt
})
return resp
}
// mergeTrendPoints combines two same-week points of one family:
// counts add, skip/taste rates re-derive from the merged counts, and
// avg_completion is weighted by each side's completion sample count.
func mergeTrendPoints(a, b trendPoint) trendPoint {
out := trendPoint{
WeekStart: a.WeekStart,
Plays: a.Plays + b.Plays,
Skips: a.Skips + b.Skips,
completionN: a.completionN + b.completionN,
}
if out.Plays > 0 {
out.SkipRate = float64(out.Skips) / float64(out.Plays)
out.TasteHitRate = (a.TasteHitRate*float64(a.Plays) + b.TasteHitRate*float64(b.Plays)) /
float64(out.Plays)
}
if out.completionN > 0 {
out.AvgCompletion = (a.AvgCompletion*float64(a.completionN) + b.AvgCompletion*float64(b.completionN)) /
float64(out.completionN)
}
return out
}
@@ -0,0 +1,152 @@
package api
import (
"context"
"encoding/json"
"net/http"
"net/http/httptest"
"testing"
"time"
"github.com/go-chi/chi/v5"
"github.com/jackc/pgx/v5/pgtype"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
)
func date(s string) pgtype.Date {
t, _ := time.Parse("2006-01-02", s)
return pgtype.Date{Time: t, Valid: true}
}
func TestBuildTrendsResponse_SeriesMergingAndMarkers(t *testing.T) {
src := func(s string) *string { return &s }
now := time.Date(2026, 7, 3, 12, 0, 0, 0, time.UTC)
rows := []dbq.RecommendationWeeklyTrendsRow{
// Two radio session sources in the same week collapse into one
// family point; completion weighting by sample count.
{WeekStart: date("2026-06-22"), Source: src("radio:aaaa"),
Plays: 3, Skips: 1, CompletionN: 3, AvgCompletion: 0.6, TasteHits: 3},
{WeekStart: date("2026-06-22"), Source: src("radio:bbbb"),
Plays: 1, Skips: 1, CompletionN: 1, AvgCompletion: 0.2, TasteHits: 0},
{WeekStart: date("2026-06-29"), Source: src("radio:aaaa"),
Plays: 2, Skips: 0, CompletionN: 2, AvgCompletion: 0.9, TasteHits: 1},
// NULL source = manual baseline family.
{WeekStart: date("2026-06-29"), Source: nil,
Plays: 5, Skips: 1, CompletionN: 5, AvgCompletion: 0.8, TasteHits: 4},
}
audits := []dbq.RecommendationTuningAudit{
{ID: 2, ChangedAt: pgtype.Timestamptz{Time: now.Add(-24 * time.Hour), Valid: true},
Scope: "radio", Action: "update", Changes: []byte(`[{"field":"taste_weight","old":1,"new":2}]`)},
// Older than the window → dropped.
{ID: 1, ChangedAt: pgtype.Timestamptz{Time: now.Add(-100 * 7 * 24 * time.Hour), Valid: true},
Scope: "taste", Action: "reset", Changes: []byte(`[]`)},
}
resp := buildTrendsResponse(12, now, rows, audits)
if len(resp.Series) != 2 {
t.Fatalf("series = %d, want 2 (radio + manual)", len(resp.Series))
}
radio := resp.Series[0] // 6 plays > manual's 5 → sorted first
if radio.Key != "radio" || radio.Plays != 6 {
t.Fatalf("series[0] = %s/%d, want radio/6", radio.Key, radio.Plays)
}
if len(radio.Points) != 2 {
t.Fatalf("radio points = %d, want 2 weeks", len(radio.Points))
}
wk1 := radio.Points[0]
if wk1.WeekStart != "2026-06-22" || wk1.Plays != 4 || wk1.Skips != 2 {
t.Errorf("week1 = %+v, want 2026-06-22 with 4 plays / 2 skips", wk1)
}
if wk1.SkipRate != 0.5 {
t.Errorf("week1 skip_rate = %v, want 0.5", wk1.SkipRate)
}
// Completion weighted by sample count: (0.6*3 + 0.2*1) / 4 = 0.5.
if wk1.AvgCompletion < 0.49 || wk1.AvgCompletion > 0.51 {
t.Errorf("week1 avg_completion = %v, want 0.5", wk1.AvgCompletion)
}
// Taste hits: 3 of 4 plays.
if wk1.TasteHitRate != 0.75 {
t.Errorf("week1 taste_hit_rate = %v, want 0.75", wk1.TasteHitRate)
}
if manual := resp.Series[1]; manual.Key != "manual" || manual.Intent != "" {
t.Errorf("series[1] = %+v, want the manual baseline family", manual)
}
if len(resp.Markers) != 1 {
t.Fatalf("markers = %d, want 1 (out-of-window marker dropped)", len(resp.Markers))
}
if resp.Markers[0].Scope != "radio" || resp.Markers[0].Action != "update" {
t.Errorf("marker = %+v, want radio/update", resp.Markers[0])
}
}
func newTrendsRouter(h *handlers) chi.Router {
r := chi.NewRouter()
r.Get("/api/admin/recommendation-trends", h.handleGetRecommendationTrends)
return r
}
func TestRecommendationTrends_EndToEnd(t *testing.T) {
h, pool := testHandlers(t)
user := seedUser(t, pool, "trends", "pw", true)
artist := seedArtist(t, pool, "TrendArtist")
album := seedAlbum(t, pool, artist.ID, "TrendAlbum", 2020)
tk := seedTrack(t, pool, album.ID, artist.ID, "TrendTrack", 1, 200000)
session := seedPlaySession(t, pool, user.ID, time.Now())
// Positive taste weight for the artist → plays count as taste hits.
if _, err := pool.Exec(context.Background(),
`INSERT INTO taste_profile_artists (user_id, artist_id, weight) VALUES ($1, $2, 1.5)`,
user.ID, artist.ID); err != nil {
t.Fatalf("seed taste profile: %v", err)
}
radio := "radio"
seedSourcedPlay(t, h, user.ID, tk.ID, session, &radio, nil, 0.9, false)
// A knob turn to mark the timeline.
if err := h.recSettings.UpdateProfile(context.Background(), "radio",
map[string]float64{"taste_weight": 2}); err != nil {
t.Fatalf("UpdateProfile: %v", err)
}
req := httptest.NewRequest(http.MethodGet, "/api/admin/recommendation-trends", nil)
rec := httptest.NewRecorder()
newTrendsRouter(h).ServeHTTP(rec, req)
if rec.Code != http.StatusOK {
t.Fatalf("status = %d, want 200 (%s)", rec.Code, rec.Body.String())
}
var resp trendsResp
if err := json.Unmarshal(rec.Body.Bytes(), &resp); err != nil {
t.Fatalf("decode: %v", err)
}
if resp.Weeks != trendsDefaultWeeks {
t.Errorf("weeks = %d, want %d", resp.Weeks, trendsDefaultWeeks)
}
var radioSeries *trendSeries
for i := range resp.Series {
if resp.Series[i].Key == "radio" {
radioSeries = &resp.Series[i]
}
}
if radioSeries == nil || len(radioSeries.Points) != 1 {
t.Fatalf("radio series = %+v, want one point", radioSeries)
}
if radioSeries.Points[0].TasteHitRate != 1.0 {
t.Errorf("taste_hit_rate = %v, want 1.0 (positive-weight artist)",
radioSeries.Points[0].TasteHitRate)
}
if len(resp.Markers) != 1 || resp.Markers[0].Scope != "radio" {
t.Errorf("markers = %+v, want the radio knob turn", resp.Markers)
}
}
func TestRecommendationTrends_InvalidWeeks(t *testing.T) {
h, _ := testHandlers(t)
req := httptest.NewRequest(http.MethodGet, "/api/admin/recommendation-trends?weeks=zero", nil)
rec := httptest.NewRecorder()
newTrendsRouter(h).ServeHTTP(rec, req)
if rec.Code != http.StatusBadRequest {
t.Errorf("status = %d, want 400", rec.Code)
}
}
+152
View File
@@ -0,0 +1,152 @@
// Admin recommendation-tuning endpoints (#1250): the defaults-
// discovery lab. GET returns current values + shipped defaults for
// every scope; PATCH applies a partial update to one scope; reset
// restores a scope to shipped defaults. Every change writes an audit
// row (consumed by the metrics trend view, #1251).
package api
import (
"encoding/json"
"errors"
"net/http"
"github.com/go-chi/chi/v5"
"git.fabledsword.com/bvandeusen/minstrel/internal/apierror"
"git.fabledsword.com/bvandeusen/minstrel/internal/recommendation"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
)
// weightsResp is one weight profile on the wire, keyed by the same
// snake_case field names the PATCH body accepts.
type weightsResp struct {
BaseWeight float64 `json:"base_weight"`
LikeBoost float64 `json:"like_boost"`
RecencyWeight float64 `json:"recency_weight"`
SkipPenalty float64 `json:"skip_penalty"`
JitterMagnitude float64 `json:"jitter_magnitude"`
ContextWeight float64 `json:"context_weight"`
SimilarityWeight float64 `json:"similarity_weight"`
TasteWeight float64 `json:"taste_weight"`
}
func weightsRespFrom(w recommendation.ScoringWeights) weightsResp {
return weightsResp{
BaseWeight: w.BaseWeight,
LikeBoost: w.LikeBoost,
RecencyWeight: w.RecencyWeight,
SkipPenalty: w.SkipPenalty,
JitterMagnitude: w.JitterMagnitude,
ContextWeight: w.ContextWeight,
SimilarityWeight: w.SimilarityWeight,
TasteWeight: w.TasteWeight,
}
}
type tasteTuningResp struct {
HalfLifeDays float64 `json:"half_life_days"`
EngagementHardSkip float64 `json:"engagement_hard_skip"`
EngagementNeutral float64 `json:"engagement_neutral"`
EngagementFull float64 `json:"engagement_full"`
}
func tasteRespFrom(t recsettings.TasteTuning) tasteTuningResp {
return tasteTuningResp{
HalfLifeDays: t.HalfLifeDays,
EngagementHardSkip: t.EngagementHardSkip,
EngagementNeutral: t.EngagementNeutral,
EngagementFull: t.EngagementFull,
}
}
// tuningSnapshot is both the GET response and the post-mutation echo:
// current values alongside shipped defaults so the card can mark
// which knobs deviate.
type tuningSnapshot struct {
Profiles map[string]weightsResp `json:"profiles"`
Taste tasteTuningResp `json:"taste"`
Shipped struct {
Profiles map[string]weightsResp `json:"profiles"`
Taste tasteTuningResp `json:"taste"`
} `json:"shipped"`
}
func (h *handlers) tuningSnapshot() tuningSnapshot {
var out tuningSnapshot
out.Profiles = map[string]weightsResp{
recsettings.ScopeRadio: weightsRespFrom(h.recSettings.Weights(recsettings.ScopeRadio)),
recsettings.ScopeDailyMix: weightsRespFrom(h.recSettings.Weights(recsettings.ScopeDailyMix)),
}
out.Taste = tasteRespFrom(h.recSettings.Taste())
out.Shipped.Profiles = map[string]weightsResp{
recsettings.ScopeRadio: weightsRespFrom(recsettings.ShippedRadioWeights()),
recsettings.ScopeDailyMix: weightsRespFrom(recsettings.ShippedDailyMixWeights()),
}
out.Shipped.Taste = tasteRespFrom(recsettings.ShippedTasteTuning())
return out
}
// handleGetRecommendationTuning implements GET /api/admin/recommendation-tuning.
func (h *handlers) handleGetRecommendationTuning(w http.ResponseWriter, _ *http.Request) {
writeJSON(w, http.StatusOK, h.tuningSnapshot())
}
// patchTuningReq carries the partial update: field name → new value,
// using the same snake_case names the GET response emits.
type patchTuningReq struct {
Values map[string]float64 `json:"values"`
}
// handlePatchRecommendationTuning implements
// PATCH /api/admin/recommendation-tuning/{scope}.
func (h *handlers) handlePatchRecommendationTuning(w http.ResponseWriter, r *http.Request) {
scope := chi.URLParam(r, "scope")
var body patchTuningReq
if err := json.NewDecoder(r.Body).Decode(&body); err != nil {
writeErr(w, apierror.BadRequest("bad_body", "invalid JSON"))
return
}
if len(body.Values) == 0 {
writeErr(w, apierror.BadRequest("bad_body", "values is empty"))
return
}
var err error
if scope == recsettings.ScopeTaste {
err = h.recSettings.UpdateTaste(r.Context(), body.Values)
} else {
err = h.recSettings.UpdateProfile(r.Context(), scope, body.Values)
}
if err != nil {
writeTuningErr(w, h, scope, err)
return
}
writeJSON(w, http.StatusOK, h.tuningSnapshot())
}
// handleResetRecommendationTuning implements
// POST /api/admin/recommendation-tuning/{scope}/reset.
func (h *handlers) handleResetRecommendationTuning(w http.ResponseWriter, r *http.Request) {
scope := chi.URLParam(r, "scope")
if err := h.recSettings.Reset(r.Context(), scope); err != nil {
writeTuningErr(w, h, scope, err)
return
}
writeJSON(w, http.StatusOK, h.tuningSnapshot())
}
// writeTuningErr maps recsettings validation errors to 400s and
// everything else to a logged 500.
func writeTuningErr(w http.ResponseWriter, h *handlers, scope string, err error) {
switch {
case errors.Is(err, recsettings.ErrUnknownScope):
writeErr(w, &apierror.Error{
Status: http.StatusNotFound, Code: "not_found", Message: "no such tuning scope",
})
case errors.Is(err, recsettings.ErrUnknownField), errors.Is(err, recsettings.ErrOutOfRange):
writeErr(w, apierror.BadRequest("invalid_tuning", err.Error()))
default:
h.logger.Error("admin: recommendation tuning", "scope", scope, "err", err)
writeErr(w, apierror.InternalMsg("tuning update failed", err))
}
}
@@ -0,0 +1,110 @@
package api
import (
"encoding/json"
"net/http"
"net/http/httptest"
"strings"
"testing"
"github.com/go-chi/chi/v5"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
)
func newTuningRouter(h *handlers) chi.Router {
r := chi.NewRouter()
r.Get("/api/admin/recommendation-tuning", h.handleGetRecommendationTuning)
r.Patch("/api/admin/recommendation-tuning/{scope}", h.handlePatchRecommendationTuning)
r.Post("/api/admin/recommendation-tuning/{scope}/reset", h.handleResetRecommendationTuning)
return r
}
func decodeTuning(t *testing.T, rec *httptest.ResponseRecorder) tuningSnapshot {
t.Helper()
var snap tuningSnapshot
if err := json.Unmarshal(rec.Body.Bytes(), &snap); err != nil {
t.Fatalf("decode: %v", err)
}
return snap
}
func TestRecommendationTuning_GetReturnsShippedDefaults(t *testing.T) {
h, _ := testHandlers(t)
req := httptest.NewRequest(http.MethodGet, "/api/admin/recommendation-tuning", nil)
rec := httptest.NewRecorder()
newTuningRouter(h).ServeHTTP(rec, req)
if rec.Code != http.StatusOK {
t.Fatalf("status = %d, want 200", rec.Code)
}
snap := decodeTuning(t, rec)
if snap.Profiles["radio"].TasteWeight != 1.0 || snap.Profiles["daily_mix"].TasteWeight != 1.5 {
t.Errorf("profiles = %+v, want shipped taste weights 1.0 / 1.5", snap.Profiles)
}
if snap.Taste.HalfLifeDays != 75 {
t.Errorf("taste half-life = %v, want shipped 75", snap.Taste.HalfLifeDays)
}
if snap.Shipped.Profiles["radio"] != snap.Profiles["radio"] {
t.Error("untouched values must equal shipped defaults")
}
}
func TestRecommendationTuning_PatchAndReset(t *testing.T) {
h, _ := testHandlers(t)
r := newTuningRouter(h)
req := httptest.NewRequest(http.MethodPatch, "/api/admin/recommendation-tuning/radio",
strings.NewReader(`{"values":{"taste_weight": 2.5}}`))
rec := httptest.NewRecorder()
r.ServeHTTP(rec, req)
if rec.Code != http.StatusOK {
t.Fatalf("patch status = %d, want 200 (%s)", rec.Code, rec.Body.String())
}
snap := decodeTuning(t, rec)
if snap.Profiles["radio"].TasteWeight != 2.5 {
t.Errorf("patched taste_weight = %v, want 2.5", snap.Profiles["radio"].TasteWeight)
}
// The change is live for the radio scoring path.
if got := h.recSettings.Weights(recsettings.ScopeRadio).TasteWeight; got != 2.5 {
t.Errorf("service taste_weight = %v, want 2.5 (live effect)", got)
}
req = httptest.NewRequest(http.MethodPost, "/api/admin/recommendation-tuning/radio/reset", nil)
rec = httptest.NewRecorder()
r.ServeHTTP(rec, req)
if rec.Code != http.StatusOK {
t.Fatalf("reset status = %d, want 200", rec.Code)
}
snap = decodeTuning(t, rec)
if snap.Profiles["radio"].TasteWeight != 1.0 {
t.Errorf("reset taste_weight = %v, want shipped 1.0", snap.Profiles["radio"].TasteWeight)
}
}
func TestRecommendationTuning_PatchErrors(t *testing.T) {
h, _ := testHandlers(t)
r := newTuningRouter(h)
cases := []struct {
name, path, body string
want int
}{
{"unknown scope", "/api/admin/recommendation-tuning/banana",
`{"values":{"taste_weight":1}}`, http.StatusNotFound},
{"unknown field", "/api/admin/recommendation-tuning/radio",
`{"values":{"vibes":1}}`, http.StatusBadRequest},
{"out of range", "/api/admin/recommendation-tuning/taste",
`{"values":{"engagement_neutral":2}}`, http.StatusBadRequest},
{"empty values", "/api/admin/recommendation-tuning/radio",
`{"values":{}}`, http.StatusBadRequest},
{"bad json", "/api/admin/recommendation-tuning/radio",
`{`, http.StatusBadRequest},
}
for _, c := range cases {
req := httptest.NewRequest(http.MethodPatch, c.path, strings.NewReader(c.body))
rec := httptest.NewRecorder()
r.ServeHTTP(rec, req)
if rec.Code != c.want {
t.Errorf("%s: status = %d, want %d", c.name, rec.Code, c.want)
}
}
}
+12 -1
View File
@@ -22,16 +22,18 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/mailer"
"git.fabledsword.com/bvandeusen/minstrel/internal/playevents"
"git.fabledsword.com/bvandeusen/minstrel/internal/playlists"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
"git.fabledsword.com/bvandeusen/minstrel/internal/tracks"
)
// Mount attaches /api/* handlers to r. Public endpoints (login) are outside
// RequireUser; everything else is gated by the middleware. The events writer
// is shared with the Subsonic mount so /rest/scrobble feeds the same store.
func Mount(r chi.Router, pool *pgxpool.Pool, logger *slog.Logger, events *playevents.Writer, recCfg config.RecommendationConfig, lidarrCfg *lidarrconfig.Service, lidarrReqs *lidarrrequests.Service, lidarrQuar *lidarrquarantine.Service, tracksSvc *tracks.Service, playlistsSvc *playlists.Service, coverEnricher *coverart.Enricher, coverSettings *coverart.SettingsService, scanner *library.Scanner, scanCfg library.RunScanConfig, dataDir string, sender mailer.Sender, bus *eventbus.Bus, playlistScheduler *playlists.Scheduler, streamSecret []byte) {
func Mount(r chi.Router, pool *pgxpool.Pool, logger *slog.Logger, events *playevents.Writer, recCfg config.RecommendationConfig, recSettings *recsettings.Service, lidarrCfg *lidarrconfig.Service, lidarrReqs *lidarrrequests.Service, lidarrQuar *lidarrquarantine.Service, tracksSvc *tracks.Service, playlistsSvc *playlists.Service, coverEnricher *coverart.Enricher, coverSettings *coverart.SettingsService, scanner *library.Scanner, scanCfg library.RunScanConfig, dataDir string, sender mailer.Sender, bus *eventbus.Bus, playlistScheduler *playlists.Scheduler, streamSecret []byte) {
rng := rand.New(rand.NewSource(rand.Int63()))
h := &handlers{
pool: pool, logger: logger, events: events, recCfg: recCfg,
recSettings: recSettings,
rng: rng.Float64,
lidarrCfg: lidarrCfg,
lidarrRequests: lidarrReqs,
@@ -201,6 +203,14 @@ func Mount(r chi.Router, pool *pgxpool.Pool, logger *slog.Logger, events *playev
admin.Get("/smtp-config", h.handleGetSMTPConfig)
admin.Put("/smtp-config", h.handleUpdateSMTPConfig)
admin.Post("/smtp-config/test", h.handleTestSMTPConfig)
// Recommendation tuning lab (#1250): scoring-weight
// profiles + taste-build knobs, DB-backed, live effect.
admin.Get("/recommendation-tuning", h.handleGetRecommendationTuning)
admin.Patch("/recommendation-tuning/{scope}", h.handlePatchRecommendationTuning)
admin.Post("/recommendation-tuning/{scope}/reset", h.handleResetRecommendationTuning)
// Weekly outcome trends + knob-turn markers (#1251).
admin.Get("/recommendation-trends", h.handleGetRecommendationTrends)
})
authed.Get("/playlists", h.handleListPlaylists)
@@ -223,6 +233,7 @@ type handlers struct {
logger *slog.Logger
events *playevents.Writer
recCfg config.RecommendationConfig
recSettings *recsettings.Service
rng func() float64
lidarrCfg *lidarrconfig.Service
lidarrRequests *lidarrrequests.Service
+6 -4
View File
@@ -29,6 +29,7 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/mailer"
"git.fabledsword.com/bvandeusen/minstrel/internal/playevents"
"git.fabledsword.com/bvandeusen/minstrel/internal/playlists"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
"git.fabledsword.com/bvandeusen/minstrel/internal/tracks"
)
@@ -56,11 +57,12 @@ func testHandlers(t *testing.T) (*handlers, *pgxpool.Pool) {
dbtest.ResetDB(t, pool)
w := playevents.NewWriter(pool, logger, 30*time.Minute, 0.5, 30000)
recCfg := config.RecommendationConfig{
BaseWeight: 1.0, LikeBoost: 2.0, RecencyWeight: 1.0,
SkipPenalty: 1.0, JitterMagnitude: 0.1,
ContextWeight: 2.0, SimilarityWeight: 2.0,
RecentlyPlayedHours: 1, RadioSize: 50, RadioSizeMax: 200,
}
recSettings, err := recsettings.New(context.Background(), pool, logger)
if err != nil {
t.Fatalf("recsettings: %v", err)
}
lidarrCfg := lidarrconfig.New(pool)
lidarrReqs := lidarrrequests.NewService(pool, lidarrCfg, nil, nil)
lidarrQuar := lidarrquarantine.NewService(pool, lidarrCfg, nil)
@@ -70,7 +72,7 @@ func testHandlers(t *testing.T) (*handlers, *pgxpool.Pool) {
dataDir := t.TempDir()
tracksSvc := tracks.NewService(pool, logger, nil, dataDir)
playlistsSvc := playlists.NewService(pool, logger, dataDir)
h := &handlers{pool: pool, logger: logger, events: w, recCfg: recCfg, rng: func() float64 { return 0.5 }, lidarrCfg: lidarrCfg, lidarrRequests: lidarrReqs, lidarrQuarantine: lidarrQuar, tracks: tracksSvc, playlists: playlistsSvc, dataDir: dataDir, scanner: nil, scanCfg: library.RunScanConfig{}, mailer: &mailer.FakeSender{}}
h := &handlers{pool: pool, logger: logger, events: w, recCfg: recCfg, recSettings: recSettings, rng: func() float64 { return 0.5 }, lidarrCfg: lidarrCfg, lidarrRequests: lidarrReqs, lidarrQuarantine: lidarrQuar, tracks: tracksSvc, playlists: playlistsSvc, dataDir: dataDir, scanner: nil, scanCfg: library.RunScanConfig{}, mailer: &mailer.FakeSender{}}
return h, pool
}
+21 -1
View File
@@ -31,6 +31,26 @@ type playStartedResponse struct {
SessionID string `json:"session_id"`
}
// eventFutureSkew is the tolerated client-clock drift into the future
// before a timestamp is treated as bogus and replaced with now.
const eventFutureSkew = 5 * time.Minute
// clampEventTime bounds a client-supplied event timestamp to sanity:
// no earlier than the account's creation (offline replays can
// legitimately be days old, but no play can predate the user) and no
// later than now + a small skew allowance. Unbounded client clocks
// previously let a skewed device write arbitrarily old plays, which
// poisoned Rediscover's "not played in 6 months" ordering (#1246).
func clampEventTime(at, userCreatedAt, now time.Time) time.Time {
if at.Before(userCreatedAt) {
return userCreatedAt
}
if at.After(now.Add(eventFutureSkew)) {
return now
}
return at
}
type okResponse struct {
OK bool `json:"ok"`
}
@@ -52,7 +72,7 @@ func (h *handlers) handleEvents(w http.ResponseWriter, r *http.Request) {
writeErr(w, apierror.BadRequest("bad_request", "invalid `at` timestamp"))
return
}
at = parsed
at = clampEventTime(parsed.UTC(), user.CreatedAt.Time, time.Now().UTC())
}
clientID := ""
if req.ClientID != nil {
+1 -1
View File
@@ -465,7 +465,7 @@ func TestRoutesRegisteredInMount(t *testing.T) {
r := chi.NewRouter()
w := playevents.NewWriter(h.pool, slog.New(slog.NewTextHandler(io.Discard, nil)),
30*time.Minute, 0.5, 30000)
Mount(r, h.pool, h.logger, w, config.RecommendationConfig{RadioSize: 50, RadioSizeMax: 200, RecentlyPlayedHours: 1}, h.lidarrCfg, h.lidarrRequests, h.lidarrQuarantine, h.tracks, h.playlists, h.coverart, h.coverSettings, h.scanner, h.scanCfg, h.dataDir, nil, eventbus.New(), nil, nil)
Mount(r, h.pool, h.logger, w, config.RecommendationConfig{RadioSize: 50, RadioSizeMax: 200, RecentlyPlayedHours: 1}, h.recSettings, h.lidarrCfg, h.lidarrRequests, h.lidarrQuarantine, h.tracks, h.playlists, h.coverart, h.coverSettings, h.scanner, h.scanCfg, h.dataDir, nil, eventbus.New(), nil, nil)
paths := []string{
"/api/artists",
+255 -26
View File
@@ -2,7 +2,9 @@ package api
import (
"net/http"
"sort"
"strconv"
"strings"
"git.fabledsword.com/bvandeusen/minstrel/internal/apierror"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
@@ -11,28 +13,142 @@ import (
const (
recMetricsDefaultDays = 30
recMetricsMaxDays = 365
// recMetricsLowVolume marks a family as low-confidence rather than
// hiding it: with fewer plays than this a skip rate is anecdote, not
// signal, but silently dropping the row would misread as "surface
// unused". The web renders low-confidence rows dimmed.
recMetricsLowVolume = 20
)
// recommendationMetric is one recommendation surface's outcomes.
type recommendationMetric struct {
Source string `json:"source"` // 'for_you' | 'discover' | mixes
Plays int64 `json:"plays"` // plays launched from this surface
// Surface intents (milestone #127): each family is judged against its
// job, not one global bar — discovery mixes are EXPECTED to run higher
// skip rates than the go-to surfaces.
const (
intentGoTo = "go_to"
intentDiscovery = "discovery"
intentDirect = "direct"
)
// surfaceMetric is one bucketed surface family's outcomes.
type surfaceMetric struct {
Key string `json:"key"` // stable family key ("for_you", "radio", …)
Label string `json:"label"` // display label
Plays int64 `json:"plays"` // plays launched from this family
Skips int64 `json:"skips"` // of those, marked skipped
SkipRate float64 `json:"skip_rate"` // skips / plays, [0,1]
AvgCompletion float64 `json:"avg_completion"` // mean completion ratio, [0,1]
LowConfidence bool `json:"low_confidence"` // plays < recMetricsLowVolume
// Breakdown splits the family into the pick-kind populations its
// builder stamped (#1249, generalized #1270): For You's taste/fresh,
// Discover's buckets, tier1-3 for tiered mixes — plus earlier plays
// that predate attribution. Present only when the family has at
// least one attributed play; the parent row remains the sum of its
// breakdown.
Breakdown []surfaceMetric `json:"breakdown,omitempty"`
}
// surfaceGroup is one intent band of surface families.
type surfaceGroup struct {
Intent string `json:"intent"` // go_to | discovery | direct
Label string `json:"label"`
Surfaces []surfaceMetric `json:"surfaces"`
}
type recommendationMetricsResp struct {
WindowDays int `json:"window_days"`
Sources []recommendationMetric `json:"sources"`
// Baseline is the control group: plays the user picked manually
// (source IS NULL). Surfaces are judged as deltas against it; nil
// when the window holds no manual plays.
Baseline *surfaceMetric `json:"baseline"`
Groups []surfaceGroup `json:"groups"`
}
// recFamily is the bucketing target for a raw play_events.source value.
type recFamily struct {
key string
label string
intent string
}
// bucketRecSource maps a raw client-stamped source string to its stable
// family. One-off sources (album:<uuid>, radio:<uuid>) collapse into
// their family so the table stays readable at any library size.
func bucketRecSource(src string) recFamily {
switch {
case src == "for_you":
return recFamily{"for_you", "For You", intentGoTo}
case src == "songs_like_artist":
return recFamily{"songs_like_artist", "Songs like…", intentGoTo}
case src == "radio" || strings.HasPrefix(src, "radio:"):
return recFamily{"radio", "Radio", intentGoTo}
case src == "discover":
return recFamily{"discover", "Discover", intentDiscovery}
case src == "deep_cuts":
return recFamily{"deep_cuts", "Deep cuts", intentDiscovery}
case src == "rediscover":
return recFamily{"rediscover", "Rediscover", intentDiscovery}
case src == "new_for_you":
return recFamily{"new_for_you", "New for you", intentDiscovery}
case src == "on_this_day":
return recFamily{"on_this_day", "On this day", intentDiscovery}
case src == "first_listens":
return recFamily{"first_listens", "First listens", intentDiscovery}
case strings.HasPrefix(src, "album:"):
return recFamily{"direct_album", "Album plays", intentDirect}
case strings.HasPrefix(src, "artist:"):
return recFamily{"direct_artist", "Artist plays", intentDirect}
case strings.HasPrefix(src, "offline:"):
return recFamily{"offline", "Offline pools", intentDirect}
case strings.HasPrefix(src, "home:"):
return recFamily{"home", "Home sections", intentDirect}
case src == "history":
return recFamily{"history", "History", intentDirect}
default:
return recFamily{"other", "Other", intentDirect}
}
}
// familyAccum merges raw source rows into one family, carrying the
// completion sample count so the merged average stays play-weighted.
type familyAccum struct {
fam recFamily
plays int64
skips int64
completionN int64
// completionSum is avg*count re-expanded, so merging N raw rows
// reduces to a single weighted division at the end.
completionSum float64
}
func (a *familyAccum) add(row dbq.RecommendationSourceMetricsForUserRow) {
a.plays += row.Plays
a.skips += row.Skips
a.completionN += row.CompletionN
a.completionSum += row.AvgCompletion * float64(row.CompletionN)
}
func (a *familyAccum) metric() surfaceMetric {
m := surfaceMetric{
Key: a.fam.key,
Label: a.fam.label,
Plays: a.plays,
Skips: a.skips,
LowConfidence: a.plays < recMetricsLowVolume,
}
if a.plays > 0 {
m.SkipRate = float64(a.skips) / float64(a.plays)
}
if a.completionN > 0 {
m.AvgCompletion = a.completionSum / float64(a.completionN)
}
return m
}
// handleGetRecommendationMetrics implements GET /api/me/recommendation-metrics.
// Per-source play outcomes (plays / skips / skip-rate / avg-completion) for the
// caller over the last `days` (default 30, capped at 365), so the operator can
// see which recommendation surfaces are landing and tune the taste weights.
// Only plays tagged with a system-playlist source count; library/radio plays
// (no source) are excluded.
// Bucketed per-surface-family outcomes for the caller over the last `days`
// (default 30, capped at 365), grouped by surface intent and anchored by the
// manual-plays baseline so the numbers are judgeable, not just observable.
func (h *handlers) handleGetRecommendationMetrics(w http.ResponseWriter, r *http.Request) {
caller, ok := requireUser(w, r)
if !ok {
@@ -51,28 +167,141 @@ func (h *handlers) handleGetRecommendationMetrics(w http.ResponseWriter, r *http
return
}
out := recommendationMetricsResp{
WindowDays: days,
Sources: make([]recommendationMetric, 0, len(rows)),
writeJSON(w, http.StatusOK, bucketMetricsResponse(days, rows))
}
// pickKindLabels is the display vocabulary for play_events.pick_kind
// values (mirrors the CHECK in migration 0039). Every system mix that
// stamps provenance gets its breakdown from this one map — adding a
// stamping mix needs no metrics change.
var pickKindLabels = map[string]string{
"taste": "Taste picks",
"fresh": "Fresh picks",
"dormant": "Dormant artists",
"cross_user": "Liked by others",
"random": "Random unheard",
"tier1": "Tier 1 (exact)",
"tier2": "Tier 2 (relaxed)",
"tier3": "Tier 3 (stretched)",
}
// pickKindOrder fixes breakdown row order; unattributed ("", i.e. NULL
// pick_kind — plays recorded before the mix stamped provenance, or
// whose track had rotated out of the snapshot at ingestion) renders
// last, kept visible so the parent row's sums stay transparent instead
// of silently shrinking. The DB CHECK gates pick_kind to exactly this
// vocabulary, so iterating the list is exhaustive.
var pickKindOrder = []string{
"taste", "fresh", "dormant", "cross_user", "random",
"tier1", "tier2", "tier3", "",
}
// pickKindFamily derives the sub-family for one (family, pick_kind)
// population, e.g. ("for_you", "taste") → for_you_taste "Taste picks".
func pickKindFamily(parent recFamily, kind string) recFamily {
if kind == "" {
return recFamily{parent.key + "_unattributed", "Earlier plays", parent.intent}
}
label, ok := pickKindLabels[kind]
if !ok {
label = kind
}
return recFamily{parent.key + "_" + kind, label, parent.intent}
}
// pickKindBreakdown folds a family's per-pick-kind accums into its
// Breakdown rows. Attached only when at least one attributed play
// exists — an all-unattributed breakdown would just repeat the parent
// row, and families that never stamp (radio, direct plays) stay flat.
func pickKindBreakdown(picks map[string]*familyAccum) []surfaceMetric {
attributed := int64(0)
for kind, acc := range picks {
if kind != "" {
attributed += acc.plays
}
}
if attributed == 0 {
return nil
}
out := make([]surfaceMetric, 0, len(picks))
for _, kind := range pickKindOrder {
if acc, ok := picks[kind]; ok && acc.plays > 0 {
out = append(out, acc.metric())
}
}
return out
}
// bucketMetricsResponse folds the raw per-source rows into the grouped,
// baseline-anchored response shape. Split from the handler for pure-unit
// testability.
func bucketMetricsResponse(
days int, rows []dbq.RecommendationSourceMetricsForUserRow,
) recommendationMetricsResp {
baseline := &familyAccum{fam: recFamily{"manual", "Manual library plays", ""}}
families := map[string]*familyAccum{}
picks := map[string]map[string]*familyAccum{}
for _, row := range rows {
source := ""
if row.Source != nil {
source = *row.Source
if row.Source == nil || *row.Source == "" {
baseline.add(row)
continue
}
var skipRate float64
if row.Plays > 0 {
skipRate = float64(row.Skips) / float64(row.Plays)
fam := bucketRecSource(*row.Source)
acc, exists := families[fam.key]
if !exists {
acc = &familyAccum{fam: fam}
families[fam.key] = acc
}
out.Sources = append(out.Sources, recommendationMetric{
Source: source,
Plays: row.Plays,
Skips: row.Skips,
SkipRate: skipRate,
AvgCompletion: row.AvgCompletion,
acc.add(row)
// Accumulate the pick-kind population unconditionally; families
// that never stamp end up all-unattributed and get no breakdown.
kind := ""
if row.PickKind != nil {
kind = *row.PickKind
}
byKind, ok := picks[fam.key]
if !ok {
byKind = map[string]*familyAccum{}
picks[fam.key] = byKind
}
pick, ok := byKind[kind]
if !ok {
pick = &familyAccum{fam: pickKindFamily(fam, kind)}
byKind[kind] = pick
}
pick.add(row)
}
resp := recommendationMetricsResp{WindowDays: days, Groups: []surfaceGroup{}}
if baseline.plays > 0 {
m := baseline.metric()
resp.Baseline = &m
}
for _, g := range []struct{ intent, label string }{
{intentGoTo, "Go-to surfaces"},
{intentDiscovery, "Discovery mixes"},
{intentDirect, "Direct plays"},
} {
group := surfaceGroup{Intent: g.intent, Label: g.label}
for _, acc := range families {
if acc.fam.intent == g.intent {
m := acc.metric()
m.Breakdown = pickKindBreakdown(picks[acc.fam.key])
group.Surfaces = append(group.Surfaces, m)
}
}
if len(group.Surfaces) == 0 {
continue
}
sort.Slice(group.Surfaces, func(i, j int) bool {
if group.Surfaces[i].Plays != group.Surfaces[j].Plays {
return group.Surfaces[i].Plays > group.Surfaces[j].Plays
}
return group.Surfaces[i].Key < group.Surfaces[j].Key
})
resp.Groups = append(resp.Groups, group)
}
writeJSON(w, http.StatusOK, out)
return resp
}
// parseMetricsDays reads the `days` query param (default 30, capped at 365).
+197 -24
View File
@@ -11,6 +11,8 @@ import (
"github.com/go-chi/chi/v5"
"github.com/jackc/pgx/v5/pgtype"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
)
func newMetricsRouter(h *handlers) chi.Router {
@@ -20,17 +22,19 @@ func newMetricsRouter(h *handlers) chi.Router {
}
// seedSourcedPlay inserts a play_event with an explicit source + completion +
// skip flag. A nil source inserts NULL (library/radio play).
// skip flag. A nil source inserts NULL (manual library play → baseline).
// pickKind is the For You taste/fresh attribution (#1249); nil everywhere
// except attributed for_you plays.
func seedSourcedPlay(
t *testing.T, h *handlers, userID, trackID, sessionID pgtype.UUID,
source *string, completion float64, skipped bool,
source, pickKind *string, completion float64, skipped bool,
) {
t.Helper()
if _, err := h.pool.Exec(context.Background(),
`INSERT INTO play_events
(user_id, track_id, session_id, started_at, source, completion_ratio, was_skipped)
VALUES ($1, $2, $3, now(), $4, $5, $6)`,
userID, trackID, sessionID, source, completion, skipped); err != nil {
(user_id, track_id, session_id, started_at, source, pick_kind, completion_ratio, was_skipped)
VALUES ($1, $2, $3, now(), $4, $5, $6, $7)`,
userID, trackID, sessionID, source, pickKind, completion, skipped); err != nil {
t.Fatalf("seed sourced play: %v", err)
}
}
@@ -45,7 +49,168 @@ func TestRecommendationMetrics_NoSession401(t *testing.T) {
}
}
func TestRecommendationMetrics_AggregatesBySourceExcludingNull(t *testing.T) {
// findSurface returns the named family from any group, or nil.
func findSurface(resp recommendationMetricsResp, key string) *surfaceMetric {
for _, g := range resp.Groups {
for i := range g.Surfaces {
if g.Surfaces[i].Key == key {
return &g.Surfaces[i]
}
}
}
return nil
}
func TestBucketMetricsResponse_FamiliesGroupsBaseline(t *testing.T) {
src := func(s string) *string { return &s }
rows := []dbq.RecommendationSourceMetricsForUserRow{
{Source: src("for_you"), Plays: 3, Skips: 1, CompletionN: 3, AvgCompletion: 2.0 / 3},
// Two radio sessions collapse into one "radio" family; weighted
// completion = (0.2*1 + 0.8*1) / 2 = 0.5.
{Source: src("radio:aaaa"), Plays: 1, Skips: 1, CompletionN: 1, AvgCompletion: 0.2},
{Source: src("radio:bbbb"), Plays: 1, Skips: 0, CompletionN: 1, AvgCompletion: 0.8},
{Source: src("album:cccc"), Plays: 1, Skips: 0, CompletionN: 0, AvgCompletion: 0},
{Source: src("discover"), Plays: 1, Skips: 0, CompletionN: 1, AvgCompletion: 0.8},
// NULL source = manual plays → baseline, not a group row.
{Source: nil, Plays: 25, Skips: 5, CompletionN: 20, AvgCompletion: 0.9},
}
resp := bucketMetricsResponse(recMetricsDefaultDays, rows)
if resp.Baseline == nil {
t.Fatal("baseline missing")
}
if resp.Baseline.Plays != 25 || resp.Baseline.SkipRate != 0.2 {
t.Errorf("baseline = %+v, want plays=25 skip_rate=0.2", resp.Baseline)
}
if resp.Baseline.LowConfidence {
t.Error("baseline with 25 plays should not be low-confidence")
}
radio := findSurface(resp, "radio")
if radio == nil {
t.Fatal("radio family missing")
}
if radio.Plays != 2 || radio.Skips != 1 {
t.Errorf("radio plays/skips = %d/%d, want 2/1", radio.Plays, radio.Skips)
}
if radio.AvgCompletion < 0.49 || radio.AvgCompletion > 0.51 {
t.Errorf("radio avg_completion = %.3f, want 0.5 (play-weighted merge)", radio.AvgCompletion)
}
if !radio.LowConfidence {
t.Error("radio with 2 plays should be low-confidence")
}
if s := findSurface(resp, "direct_album"); s == nil || s.Plays != 1 {
t.Errorf("direct_album = %+v, want plays=1", s)
}
if s := findSurface(resp, ""); s != nil {
t.Error("NULL source must not appear as a surface family")
}
// No attributed (taste/fresh) plays in this fixture → no breakdown;
// an all-unattributed breakdown would just repeat the parent row.
if fy := findSurface(resp, "for_you"); fy == nil || fy.Breakdown != nil {
t.Errorf("for_you breakdown = %+v, want nil without attributed plays", fy)
}
// Group ordering is intent-banded: go_to before discovery before direct.
wantOrder := []string{intentGoTo, intentDiscovery, intentDirect}
if len(resp.Groups) != len(wantOrder) {
t.Fatalf("groups = %d, want %d", len(resp.Groups), len(wantOrder))
}
for i, g := range resp.Groups {
if g.Intent != wantOrder[i] {
t.Errorf("group[%d].intent = %s, want %s", i, g.Intent, wantOrder[i])
}
}
}
func TestBucketMetricsResponse_ForYouBreakdown(t *testing.T) {
src := func(s string) *string { return &s }
kind := func(s string) *string { return &s }
rows := []dbq.RecommendationSourceMetricsForUserRow{
{Source: src("for_you"), PickKind: kind("taste"), Plays: 30, Skips: 3,
CompletionN: 30, AvgCompletion: 0.9},
{Source: src("for_you"), PickKind: kind("fresh"), Plays: 10, Skips: 4,
CompletionN: 10, AvgCompletion: 0.5},
// NULL pick_kind = plays that predate attribution.
{Source: src("for_you"), Plays: 5, Skips: 1, CompletionN: 5, AvgCompletion: 0.7},
}
resp := bucketMetricsResponse(recMetricsDefaultDays, rows)
fy := findSurface(resp, "for_you")
if fy == nil {
t.Fatal("for_you family missing")
}
// The parent row stays the sum of its breakdown.
if fy.Plays != 45 || fy.Skips != 8 {
t.Errorf("for_you plays/skips = %d/%d, want 45/8", fy.Plays, fy.Skips)
}
if len(fy.Breakdown) != 3 {
t.Fatalf("breakdown rows = %d, want 3 (taste, fresh, earlier)", len(fy.Breakdown))
}
wantKeys := []string{"for_you_taste", "for_you_fresh", "for_you_unattributed"}
for i, k := range wantKeys {
if fy.Breakdown[i].Key != k {
t.Errorf("breakdown[%d].key = %s, want %s", i, fy.Breakdown[i].Key, k)
}
}
taste, fresh := fy.Breakdown[0], fy.Breakdown[1]
if taste.Plays != 30 || taste.SkipRate != 0.1 || taste.LowConfidence {
t.Errorf("taste = %+v, want plays=30 skip_rate=0.1 confident", taste)
}
if fresh.Plays != 10 || fresh.SkipRate != 0.4 || !fresh.LowConfidence {
t.Errorf("fresh = %+v, want plays=10 skip_rate=0.4 low-confidence", fresh)
}
// Breakdown rows never appear as their own surface families.
if s := findSurface(resp, "for_you_taste"); s != nil {
t.Error("for_you_taste must not be a top-level surface")
}
}
func TestBucketMetricsResponse_DiscoverBucketBreakdown(t *testing.T) {
// Provenance is standard (#1270): Discover's bucket stamps surface as
// a breakdown exactly like For You's taste/fresh — this is what makes
// the 40/30/30 allocation judgeable instead of a guess.
src := func(s string) *string { return &s }
kind := func(s string) *string { return &s }
rows := []dbq.RecommendationSourceMetricsForUserRow{
{Source: src("discover"), PickKind: kind("dormant"), Plays: 8, Skips: 2,
CompletionN: 8, AvgCompletion: 0.8},
{Source: src("discover"), PickKind: kind("cross_user"), Plays: 6, Skips: 3,
CompletionN: 6, AvgCompletion: 0.6},
{Source: src("discover"), PickKind: kind("random"), Plays: 4, Skips: 3,
CompletionN: 4, AvgCompletion: 0.4},
// NULL pick_kind = plays that predate bucket stamping.
{Source: src("discover"), Plays: 2, Skips: 0, CompletionN: 2, AvgCompletion: 0.9},
}
resp := bucketMetricsResponse(recMetricsDefaultDays, rows)
d := findSurface(resp, "discover")
if d == nil {
t.Fatal("discover family missing")
}
if d.Plays != 20 || d.Skips != 8 {
t.Errorf("discover plays/skips = %d/%d, want 20/8", d.Plays, d.Skips)
}
wantKeys := []string{
"discover_dormant", "discover_cross_user", "discover_random",
"discover_unattributed",
}
if len(d.Breakdown) != len(wantKeys) {
t.Fatalf("breakdown rows = %d, want %d", len(d.Breakdown), len(wantKeys))
}
for i, k := range wantKeys {
if d.Breakdown[i].Key != k {
t.Errorf("breakdown[%d].key = %s, want %s", i, d.Breakdown[i].Key, k)
}
}
if b := d.Breakdown[0]; b.Label != "Dormant artists" || b.Plays != 8 {
t.Errorf("dormant row = %+v, want label=Dormant artists plays=8", b)
}
}
func TestRecommendationMetrics_BucketsWithBaseline(t *testing.T) {
if os.Getenv("MINSTREL_TEST_DATABASE_URL") == "" {
t.Skip("MINSTREL_TEST_DATABASE_URL not set")
}
@@ -57,15 +222,17 @@ func TestRecommendationMetrics_AggregatesBySourceExcludingNull(t *testing.T) {
session := seedPlaySession(t, pool, user.ID, time.Now())
forYou := "for_you"
discover := "discover"
radioA := "radio:11111111-1111-1111-1111-111111111111"
taste, fresh := "taste", "fresh"
// for_you: 3 plays, 1 skipped; completions 1.0, 0.95, 0.05 → mean 0.6667.
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, 1.0, false)
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, 0.95, false)
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, 0.05, true)
// discover: 1 play.
seedSourcedPlay(t, h, user.ID, tk.ID, session, &discover, 0.8, false)
// library play (NULL source) — must be excluded.
seedSourcedPlay(t, h, user.ID, tk.ID, session, nil, 1.0, false)
// Two attributed as taste picks, the skipped one as a fresh pick (#1249).
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, &taste, 1.0, false)
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, &taste, 0.95, false)
seedSourcedPlay(t, h, user.ID, tk.ID, session, &forYou, &fresh, 0.05, true)
// A radio session play collapses into the "radio" family.
seedSourcedPlay(t, h, user.ID, tk.ID, session, &radioA, nil, 0.8, false)
// Manual play (NULL source) — the baseline row.
seedSourcedPlay(t, h, user.ID, tk.ID, session, nil, nil, 1.0, false)
req := httptest.NewRequest(http.MethodGet, "/api/me/recommendation-metrics", nil)
req = withUser(req, user)
@@ -82,15 +249,11 @@ func TestRecommendationMetrics_AggregatesBySourceExcludingNull(t *testing.T) {
if resp.WindowDays != recMetricsDefaultDays {
t.Errorf("window_days = %d, want %d", resp.WindowDays, recMetricsDefaultDays)
}
bySource := map[string]recommendationMetric{}
for _, m := range resp.Sources {
bySource[m.Source] = m
if resp.Baseline == nil || resp.Baseline.Plays != 1 {
t.Fatalf("baseline = %+v, want plays=1", resp.Baseline)
}
if _, present := bySource[""]; present {
t.Error("NULL-source (library) plays should be excluded")
}
fy, ok := bySource["for_you"]
if !ok {
fy := findSurface(resp, "for_you")
if fy == nil {
t.Fatal("for_you metrics missing")
}
if fy.Plays != 3 || fy.Skips != 1 {
@@ -102,7 +265,17 @@ func TestRecommendationMetrics_AggregatesBySourceExcludingNull(t *testing.T) {
if fy.AvgCompletion < 0.66 || fy.AvgCompletion > 0.67 {
t.Errorf("for_you avg_completion = %.4f, want ~0.6667", fy.AvgCompletion)
}
if d, ok := bySource["discover"]; !ok || d.Plays != 1 || d.Skips != 0 {
t.Errorf("discover metrics = %+v, want plays=1 skips=0", d)
// Pick-kind attribution surfaces as the For You breakdown (#1249).
if len(fy.Breakdown) != 2 {
t.Fatalf("for_you breakdown rows = %d, want 2 (taste, fresh)", len(fy.Breakdown))
}
if b := fy.Breakdown[0]; b.Key != "for_you_taste" || b.Plays != 2 || b.Skips != 0 {
t.Errorf("breakdown[0] = %+v, want for_you_taste plays=2 skips=0", b)
}
if b := fy.Breakdown[1]; b.Key != "for_you_fresh" || b.Plays != 1 || b.Skips != 1 {
t.Errorf("breakdown[1] = %+v, want for_you_fresh plays=1 skips=1", b)
}
if radio := findSurface(resp, "radio"); radio == nil || radio.Plays != 1 {
t.Errorf("radio family = %+v, want plays=1 (collapsed from radio:<uuid>)", radio)
}
}
+4 -10
View File
@@ -15,6 +15,7 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/apierror"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
"git.fabledsword.com/bvandeusen/minstrel/internal/recommendation"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
)
// RadioResponse is the body of GET /api/radio.
@@ -100,16 +101,9 @@ func (h *handlers) handleRadio(w http.ResponseWriter, r *http.Request) {
}
}
weights := recommendation.ScoringWeights{
BaseWeight: h.recCfg.BaseWeight,
LikeBoost: h.recCfg.LikeBoost,
RecencyWeight: h.recCfg.RecencyWeight,
SkipPenalty: h.recCfg.SkipPenalty,
JitterMagnitude: h.recCfg.JitterMagnitude,
ContextWeight: h.recCfg.ContextWeight,
SimilarityWeight: h.recCfg.SimilarityWeight,
TasteWeight: h.recCfg.TasteWeight,
}
// Scoring weights come from the DB-backed tuning lab (#1250) —
// read per request so an admin change takes effect live.
weights := h.recSettings.Weights(recsettings.ScopeRadio)
picks := recommendation.Shuffle(candidates, weights, time.Now().UTC(), h.rng, limit-1)
out := make([]TrackRef, 0, len(picks)+1)
+4 -20
View File
@@ -88,17 +88,11 @@ type EventsConfig struct {
SkipMaxDurationPlayedMs int `yaml:"skip_max_duration_played_ms"`
}
// RecommendationConfig governs the M3 weighted-shuffle scoring (spec §6).
// All weights are operator-tunable; defaults match the spec recommendations.
// RecommendationConfig holds the radio path's operational knobs. The
// scoring WEIGHTS moved to DB-backed admin settings (#1250,
// internal/recsettings) — YAML is bootstrap-only; anything an operator
// tunes lives in the UI with live effect.
type RecommendationConfig struct {
BaseWeight float64 `yaml:"base_weight"`
LikeBoost float64 `yaml:"like_boost"`
RecencyWeight float64 `yaml:"recency_weight"`
SkipPenalty float64 `yaml:"skip_penalty"`
JitterMagnitude float64 `yaml:"jitter_magnitude"`
ContextWeight float64 `yaml:"context_weight"`
SimilarityWeight float64 `yaml:"similarity_weight"`
TasteWeight float64 `yaml:"taste_weight"`
RecentlyPlayedHours int `yaml:"recently_played_hours"`
RadioSize int `yaml:"radio_size"`
RadioSizeMax int `yaml:"radio_size_max"`
@@ -120,16 +114,6 @@ func Default() Config {
SkipMaxDurationPlayedMs: 30000,
},
Recommendation: RecommendationConfig{
BaseWeight: 1.0,
LikeBoost: 2.0,
RecencyWeight: 1.0,
SkipPenalty: 1.0,
JitterMagnitude: 0.1,
ContextWeight: 2.0,
SimilarityWeight: 2.0,
// Radio is seed-directed (the user picked a direction), so taste
// is a lighter nudge here than in the daily mixes (1.5).
TasteWeight: 1.0,
RecentlyPlayedHours: 1,
RadioSize: 50,
RadioSizeMax: 200,
+47 -6
View File
@@ -54,7 +54,7 @@ func (q *Queries) GetMostRecentPlaySessionForUser(ctx context.Context, userID pg
}
const getOpenPlayEventForUser = `-- name: GetOpenPlayEventForUser :one
SELECT id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source FROM play_events
SELECT id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source, pick_kind FROM play_events
WHERE user_id = $1 AND ended_at IS NULL
ORDER BY started_at DESC
LIMIT 1
@@ -79,12 +79,13 @@ func (q *Queries) GetOpenPlayEventForUser(ctx context.Context, userID pgtype.UUI
&i.SessionVectorAtPlay,
&i.ScrobbledAt,
&i.Source,
&i.PickKind,
)
return i, err
}
const getPlayEventByID = `-- name: GetPlayEventByID :one
SELECT id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source FROM play_events WHERE id = $1
SELECT id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source, pick_kind FROM play_events WHERE id = $1
`
func (q *Queries) GetPlayEventByID(ctx context.Context, id pgtype.UUID) (PlayEvent, error) {
@@ -104,15 +105,48 @@ func (q *Queries) GetPlayEventByID(ctx context.Context, id pgtype.UUID) (PlayEve
&i.SessionVectorAtPlay,
&i.ScrobbledAt,
&i.Source,
&i.PickKind,
)
return i, err
}
const getSystemPickKindForTrack = `-- name: GetSystemPickKindForTrack :one
SELECT pt.pick_kind
FROM playlist_tracks pt
JOIN playlists p ON p.id = pt.playlist_id
WHERE p.user_id = $1
AND p.system_variant = $2
AND pt.track_id = $3
LIMIT 1
`
type GetSystemPickKindForTrackParams struct {
UserID pgtype.UUID
SystemVariant *string
TrackID pgtype.UUID
}
// Looks a track up in the user's CURRENT snapshot of the given system
// variant and returns its pick_kind. Used at play-ingestion time to
// freeze provenance onto the play_event — snapshots rebuild daily, so
// attribution can't be reconstructed at read time (#1249/#1270). No
// row = track not in today's snapshot (caller stores NULL); a row with
// NULL pick_kind = the variant doesn't stamp (yet). songs_like_artist
// is non-singleton (up to 3 mixes/user); a track in two of them takes
// whichever LIMIT 1 hits — acceptable, tier stamps there describe the
// same eligibility ladder.
func (q *Queries) GetSystemPickKindForTrack(ctx context.Context, arg GetSystemPickKindForTrackParams) (*string, error) {
row := q.db.QueryRow(ctx, getSystemPickKindForTrack, arg.UserID, arg.SystemVariant, arg.TrackID)
var pick_kind *string
err := row.Scan(&pick_kind)
return pick_kind, err
}
const insertPlayEvent = `-- name: InsertPlayEvent :one
INSERT INTO play_events (
user_id, track_id, session_id, started_at, client_id, source
) VALUES ($1, $2, $3, $4, $5, $6)
RETURNING id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source
user_id, track_id, session_id, started_at, client_id, source, pick_kind
) VALUES ($1, $2, $3, $4, $5, $6, $7::text)
RETURNING id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source, pick_kind
`
type InsertPlayEventParams struct {
@@ -122,8 +156,12 @@ type InsertPlayEventParams struct {
StartedAt pgtype.Timestamptz
ClientID *string
Source *string
PickKind *string
}
// pick_kind is non-NULL only for system-playlist plays whose track was
// found (with a stamped kind) in the user's live snapshot for that
// variant at ingestion time (#1249, generalized in #1270).
func (q *Queries) InsertPlayEvent(ctx context.Context, arg InsertPlayEventParams) (PlayEvent, error) {
row := q.db.QueryRow(ctx, insertPlayEvent,
arg.UserID,
@@ -132,6 +170,7 @@ func (q *Queries) InsertPlayEvent(ctx context.Context, arg InsertPlayEventParams
arg.StartedAt,
arg.ClientID,
arg.Source,
arg.PickKind,
)
var i PlayEvent
err := row.Scan(
@@ -148,6 +187,7 @@ func (q *Queries) InsertPlayEvent(ctx context.Context, arg InsertPlayEventParams
&i.SessionVectorAtPlay,
&i.ScrobbledAt,
&i.Source,
&i.PickKind,
)
return i, err
}
@@ -291,7 +331,7 @@ SET ended_at = $2,
completion_ratio = $4,
was_skipped = $5
WHERE id = $1
RETURNING id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source
RETURNING id, user_id, track_id, session_id, started_at, ended_at, duration_played_ms, completion_ratio, was_skipped, client_id, session_vector_at_play, scrobbled_at, source, pick_kind
`
type UpdatePlayEventEndedParams struct {
@@ -328,6 +368,7 @@ func (q *Queries) UpdatePlayEventEnded(ctx context.Context, arg UpdatePlayEventE
&i.SessionVectorAtPlay,
&i.ScrobbledAt,
&i.Source,
&i.PickKind,
)
return i, err
}
+32
View File
@@ -383,6 +383,7 @@ type PlayEvent struct {
SessionVectorAtPlay []byte
ScrobbledAt pgtype.Timestamptz
Source *string
PickKind *string
}
type PlaySession struct {
@@ -433,6 +434,28 @@ type PlaylistTrack struct {
AlbumTitle string
DurationSec int32
AddedAt pgtype.Timestamptz
PickKind *string
}
type RecommendationTuningAudit struct {
ID int64
ChangedAt pgtype.Timestamptz
Scope string
Action string
Changes []byte
}
type RecommendationWeightProfile struct {
Profile string
BaseWeight float64
LikeBoost float64
RecencyWeight float64
SkipPenalty float64
JitterMagnitude float64
ContextWeight float64
SimilarityWeight float64
TasteWeight float64
UpdatedAt pgtype.Timestamptz
}
type RegistrationSetting struct {
@@ -523,6 +546,15 @@ type TasteProfileTag struct {
UpdatedAt pgtype.Timestamptz
}
type TasteTuning struct {
Singleton bool
HalfLifeDays float64
EngagementHardSkip float64
EngagementNeutral float64
EngagementFull float64
UpdatedAt pgtype.Timestamptz
}
type Track struct {
ID pgtype.UUID
Title string
+13 -7
View File
@@ -12,7 +12,7 @@ import (
)
const appendPlaylistTrack = `-- name: AppendPlaylistTrack :one
INSERT INTO playlist_tracks (playlist_id, position, track_id, title, artist_name, album_title, duration_sec)
INSERT INTO playlist_tracks (playlist_id, position, track_id, title, artist_name, album_title, duration_sec, pick_kind)
SELECT
$1::uuid,
COALESCE((SELECT MAX(position) + 1 FROM playlist_tracks WHERE playlist_id = $1::uuid), 0),
@@ -20,24 +20,27 @@ SELECT
t.title,
artists.name,
albums.title,
(t.duration_ms / 1000)::integer
(t.duration_ms / 1000)::integer,
$2::text
FROM tracks t
JOIN albums ON albums.id = t.album_id
JOIN artists ON artists.id = t.artist_id
WHERE t.id = $2::uuid
RETURNING playlist_id, position, track_id, title, artist_name, album_title, duration_sec, added_at
WHERE t.id = $3::uuid
RETURNING playlist_id, position, track_id, title, artist_name, album_title, duration_sec, added_at, pick_kind
`
type AppendPlaylistTrackParams struct {
PlaylistID pgtype.UUID
PickKind *string
TrackID pgtype.UUID
}
// Inserts at the next available position. Snapshot fields are copied
// from the tracks/albums/artists join at insert time. tracks.duration_ms
// is converted to seconds for the snapshot.
// is converted to seconds for the snapshot. pick_kind is NULL for manual
// playlists; the For You builder stamps 'taste'/'fresh' (#1249).
func (q *Queries) AppendPlaylistTrack(ctx context.Context, arg AppendPlaylistTrackParams) (PlaylistTrack, error) {
row := q.db.QueryRow(ctx, appendPlaylistTrack, arg.PlaylistID, arg.TrackID)
row := q.db.QueryRow(ctx, appendPlaylistTrack, arg.PlaylistID, arg.PickKind, arg.TrackID)
var i PlaylistTrack
err := row.Scan(
&i.PlaylistID,
@@ -48,6 +51,7 @@ func (q *Queries) AppendPlaylistTrack(ctx context.Context, arg AppendPlaylistTra
&i.AlbumTitle,
&i.DurationSec,
&i.AddedAt,
&i.PickKind,
)
return i, err
}
@@ -257,7 +261,7 @@ func (q *Queries) ListAllPlaylistTracksForCollage(ctx context.Context, arg ListA
}
const listPlaylistTracks = `-- name: ListPlaylistTracks :many
SELECT pt.playlist_id, pt.position, pt.track_id, pt.title, pt.artist_name, pt.album_title, pt.duration_sec, pt.added_at,
SELECT pt.playlist_id, pt.position, pt.track_id, pt.title, pt.artist_name, pt.album_title, pt.duration_sec, pt.added_at, pt.pick_kind,
t.id AS live_track_id,
albums.id AS album_id,
artists.id AS artist_id
@@ -278,6 +282,7 @@ type ListPlaylistTracksRow struct {
AlbumTitle string
DurationSec int32
AddedAt pgtype.Timestamptz
PickKind *string
LiveTrackID pgtype.UUID
AlbumID pgtype.UUID
ArtistID pgtype.UUID
@@ -305,6 +310,7 @@ func (q *Queries) ListPlaylistTracks(ctx context.Context, playlistID pgtype.UUID
&i.AlbumTitle,
&i.DurationSec,
&i.AddedAt,
&i.PickKind,
&i.LiveTrackID,
&i.AlbumID,
&i.ArtistID,
+91 -12
View File
@@ -12,19 +12,17 @@ import (
)
const recommendationSourceMetricsForUser = `-- name: RecommendationSourceMetricsForUser :many
SELECT
pe.source,
pe.pick_kind,
count(*)::bigint AS plays,
count(*) FILTER (WHERE pe.was_skipped)::bigint AS skips,
COALESCE(
avg(pe.completion_ratio) FILTER (WHERE pe.completion_ratio IS NOT NULL),
0)::float8 AS avg_completion
count(pe.completion_ratio)::bigint AS completion_n,
COALESCE(avg(pe.completion_ratio), 0)::float8 AS avg_completion
FROM play_events pe
WHERE pe.user_id = $1
AND pe.source IS NOT NULL
AND pe.started_at > now() - ($2::float8 * INTERVAL '1 day')
GROUP BY pe.source
GROUP BY pe.source, pe.pick_kind
ORDER BY plays DESC
`
@@ -35,18 +33,17 @@ type RecommendationSourceMetricsForUserParams struct {
type RecommendationSourceMetricsForUserRow struct {
Source *string
PickKind *string
Plays int64
Skips int64
CompletionN int64
AvgCompletion float64
}
// Recommendation observability (#796 phase 4). Per-source play outcomes so the
// operator can see whether each recommendation surface is landing and tune the
// taste weights. Source is stamped on play_events when a play is launched from
// a system-playlist surface ('for_you' | 'discover' | the discovery mixes);
// NULL for library / radio / user-playlist plays, which are excluded here.
// $1 user_id, $2 window_days. plays/skips are counts; avg_completion is the
// mean completion ratio over plays that recorded one (0 when none did).
// mean completion ratio over the completion_n plays that recorded one.
// pick_kind splits For You plays into taste/fresh/unattributed (#1249);
// it is NULL for every other source, so those still group to one row.
func (q *Queries) RecommendationSourceMetricsForUser(ctx context.Context, arg RecommendationSourceMetricsForUserParams) ([]RecommendationSourceMetricsForUserRow, error) {
rows, err := q.db.Query(ctx, recommendationSourceMetricsForUser, arg.UserID, arg.Column2)
if err != nil {
@@ -58,8 +55,10 @@ func (q *Queries) RecommendationSourceMetricsForUser(ctx context.Context, arg Re
var i RecommendationSourceMetricsForUserRow
if err := rows.Scan(
&i.Source,
&i.PickKind,
&i.Plays,
&i.Skips,
&i.CompletionN,
&i.AvgCompletion,
); err != nil {
return nil, err
@@ -71,3 +70,83 @@ func (q *Queries) RecommendationSourceMetricsForUser(ctx context.Context, arg Re
}
return items, nil
}
const recommendationWeeklyTrends = `-- name: RecommendationWeeklyTrends :many
SELECT
date_trunc('week', pe.started_at)::date AS week_start,
pe.source,
count(*)::bigint AS plays,
count(*) FILTER (WHERE pe.was_skipped)::bigint AS skips,
count(pe.completion_ratio)::bigint AS completion_n,
COALESCE(avg(pe.completion_ratio), 0)::float8 AS avg_completion,
count(*) FILTER (WHERE tpa.artist_id IS NOT NULL)::bigint AS taste_hits
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
LEFT JOIN taste_profile_artists tpa
ON tpa.user_id = pe.user_id
AND tpa.artist_id = t.artist_id
AND tpa.weight > 0
WHERE pe.started_at > now() - ($1::int * INTERVAL '1 week')
GROUP BY 1, 2
ORDER BY 1, 2
`
type RecommendationWeeklyTrendsRow struct {
WeekStart pgtype.Date
Source *string
Plays int64
Skips int64
CompletionN int64
AvgCompletion float64
TasteHits int64
}
// Recommendation observability (#796 phase 4). Per-source play outcomes so the
// operator can see whether each recommendation surface is landing and tune the
// taste weights. Source is stamped on play_events when a play is launched from
// a recommendation surface; NULL means the user picked the track manually —
// those rows are INCLUDED here as the baseline control group the surfaces are
// judged against (milestone #127: delta-vs-baseline is what makes the numbers
// actionable). Raw source strings are bucketed into stable surface families in
// the Go handler; completion_n is carried so family merges can weight
// avg_completion correctly.
// Weekly per-source outcome series for the tuning lab's trend view
// (#1251). Aggregated across ALL users: the tuning knobs are global,
// so judging a knob turn needs global outcomes — rows carry rates
// only, no track or user identity. NULL-source (manual) rows are
// included as the baseline family.
//
// taste_hits counts plays whose track's artist has a positive weight
// in that user's CURRENT taste profile — the "cheap recompute" option:
// retroactive over the whole window, at the cost of drift (the profile
// is today's, the play may be weeks old). Good enough to read whether
// a surface is feeding taste-fitting tracks.
// $1 window in weeks.
func (q *Queries) RecommendationWeeklyTrends(ctx context.Context, weeks int32) ([]RecommendationWeeklyTrendsRow, error) {
rows, err := q.db.Query(ctx, recommendationWeeklyTrends, weeks)
if err != nil {
return nil, err
}
defer rows.Close()
var items []RecommendationWeeklyTrendsRow
for rows.Next() {
var i RecommendationWeeklyTrendsRow
if err := rows.Scan(
&i.WeekStart,
&i.Source,
&i.Plays,
&i.Skips,
&i.CompletionN,
&i.AvgCompletion,
&i.TasteHits,
); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
@@ -0,0 +1,272 @@
// Code generated by sqlc. DO NOT EDIT.
// versions:
// sqlc v1.31.1
// source: recommendation_tuning.sql
package dbq
import (
"context"
)
const getTasteTuning = `-- name: GetTasteTuning :one
SELECT singleton, half_life_days, engagement_hard_skip, engagement_neutral, engagement_full, updated_at FROM taste_tuning WHERE singleton = true
`
func (q *Queries) GetTasteTuning(ctx context.Context) (TasteTuning, error) {
row := q.db.QueryRow(ctx, getTasteTuning)
var i TasteTuning
err := row.Scan(
&i.Singleton,
&i.HalfLifeDays,
&i.EngagementHardSkip,
&i.EngagementNeutral,
&i.EngagementFull,
&i.UpdatedAt,
)
return i, err
}
const insertTuningAudit = `-- name: InsertTuningAudit :exec
INSERT INTO recommendation_tuning_audit (scope, action, changes)
VALUES ($1, $2, $3)
`
type InsertTuningAuditParams struct {
Scope string
Action string
Changes []byte
}
// changes is a jsonb array of {field, old, new} objects.
func (q *Queries) InsertTuningAudit(ctx context.Context, arg InsertTuningAuditParams) error {
_, err := q.db.Exec(ctx, insertTuningAudit, arg.Scope, arg.Action, arg.Changes)
return err
}
const listTuningAudit = `-- name: ListTuningAudit :many
SELECT id, changed_at, scope, action, changes
FROM recommendation_tuning_audit
ORDER BY changed_at DESC, id DESC
LIMIT $1
`
// Newest first; consumed by the metrics trend view (#1251) to annotate
// knob turns on the timeline.
func (q *Queries) ListTuningAudit(ctx context.Context, limit int32) ([]RecommendationTuningAudit, error) {
rows, err := q.db.Query(ctx, listTuningAudit, limit)
if err != nil {
return nil, err
}
defer rows.Close()
var items []RecommendationTuningAudit
for rows.Next() {
var i RecommendationTuningAudit
if err := rows.Scan(
&i.ID,
&i.ChangedAt,
&i.Scope,
&i.Action,
&i.Changes,
); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const listWeightProfiles = `-- name: ListWeightProfiles :many
SELECT profile, base_weight, like_boost, recency_weight, skip_penalty, jitter_magnitude, context_weight, similarity_weight, taste_weight, updated_at FROM recommendation_weight_profiles ORDER BY profile
`
func (q *Queries) ListWeightProfiles(ctx context.Context) ([]RecommendationWeightProfile, error) {
rows, err := q.db.Query(ctx, listWeightProfiles)
if err != nil {
return nil, err
}
defer rows.Close()
var items []RecommendationWeightProfile
for rows.Next() {
var i RecommendationWeightProfile
if err := rows.Scan(
&i.Profile,
&i.BaseWeight,
&i.LikeBoost,
&i.RecencyWeight,
&i.SkipPenalty,
&i.JitterMagnitude,
&i.ContextWeight,
&i.SimilarityWeight,
&i.TasteWeight,
&i.UpdatedAt,
); err != nil {
return nil, err
}
items = append(items, i)
}
if err := rows.Err(); err != nil {
return nil, err
}
return items, nil
}
const updateTasteTuning = `-- name: UpdateTasteTuning :one
UPDATE taste_tuning
SET half_life_days = $1,
engagement_hard_skip = $2,
engagement_neutral = $3,
engagement_full = $4,
updated_at = now()
WHERE singleton = true
RETURNING singleton, half_life_days, engagement_hard_skip, engagement_neutral, engagement_full, updated_at
`
type UpdateTasteTuningParams struct {
HalfLifeDays float64
EngagementHardSkip float64
EngagementNeutral float64
EngagementFull float64
}
func (q *Queries) UpdateTasteTuning(ctx context.Context, arg UpdateTasteTuningParams) (TasteTuning, error) {
row := q.db.QueryRow(ctx, updateTasteTuning,
arg.HalfLifeDays,
arg.EngagementHardSkip,
arg.EngagementNeutral,
arg.EngagementFull,
)
var i TasteTuning
err := row.Scan(
&i.Singleton,
&i.HalfLifeDays,
&i.EngagementHardSkip,
&i.EngagementNeutral,
&i.EngagementFull,
&i.UpdatedAt,
)
return i, err
}
const updateWeightProfile = `-- name: UpdateWeightProfile :one
UPDATE recommendation_weight_profiles
SET base_weight = $2,
like_boost = $3,
recency_weight = $4,
skip_penalty = $5,
jitter_magnitude = $6,
context_weight = $7,
similarity_weight = $8,
taste_weight = $9,
updated_at = now()
WHERE profile = $1
RETURNING profile, base_weight, like_boost, recency_weight, skip_penalty, jitter_magnitude, context_weight, similarity_weight, taste_weight, updated_at
`
type UpdateWeightProfileParams struct {
Profile string
BaseWeight float64
LikeBoost float64
RecencyWeight float64
SkipPenalty float64
JitterMagnitude float64
ContextWeight float64
SimilarityWeight float64
TasteWeight float64
}
func (q *Queries) UpdateWeightProfile(ctx context.Context, arg UpdateWeightProfileParams) (RecommendationWeightProfile, error) {
row := q.db.QueryRow(ctx, updateWeightProfile,
arg.Profile,
arg.BaseWeight,
arg.LikeBoost,
arg.RecencyWeight,
arg.SkipPenalty,
arg.JitterMagnitude,
arg.ContextWeight,
arg.SimilarityWeight,
arg.TasteWeight,
)
var i RecommendationWeightProfile
err := row.Scan(
&i.Profile,
&i.BaseWeight,
&i.LikeBoost,
&i.RecencyWeight,
&i.SkipPenalty,
&i.JitterMagnitude,
&i.ContextWeight,
&i.SimilarityWeight,
&i.TasteWeight,
&i.UpdatedAt,
)
return i, err
}
const upsertTasteTuningDefaults = `-- name: UpsertTasteTuningDefaults :exec
INSERT INTO taste_tuning (
singleton, half_life_days, engagement_hard_skip,
engagement_neutral, engagement_full
) VALUES (true, $1, $2, $3, $4)
ON CONFLICT (singleton) DO NOTHING
`
type UpsertTasteTuningDefaultsParams struct {
HalfLifeDays float64
EngagementHardSkip float64
EngagementNeutral float64
EngagementFull float64
}
func (q *Queries) UpsertTasteTuningDefaults(ctx context.Context, arg UpsertTasteTuningDefaultsParams) error {
_, err := q.db.Exec(ctx, upsertTasteTuningDefaults,
arg.HalfLifeDays,
arg.EngagementHardSkip,
arg.EngagementNeutral,
arg.EngagementFull,
)
return err
}
const upsertWeightProfileDefaults = `-- name: UpsertWeightProfileDefaults :exec
INSERT INTO recommendation_weight_profiles (
profile, base_weight, like_boost, recency_weight, skip_penalty,
jitter_magnitude, context_weight, similarity_weight, taste_weight
) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9)
ON CONFLICT (profile) DO NOTHING
`
type UpsertWeightProfileDefaultsParams struct {
Profile string
BaseWeight float64
LikeBoost float64
RecencyWeight float64
SkipPenalty float64
JitterMagnitude float64
ContextWeight float64
SimilarityWeight float64
TasteWeight float64
}
// Recommendation tuning lab queries (#1250). Seeding happens via the
// recsettings boot reconcile; shipped defaults live in Go only.
// Boot reconcile: insert the shipped defaults for a profile if the row
// doesn't exist yet. Never overwrites operator-tuned values.
func (q *Queries) UpsertWeightProfileDefaults(ctx context.Context, arg UpsertWeightProfileDefaultsParams) error {
_, err := q.db.Exec(ctx, upsertWeightProfileDefaults,
arg.Profile,
arg.BaseWeight,
arg.LikeBoost,
arg.RecencyWeight,
arg.SkipPenalty,
arg.JitterMagnitude,
arg.ContextWeight,
arg.SimilarityWeight,
arg.TasteWeight,
)
return err
}
+162 -62
View File
@@ -27,12 +27,20 @@ play_counts AS (
FROM play_events
WHERE user_id = $1 AND was_skipped = false
GROUP BY track_id
),
skip_counts AS (
SELECT track_id, COUNT(*) AS c
FROM play_events
WHERE user_id = $1 AND was_skipped = true
GROUP BY track_id
)
SELECT t.id, t.album_id, t.artist_id
FROM tracks t
JOIN affinity_artists aa ON aa.artist_id = t.artist_id
LEFT JOIN play_counts pc ON pc.track_id = t.id
LEFT JOIN skip_counts sc ON sc.track_id = t.id
WHERE COALESCE(pc.c, 0) <= 2
AND COALESCE(sc.c, 0) < 2
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
@@ -60,6 +68,11 @@ type ListDeepCutsTracksRow struct {
// gives per-play variety on top.
// #419 Deep Cuts: low-play tracks (<=2 plays) from artists the user
// has liked OR played heavily (>=5 non-skip plays across the artist).
// Tracks the user has skipped twice or more don't qualify (#1257):
// eligibility used to count only unskipped plays, so the most actively
// rejected tracks read as "barely heard" and kept being re-offered.
// Threshold 2 so a single accidental skip doesn't banish a track;
// passive-signal only (no dislike UI, rule #101).
// $1 user_id, $2 date string.
func (q *Queries) ListDeepCutsTracks(ctx context.Context, arg ListDeepCutsTracksParams) ([]ListDeepCutsTracksRow, error) {
rows, err := q.db.Query(ctx, listDeepCutsTracks, arg.UserID, arg.Column2)
@@ -82,31 +95,49 @@ func (q *Queries) ListDeepCutsTracks(ctx context.Context, arg ListDeepCutsTracks
}
const listFirstListensTracks = `-- name: ListFirstListensTracks :many
WITH heard_albums AS (
SELECT DISTINCT t.album_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
WITH attempted AS (
SELECT DISTINCT pe.track_id
FROM play_events pe
WHERE pe.user_id = $1 AND pe.duration_played_ms >= 30000
),
played_artists AS (
album_attempts AS (
SELECT t.album_id,
COUNT(a.track_id) AS attempted_count,
COUNT(*) AS track_count
FROM tracks t
LEFT JOIN attempted a ON a.track_id = t.id
GROUP BY t.album_id
),
attempted_artists AS (
SELECT DISTINCT t.artist_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
FROM attempted a JOIN tracks t ON t.id = a.track_id
),
albums_tiered AS (
SELECT aa.album_id,
CASE
WHEN aa.attempted_count = 0 THEN 1
WHEN aa.attempted_count::float / GREATEST(aa.track_count, 1) <= 0.25 THEN 2
END AS tier
FROM album_attempts aa
)
SELECT t.id, t.album_id, t.artist_id
SELECT t.id, t.album_id, t.artist_id, alt.tier::int AS tier
FROM tracks t
JOIN albums al ON al.id = t.album_id
WHERE NOT EXISTS (SELECT 1 FROM heard_albums h WHERE h.album_id = al.id)
JOIN albums_tiered alt ON alt.album_id = al.id
WHERE alt.tier IS NOT NULL
AND NOT EXISTS (SELECT 1 FROM attempted a WHERE a.track_id = t.id)
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
ORDER BY
alt.tier,
(CASE
WHEN EXISTS (
SELECT 1 FROM general_likes_artists gla
WHERE gla.user_id = $1 AND gla.artist_id = al.artist_id
) THEN 0
WHEN al.artist_id IN (SELECT artist_id FROM played_artists) THEN 1
WHEN al.artist_id IN (SELECT artist_id FROM attempted_artists) THEN 1
ELSE 2
END),
al.id, t.disc_number NULLS FIRST, t.track_number NULLS FIRST
@@ -117,12 +148,25 @@ type ListFirstListensTracksRow struct {
ID pgtype.UUID
AlbumID pgtype.UUID
ArtistID pgtype.UUID
Tier int32
}
// #423 First Listens: albums the user has never played any track of.
// Tiered: liked-artist albums first, then played-artist albums, then
// the rest — album-coherent within each tier. Not diversity-capped
// (whole-album discovery). $1 user_id.
// #423 First Listens: songs the user has never heard or even
// attempted. "Attempted" is track-level with a >=30s listen threshold
// (#1268, operator decision 2026-07-03) — the old version disqualified
// a whole album on ANY play_event, so a 2-second accidental brush
// banished it permanently.
//
// tier 1 the exact desire: albums with ZERO attempted tracks
// tier 2 step back: barely-attempted albums (<=25% of tracks
// reached 30s) — brushed, never explored. The already-
// attempted tracks themselves are excluded; they're not
// first listens.
//
// Within each tier, affinity-ordered as before: liked artist, then
// attempted-played artist (>=30s — skip-only contact isn't trust),
// then the rest; album-coherent. The tier lands on
// playlist_tracks.pick_kind via the producer. $1 user_id.
func (q *Queries) ListFirstListensTracks(ctx context.Context, userID pgtype.UUID) ([]ListFirstListensTracksRow, error) {
rows, err := q.db.Query(ctx, listFirstListensTracks, userID)
if err != nil {
@@ -132,7 +176,12 @@ func (q *Queries) ListFirstListensTracks(ctx context.Context, userID pgtype.UUID
var items []ListFirstListensTracksRow
for rows.Next() {
var i ListFirstListensTracksRow
if err := rows.Scan(&i.ID, &i.AlbumID, &i.ArtistID); err != nil {
if err := rows.Scan(
&i.ID,
&i.AlbumID,
&i.ArtistID,
&i.Tier,
); err != nil {
return nil, err
}
items = append(items, i)
@@ -144,39 +193,76 @@ func (q *Queries) ListFirstListensTracks(ctx context.Context, userID pgtype.UUID
}
const listNewForYouTracks = `-- name: ListNewForYouTracks :many
WITH affinity_artists AS (
SELECT artist_id FROM general_likes_artists WHERE user_id = $1
WITH attempted AS (
SELECT DISTINCT pe.track_id
FROM play_events pe
WHERE pe.user_id = $1 AND pe.duration_played_ms >= 30000
),
consumed_albums AS (
SELECT DISTINCT t.album_id
FROM attempted a JOIN tracks t ON t.id = a.track_id
),
affinity_artists AS (
SELECT gla.artist_id FROM general_likes_artists gla WHERE gla.user_id = $1
UNION
SELECT t.artist_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1 AND pe.was_skipped = false
FROM attempted a JOIN tracks t ON t.id = a.track_id
GROUP BY t.artist_id
HAVING COUNT(*) >= 3
),
albums_tiered AS (
SELECT al.id AS album_id, al.created_at,
CASE
WHEN al.created_at >= now() - interval '30 days'
AND al.artist_id IN (SELECT artist_id FROM affinity_artists) THEN 1
WHEN al.artist_id IN (SELECT artist_id FROM affinity_artists) THEN 2
ELSE 3
END AS tier
FROM albums al
WHERE al.created_at >= now() - interval '90 days'
AND al.id NOT IN (SELECT album_id FROM consumed_albums)
)
SELECT t.id, t.album_id, t.artist_id
SELECT t.id, t.album_id, t.artist_id, alt.tier::int AS tier
FROM tracks t
JOIN albums al ON al.id = t.album_id
JOIN affinity_artists aa ON aa.artist_id = al.artist_id
WHERE al.created_at >= now() - interval '30 days'
AND NOT EXISTS (
JOIN albums_tiered alt ON alt.album_id = t.album_id
WHERE NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
ORDER BY al.created_at DESC, t.disc_number NULLS FIRST, t.track_number NULLS FIRST
LIMIT 200
ORDER BY alt.tier, alt.created_at DESC, t.album_id,
t.disc_number NULLS FIRST, t.track_number NULLS FIRST
LIMIT 300
`
type ListNewForYouTracksRow struct {
ID pgtype.UUID
AlbumID pgtype.UUID
ArtistID pgtype.UUID
Tier int32
}
// #421 New for you: tracks from albums added in the last 30 days
// whose artist the user has liked OR played (>=3 non-skip). Album-
// coherent (newest album first, then disc/track) — the producer does
// NOT diversity-cap these; they're meant as whole-album discovery.
// $1 user_id.
// #421 New for you, tiered per project rule #131 (#1267). The old
// single hard rule (added <30d AND direct affinity) had two failure
// modes with one root: no notion of consumption — the album you
// devoured in week one crowded the mix for three more weeks, and
// after a quiet month nothing qualified at all.
//
// "Consumed" = any track attempted for >=30s (the First Listens
// threshold, #1268): once you've meaningfully engaged, the album is
// no longer news. Applies to every tier — played albums leave the
// mix the next build.
//
// tier 1 the exact desire: unconsumed albums added <30d by
// direct-affinity artists (liked, or >=3 attempted plays)
// tier 2 step back a little: unconsumed affinity albums from the
// wider 30-90d window — added while you weren't looking
// tier 3 step back more: any unconsumed album added <90d
// regardless of affinity, newest first
//
// Album-coherent within tiers (newest album first, then disc/track);
// the producer fills top-down and rotates within tier blocks. The
// tier lands on playlist_tracks.pick_kind so metrics can price what
// each step-back trades away. $1 user_id.
func (q *Queries) ListNewForYouTracks(ctx context.Context, userID pgtype.UUID) ([]ListNewForYouTracksRow, error) {
rows, err := q.db.Query(ctx, listNewForYouTracks, userID)
if err != nil {
@@ -186,7 +272,12 @@ func (q *Queries) ListNewForYouTracks(ctx context.Context, userID pgtype.UUID) (
var items []ListNewForYouTracksRow
for rows.Next() {
var i ListNewForYouTracksRow
if err := rows.Scan(&i.ID, &i.AlbumID, &i.ArtistID); err != nil {
if err := rows.Scan(
&i.ID,
&i.AlbumID,
&i.ArtistID,
&i.Tier,
); err != nil {
return nil, err
}
items = append(items, i)
@@ -203,7 +294,10 @@ WITH windowed AS (
FROM play_events
WHERE user_id = $1 AND was_skipped = false
AND started_at < now() - interval '30 days'
AND ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM now())) <= 10
AND LEAST(
ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM $2::date)),
365 - ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM $2::date))
) <= 10
GROUP BY track_id
)
SELECT t.id, t.album_id, t.artist_id
@@ -235,6 +329,13 @@ type ListOnThisDayTracksRow struct {
// and window ±7→±10 so it surfaces on a months-old library instead
// of needing a full year of history; still skips cleanly (no rows →
// no playlist) when there's no qualifying history yet.
// Day-of-year distance is circular (#1256): plain ABS made Dec 28 vs
// Jan 3 read as 359 days apart, silently gutting the window for ~3
// weeks around every New Year — precisely when holiday nostalgia is
// the point. LEAST(d, 365-d) wraps the boundary; leap-year drift of
// ±1 is absorbed by the ±10 window. The build date ($2, already here
// for the md5 rotation) anchors "today" instead of now() so the
// window is testable and consistent with the mix's daily determinism.
// $1 user_id, $2 date string.
func (q *Queries) ListOnThisDayTracks(ctx context.Context, arg ListOnThisDayTracksParams) ([]ListOnThisDayTracksRow, error) {
rows, err := q.db.Query(ctx, listOnThisDayTracks, arg.UserID, arg.Column2)
@@ -262,37 +363,32 @@ WITH stats AS (
FROM play_events pe
WHERE pe.user_id = $1 AND pe.was_skipped = false
GROUP BY pe.track_id
),
deep AS (
SELECT t.id, t.album_id, t.artist_id, s.c, 0 AS tier
FROM tracks t
JOIN stats s ON s.track_id = t.id
WHERE s.c >= 5
AND s.last_at <= now() - interval '6 months'
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
),
shallow AS (
SELECT t.id, t.album_id, t.artist_id, s.c, 1 AS tier
SELECT t.id, t.album_id, t.artist_id
FROM tracks t
JOIN stats s ON s.track_id = t.id
WHERE s.c >= 5
WHERE s.c >= 3
AND s.last_at <= now() - interval '30 days'
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
-- Declined-resurfacing cooldown (#1258): a skip on a
-- rediscover-sourced play is the user explicitly saying "I've moved
-- on" to the exact invitation this mix extends — such tracks sit
-- out ~90 days instead of re-qualifying the next day. Passive
-- signal only (rule #101).
AND NOT EXISTS (
SELECT 1 FROM play_events rp
WHERE rp.user_id = $1
AND rp.track_id = t.id
AND rp.source = 'rediscover'
AND rp.was_skipped = true
AND rp.started_at > now() - interval '90 days'
)
SELECT id, album_id, artist_id
FROM (
SELECT id, album_id, artist_id, c, tier FROM deep
UNION ALL
SELECT id, album_id, artist_id, c, tier FROM shallow
WHERE NOT EXISTS (SELECT 1 FROM deep)
) u
ORDER BY tier, c DESC, id
ORDER BY (s.last_at <= now() - interval '6 months') DESC,
(s.c >= 5) DESC,
s.c DESC, t.id
LIMIT 200
`
@@ -302,13 +398,17 @@ type ListRediscoverTracksRow struct {
ArtistID pgtype.UUID
}
// #420 Rediscover: tracks the user played a lot (>=5 non-skip) but
// has drifted away from. Tiered so a young library still gets a mix:
//
// tier 0 not played in the last 6 months (true rediscovery)
// tier 1 (only if tier 0 empty) not played in the last 30 days
//
// Ordered by historical affection. $1 user_id.
// #420 Rediscover: tracks the user played a lot but has drifted away
// from. One blended pool (issue #1246: the old two-tier UNION was
// all-or-nothing — a single ">=6 months cold" row suppressed the whole
// ">=30 days" tier, which is how a one-song playlist shipped; the 6mo
// tier was also a strict subset of the 30d tier). Eligibility is >=3
// non-skip plays and >=30 days cold; the bar sits at 3 rather than 5
// so a weeks-old library still fields a pool — on young histories the
// >=5-play tracks are precisely the ones still in rotation. Ordering
// prefers true rediscoveries (>=6 months cold), then strong affection
// (>=5 plays), then raw play count; the producer's minimum floor
// decides whether the pool is big enough to ship at all. $1 user_id.
func (q *Queries) ListRediscoverTracks(ctx context.Context, userID pgtype.UUID) ([]ListRediscoverTracksRow, error) {
rows, err := q.db.Query(ctx, listRediscoverTracks, userID)
if err != nil {
+75 -19
View File
@@ -290,9 +290,13 @@ func (q *Queries) ListPlaylistsByUserAndKind(ctx context.Context, arg ListPlayli
}
const pickSeedArtists = `-- name: PickSeedArtists :many
WITH plays AS (
WITH liked AS (
SELECT gla.artist_id FROM general_likes_artists gla WHERE gla.user_id = $1
),
recent7 AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
0 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
@@ -300,27 +304,79 @@ WITH plays AS (
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
liked AS (
SELECT artist_id FROM general_likes_artists WHERE user_id = $1
recent30 AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
1 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
AND pe.started_at > now() - INTERVAL '30 days'
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
alltime AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
2 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
likedonly AS (
SELECT l.artist_id, 0::bigint AS play_count, 3 AS tier
FROM liked l
),
chosen AS (
SELECT artist_id, play_count, tier FROM recent7
UNION ALL
SELECT artist_id, play_count, tier FROM recent30
WHERE NOT EXISTS (SELECT 1 FROM recent7)
UNION ALL
SELECT artist_id, play_count, tier FROM alltime
WHERE NOT EXISTS (SELECT 1 FROM recent7)
AND NOT EXISTS (SELECT 1 FROM recent30)
UNION ALL
SELECT artist_id, play_count, tier FROM likedonly
WHERE NOT EXISTS (SELECT 1 FROM recent7)
AND NOT EXISTS (SELECT 1 FROM recent30)
AND NOT EXISTS (SELECT 1 FROM alltime)
)
SELECT p.artist_id,
(p.play_count + CASE WHEN l.artist_id IS NOT NULL THEN 5 ELSE 0 END)::bigint AS score
FROM plays p
LEFT JOIN liked l ON l.artist_id = p.artist_id
ORDER BY score DESC, p.artist_id
SELECT c.artist_id,
(c.play_count + CASE WHEN l.artist_id IS NOT NULL THEN 5 ELSE 0 END)::bigint AS score,
c.tier::int AS tier
FROM chosen c
LEFT JOIN liked l ON l.artist_id = c.artist_id
ORDER BY score DESC, c.artist_id
LIMIT 5
`
type PickSeedArtistsRow struct {
ArtistID pgtype.UUID
Score int64
Tier int32
}
// Top-5 most-engaged distinct artist candidates in the user's last 7
// days. The Go-side picker (pickSeedArtistsForDay) shuffles these
// daily-deterministically and takes the first 3 so the set of
// "Songs like X" mixes rotates day-to-day.
// Score = unskipped-play count + 5 if user has liked the artist.
// Top-5 most-engaged distinct artist candidates, tiered so the
// "Songs like X" mixes never silently vanish (#1255): the old hard
// 7-day window emptied the seed pool after a quiet week, and the
// daily atomic-replace build then deleted every existing mix until
// the user played something again. Same fallback shape as
// PickTopPlayedTracksForUser (For You's seeds):
//
// tier 0 engagement in the last 7 days
// tier 1 (only if tier 0 empty) last 30 days
// tier 2 (only if tiers 0+1 empty) all-time
// tier 3 (only if 0-2 empty) liked artists with no play history
//
// All returned rows share one tier; produceSeedMixes maps it onto the
// rule-#131 pick-kind ladder and stamps the built tracks, so metrics
// can compare mixes seeded from fresh vs stale engagement.
// Score = unskipped-play count + 5 if user has liked the artist. The
// Go-side picker (pickSeedArtistsForDay) shuffles the 5
// daily-deterministically and takes 3 so the mix set rotates.
func (q *Queries) PickSeedArtists(ctx context.Context, userID pgtype.UUID) ([]PickSeedArtistsRow, error) {
rows, err := q.db.Query(ctx, pickSeedArtists, userID)
if err != nil {
@@ -330,7 +386,7 @@ func (q *Queries) PickSeedArtists(ctx context.Context, userID pgtype.UUID) ([]Pi
var items []PickSeedArtistsRow
for rows.Next() {
var i PickSeedArtistsRow
if err := rows.Scan(&i.ArtistID, &i.Score); err != nil {
if err := rows.Scan(&i.ArtistID, &i.Score, &i.Tier); err != nil {
return nil, err
}
items = append(items, i)
@@ -457,10 +513,10 @@ SELECT id
// tier 2 (only if tiers 0+1 empty) liked tracks
//
// Returns up to 5 ids; tie-break by track_id for determinism. The
// Go-side picker (pickForYouSeedForDay) rotates one per day via
// userIDHash. Widened from a hard 7-day window, which made For-You
// disappear after a week of not listening and never recover on a
// self-hosted library with sparse history.
// Go-side picker (pickDailySeeds) draws the day's seeds from these.
// Widened from a hard 7-day window, which made For-You disappear
// after a week of not listening and never recover on a self-hosted
// library with sparse history.
func (q *Queries) PickTopPlayedTracksForUser(ctx context.Context, userID pgtype.UUID) ([]pgtype.UUID, error) {
rows, err := q.db.Query(ctx, pickTopPlayedTracksForUser, userID)
if err != nil {
@@ -0,0 +1,2 @@
ALTER TABLE play_events DROP COLUMN pick_kind;
ALTER TABLE playlist_tracks DROP COLUMN pick_kind;
@@ -0,0 +1,34 @@
-- For You exploration attribution (milestone #127 step 2, Scribe #1249).
--
-- For You is deliberately composed of two populations: a head of
-- top-scored similarity matches ("taste" picks) and a tail sampled from
-- deeper in the candidate ranking ("fresh" picks — the freshness
-- injection). Until now the metrics could only judge For You as one
-- blob, so a high skip rate couldn't be read as "the taste engine is
-- missing" vs "the freshness tax is too high".
--
-- Two denormalized columns, both nullable:
--
-- 1. playlist_tracks.pick_kind — stamped by the system-playlist builder
-- at snapshot build time. NULL for manual playlists and for system
-- variants that don't split (Discover, Songs like X, discovery
-- mixes). System snapshots are atomically replaced daily, so the
-- column never goes stale within a snapshot.
--
-- 2. play_events.pick_kind — stamped at play-ingestion time by looking
-- the track up in the user's CURRENT For You snapshot. Attribution
-- must be frozen at play time: For You rebuilds daily, so a 30-day
-- metrics window spans ~30 snapshots and a join-at-read against the
-- live snapshot would misattribute nearly everything. Plays that
-- predate this feature (or whose track already rotated out, e.g. a
-- late offline replay) stay NULL → reported as unattributed.
ALTER TABLE playlist_tracks
ADD COLUMN pick_kind text
CONSTRAINT playlist_tracks_pick_kind_check
CHECK (pick_kind IN ('taste', 'fresh'));
ALTER TABLE play_events
ADD COLUMN pick_kind text
CONSTRAINT play_events_pick_kind_check
CHECK (pick_kind IN ('taste', 'fresh'));
@@ -0,0 +1,19 @@
-- Rows stamped with the widened vocabulary must be cleared before the
-- narrow CHECK can be re-added; NULL reads as "unattributed", which is
-- the honest downgrade.
UPDATE play_events SET pick_kind = NULL
WHERE pick_kind NOT IN ('taste', 'fresh');
UPDATE playlist_tracks SET pick_kind = NULL
WHERE pick_kind NOT IN ('taste', 'fresh');
ALTER TABLE play_events
DROP CONSTRAINT play_events_pick_kind_check;
ALTER TABLE play_events
ADD CONSTRAINT play_events_pick_kind_check
CHECK (pick_kind IN ('taste', 'fresh'));
ALTER TABLE playlist_tracks
DROP CONSTRAINT playlist_tracks_pick_kind_check;
ALTER TABLE playlist_tracks
ADD CONSTRAINT playlist_tracks_pick_kind_check
CHECK (pick_kind IN ('taste', 'fresh'));
@@ -0,0 +1,41 @@
-- Provenance as standard (milestone #127, Scribe #1270).
--
-- #1249 introduced pick_kind as a For You one-off ('taste'/'fresh').
-- The mechanism — stamp WHY a track is in the snapshot at build time,
-- freeze it onto the play at ingestion, break it down in metrics — is
-- now the standard for every system mix, so the CHECK vocabulary
-- widens to cover:
--
-- taste / fresh For You head vs freshness-injection tail
-- dormant / cross_user / random
-- Discover's three candidate buckets — makes
-- the 40/30/30 allocation measurable instead
-- of a guess
-- tier1 / tier2 / tier3 the tiered-eligibility ladder (project rule
-- #131): tier 1 pins the mix's exact desire,
-- each higher tier steps back a little. Tier
-- provenance is how we measure what the
-- step-back trades away.
--
-- Postgres CHECK whitelists can't be altered in place: DROP + re-ADD
-- with the expanded list, in the same migration.
ALTER TABLE playlist_tracks
DROP CONSTRAINT playlist_tracks_pick_kind_check;
ALTER TABLE playlist_tracks
ADD CONSTRAINT playlist_tracks_pick_kind_check
CHECK (pick_kind IN (
'taste', 'fresh',
'dormant', 'cross_user', 'random',
'tier1', 'tier2', 'tier3'
));
ALTER TABLE play_events
DROP CONSTRAINT play_events_pick_kind_check;
ALTER TABLE play_events
ADD CONSTRAINT play_events_pick_kind_check
CHECK (pick_kind IN (
'taste', 'fresh',
'dormant', 'cross_user', 'random',
'tier1', 'tier2', 'tier3'
));
@@ -0,0 +1,3 @@
DROP TABLE recommendation_tuning_audit;
DROP TABLE taste_tuning;
DROP TABLE recommendation_weight_profiles;
@@ -0,0 +1,57 @@
-- Recommendation tuning lab (milestone #127, Scribe #1250).
--
-- The scoring weights move out of YAML (radio profile) and out of the
-- systemMixWeights hard-code (daily_mix profile) into DB-backed
-- settings with live effect — the defaults-discovery lab (decision
-- #1247): the operator turns knobs here to FIND good values, which
-- then get baked into shipped defaults; end users and other operators
-- should never need this card.
--
-- Rows are seeded by the recsettings service's boot reconcile (the
-- coverart SettingsService pattern), not by this migration, so shipped
-- defaults live in exactly one place (Go).
CREATE TABLE recommendation_weight_profiles (
profile text PRIMARY KEY
CONSTRAINT recommendation_weight_profiles_profile_check
CHECK (profile IN ('radio', 'daily_mix')),
base_weight double precision NOT NULL,
like_boost double precision NOT NULL,
recency_weight double precision NOT NULL,
skip_penalty double precision NOT NULL,
jitter_magnitude double precision NOT NULL,
context_weight double precision NOT NULL,
similarity_weight double precision NOT NULL,
taste_weight double precision NOT NULL,
updated_at timestamptz NOT NULL DEFAULT now()
);
-- Taste-profile build knobs: engagement half-life + the completion→
-- engagement curve points (taste.Config's tunable subset).
CREATE TABLE taste_tuning (
singleton boolean PRIMARY KEY DEFAULT true
CONSTRAINT taste_tuning_singleton_check CHECK (singleton),
half_life_days double precision NOT NULL,
engagement_hard_skip double precision NOT NULL,
engagement_neutral double precision NOT NULL,
engagement_full double precision NOT NULL,
updated_at timestamptz NOT NULL DEFAULT now()
);
-- Every knob turn writes one audit row; the metrics trend view
-- (#1251) annotates these on its timeline so cause→effect is visible
-- after a change. changes is a jsonb array of {field, old, new}.
CREATE TABLE recommendation_tuning_audit (
id bigserial PRIMARY KEY,
changed_at timestamptz NOT NULL DEFAULT now(),
scope text NOT NULL
CONSTRAINT recommendation_tuning_audit_scope_check
CHECK (scope IN ('radio', 'daily_mix', 'taste')),
action text NOT NULL
CONSTRAINT recommendation_tuning_audit_action_check
CHECK (action IN ('update', 'reset')),
changes jsonb NOT NULL
);
CREATE INDEX recommendation_tuning_audit_changed_at_idx
ON recommendation_tuning_audit (changed_at DESC);
+23 -2
View File
@@ -24,11 +24,32 @@ ORDER BY started_at DESC
LIMIT 1;
-- name: InsertPlayEvent :one
-- pick_kind is non-NULL only for system-playlist plays whose track was
-- found (with a stamped kind) in the user's live snapshot for that
-- variant at ingestion time (#1249, generalized in #1270).
INSERT INTO play_events (
user_id, track_id, session_id, started_at, client_id, source
) VALUES ($1, $2, $3, $4, $5, $6)
user_id, track_id, session_id, started_at, client_id, source, pick_kind
) VALUES ($1, $2, $3, $4, $5, $6, sqlc.narg(pick_kind)::text)
RETURNING *;
-- name: GetSystemPickKindForTrack :one
-- Looks a track up in the user's CURRENT snapshot of the given system
-- variant and returns its pick_kind. Used at play-ingestion time to
-- freeze provenance onto the play_event — snapshots rebuild daily, so
-- attribution can't be reconstructed at read time (#1249/#1270). No
-- row = track not in today's snapshot (caller stores NULL); a row with
-- NULL pick_kind = the variant doesn't stamp (yet). songs_like_artist
-- is non-singleton (up to 3 mixes/user); a track in two of them takes
-- whichever LIMIT 1 hits — acceptable, tier stamps there describe the
-- same eligibility ladder.
SELECT pt.pick_kind
FROM playlist_tracks pt
JOIN playlists p ON p.id = pt.playlist_id
WHERE p.user_id = $1
AND p.system_variant = $2
AND pt.track_id = $3
LIMIT 1;
-- name: UpdatePlayEventEnded :one
-- Closes a play_event by id with the given ended_at, duration, and skip flag.
-- completion_ratio is computed from duration_played_ms and the track duration
+5 -3
View File
@@ -68,8 +68,9 @@ ORDER BY pt.position;
-- name: AppendPlaylistTrack :one
-- Inserts at the next available position. Snapshot fields are copied
-- from the tracks/albums/artists join at insert time. tracks.duration_ms
-- is converted to seconds for the snapshot.
INSERT INTO playlist_tracks (playlist_id, position, track_id, title, artist_name, album_title, duration_sec)
-- is converted to seconds for the snapshot. pick_kind is NULL for manual
-- playlists; the For You builder stamps 'taste'/'fresh' (#1249).
INSERT INTO playlist_tracks (playlist_id, position, track_id, title, artist_name, album_title, duration_sec, pick_kind)
SELECT
sqlc.arg(playlist_id)::uuid,
COALESCE((SELECT MAX(position) + 1 FROM playlist_tracks WHERE playlist_id = sqlc.arg(playlist_id)::uuid), 0),
@@ -77,7 +78,8 @@ SELECT
t.title,
artists.name,
albums.title,
(t.duration_ms / 1000)::integer
(t.duration_ms / 1000)::integer,
sqlc.narg(pick_kind)::text
FROM tracks t
JOIN albums ON albums.id = t.album_id
JOIN artists ON artists.id = t.artist_id
+46 -10
View File
@@ -1,22 +1,58 @@
-- Recommendation observability (#796 phase 4). Per-source play outcomes so the
-- operator can see whether each recommendation surface is landing and tune the
-- taste weights. Source is stamped on play_events when a play is launched from
-- a system-playlist surface ('for_you' | 'discover' | the discovery mixes);
-- NULL for library / radio / user-playlist plays, which are excluded here.
-- a recommendation surface; NULL means the user picked the track manually —
-- those rows are INCLUDED here as the baseline control group the surfaces are
-- judged against (milestone #127: delta-vs-baseline is what makes the numbers
-- actionable). Raw source strings are bucketed into stable surface families in
-- the Go handler; completion_n is carried so family merges can weight
-- avg_completion correctly.
-- name: RecommendationSourceMetricsForUser :many
-- $1 user_id, $2 window_days. plays/skips are counts; avg_completion is the
-- mean completion ratio over plays that recorded one (0 when none did).
-- name: RecommendationWeeklyTrends :many
-- Weekly per-source outcome series for the tuning lab's trend view
-- (#1251). Aggregated across ALL users: the tuning knobs are global,
-- so judging a knob turn needs global outcomes — rows carry rates
-- only, no track or user identity. NULL-source (manual) rows are
-- included as the baseline family.
--
-- taste_hits counts plays whose track's artist has a positive weight
-- in that user's CURRENT taste profile — the "cheap recompute" option:
-- retroactive over the whole window, at the cost of drift (the profile
-- is today's, the play may be weeks old). Good enough to read whether
-- a surface is feeding taste-fitting tracks.
-- $1 window in weeks.
SELECT
date_trunc('week', pe.started_at)::date AS week_start,
pe.source,
count(*)::bigint AS plays,
count(*) FILTER (WHERE pe.was_skipped)::bigint AS skips,
COALESCE(
avg(pe.completion_ratio) FILTER (WHERE pe.completion_ratio IS NOT NULL),
0)::float8 AS avg_completion
count(pe.completion_ratio)::bigint AS completion_n,
COALESCE(avg(pe.completion_ratio), 0)::float8 AS avg_completion,
count(*) FILTER (WHERE tpa.artist_id IS NOT NULL)::bigint AS taste_hits
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
LEFT JOIN taste_profile_artists tpa
ON tpa.user_id = pe.user_id
AND tpa.artist_id = t.artist_id
AND tpa.weight > 0
WHERE pe.started_at > now() - (sqlc.arg(weeks)::int * INTERVAL '1 week')
GROUP BY 1, 2
ORDER BY 1, 2;
-- name: RecommendationSourceMetricsForUser :many
-- $1 user_id, $2 window_days. plays/skips are counts; avg_completion is the
-- mean completion ratio over the completion_n plays that recorded one.
-- pick_kind splits For You plays into taste/fresh/unattributed (#1249);
-- it is NULL for every other source, so those still group to one row.
SELECT
pe.source,
pe.pick_kind,
count(*)::bigint AS plays,
count(*) FILTER (WHERE pe.was_skipped)::bigint AS skips,
count(pe.completion_ratio)::bigint AS completion_n,
COALESCE(avg(pe.completion_ratio), 0)::float8 AS avg_completion
FROM play_events pe
WHERE pe.user_id = $1
AND pe.source IS NOT NULL
AND pe.started_at > now() - ($2::float8 * INTERVAL '1 day')
GROUP BY pe.source
GROUP BY pe.source, pe.pick_kind
ORDER BY plays DESC;
@@ -0,0 +1,61 @@
-- Recommendation tuning lab queries (#1250). Seeding happens via the
-- recsettings boot reconcile; shipped defaults live in Go only.
-- name: UpsertWeightProfileDefaults :exec
-- Boot reconcile: insert the shipped defaults for a profile if the row
-- doesn't exist yet. Never overwrites operator-tuned values.
INSERT INTO recommendation_weight_profiles (
profile, base_weight, like_boost, recency_weight, skip_penalty,
jitter_magnitude, context_weight, similarity_weight, taste_weight
) VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9)
ON CONFLICT (profile) DO NOTHING;
-- name: ListWeightProfiles :many
SELECT * FROM recommendation_weight_profiles ORDER BY profile;
-- name: UpdateWeightProfile :one
UPDATE recommendation_weight_profiles
SET base_weight = $2,
like_boost = $3,
recency_weight = $4,
skip_penalty = $5,
jitter_magnitude = $6,
context_weight = $7,
similarity_weight = $8,
taste_weight = $9,
updated_at = now()
WHERE profile = $1
RETURNING *;
-- name: UpsertTasteTuningDefaults :exec
INSERT INTO taste_tuning (
singleton, half_life_days, engagement_hard_skip,
engagement_neutral, engagement_full
) VALUES (true, $1, $2, $3, $4)
ON CONFLICT (singleton) DO NOTHING;
-- name: GetTasteTuning :one
SELECT * FROM taste_tuning WHERE singleton = true;
-- name: UpdateTasteTuning :one
UPDATE taste_tuning
SET half_life_days = $1,
engagement_hard_skip = $2,
engagement_neutral = $3,
engagement_full = $4,
updated_at = now()
WHERE singleton = true
RETURNING *;
-- name: InsertTuningAudit :exec
-- changes is a jsonb array of {field, old, new} objects.
INSERT INTO recommendation_tuning_audit (scope, action, changes)
VALUES ($1, $2, $3);
-- name: ListTuningAudit :many
-- Newest first; consumed by the metrics trend view (#1251) to annotate
-- knob turns on the timeline.
SELECT id, changed_at, scope, action, changes
FROM recommendation_tuning_audit
ORDER BY changed_at DESC, id DESC
LIMIT $1;
+148 -58
View File
@@ -8,6 +8,11 @@
-- name: ListDeepCutsTracks :many
-- #419 Deep Cuts: low-play tracks (<=2 plays) from artists the user
-- has liked OR played heavily (>=5 non-skip plays across the artist).
-- Tracks the user has skipped twice or more don't qualify (#1257):
-- eligibility used to count only unskipped plays, so the most actively
-- rejected tracks read as "barely heard" and kept being re-offered.
-- Threshold 2 so a single accidental skip doesn't banish a track;
-- passive-signal only (no dislike UI, rule #101).
-- $1 user_id, $2 date string.
WITH affinity_artists AS (
SELECT artist_id FROM general_likes_artists WHERE user_id = $1
@@ -23,12 +28,20 @@ play_counts AS (
FROM play_events
WHERE user_id = $1 AND was_skipped = false
GROUP BY track_id
),
skip_counts AS (
SELECT track_id, COUNT(*) AS c
FROM play_events
WHERE user_id = $1 AND was_skipped = true
GROUP BY track_id
)
SELECT t.id, t.album_id, t.artist_id
FROM tracks t
JOIN affinity_artists aa ON aa.artist_id = t.artist_id
LEFT JOIN play_counts pc ON pc.track_id = t.id
LEFT JOIN skip_counts sc ON sc.track_id = t.id
WHERE COALESCE(pc.c, 0) <= 2
AND COALESCE(sc.c, 0) < 2
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
@@ -37,75 +50,112 @@ SELECT t.id, t.album_id, t.artist_id
LIMIT 200;
-- name: ListRediscoverTracks :many
-- #420 Rediscover: tracks the user played a lot (>=5 non-skip) but
-- has drifted away from. Tiered so a young library still gets a mix:
-- tier 0 not played in the last 6 months (true rediscovery)
-- tier 1 (only if tier 0 empty) not played in the last 30 days
-- Ordered by historical affection. $1 user_id.
-- #420 Rediscover: tracks the user played a lot but has drifted away
-- from. One blended pool (issue #1246: the old two-tier UNION was
-- all-or-nothing — a single ">=6 months cold" row suppressed the whole
-- ">=30 days" tier, which is how a one-song playlist shipped; the 6mo
-- tier was also a strict subset of the 30d tier). Eligibility is >=3
-- non-skip plays and >=30 days cold; the bar sits at 3 rather than 5
-- so a weeks-old library still fields a pool — on young histories the
-- >=5-play tracks are precisely the ones still in rotation. Ordering
-- prefers true rediscoveries (>=6 months cold), then strong affection
-- (>=5 plays), then raw play count; the producer's minimum floor
-- decides whether the pool is big enough to ship at all. $1 user_id.
WITH stats AS (
SELECT pe.track_id, COUNT(*) AS c, MAX(pe.started_at) AS last_at
FROM play_events pe
WHERE pe.user_id = $1 AND pe.was_skipped = false
GROUP BY pe.track_id
),
deep AS (
SELECT t.id, t.album_id, t.artist_id, s.c, 0 AS tier
FROM tracks t
JOIN stats s ON s.track_id = t.id
WHERE s.c >= 5
AND s.last_at <= now() - interval '6 months'
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
),
shallow AS (
SELECT t.id, t.album_id, t.artist_id, s.c, 1 AS tier
SELECT t.id, t.album_id, t.artist_id
FROM tracks t
JOIN stats s ON s.track_id = t.id
WHERE s.c >= 5
WHERE s.c >= 3
AND s.last_at <= now() - interval '30 days'
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
-- Declined-resurfacing cooldown (#1258): a skip on a
-- rediscover-sourced play is the user explicitly saying "I've moved
-- on" to the exact invitation this mix extends — such tracks sit
-- out ~90 days instead of re-qualifying the next day. Passive
-- signal only (rule #101).
AND NOT EXISTS (
SELECT 1 FROM play_events rp
WHERE rp.user_id = $1
AND rp.track_id = t.id
AND rp.source = 'rediscover'
AND rp.was_skipped = true
AND rp.started_at > now() - interval '90 days'
)
SELECT id, album_id, artist_id
FROM (
SELECT id, album_id, artist_id, c, tier FROM deep
UNION ALL
SELECT id, album_id, artist_id, c, tier FROM shallow
WHERE NOT EXISTS (SELECT 1 FROM deep)
) u
ORDER BY tier, c DESC, id
ORDER BY (s.last_at <= now() - interval '6 months') DESC,
(s.c >= 5) DESC,
s.c DESC, t.id
LIMIT 200;
-- name: ListNewForYouTracks :many
-- #421 New for you: tracks from albums added in the last 30 days
-- whose artist the user has liked OR played (>=3 non-skip). Album-
-- coherent (newest album first, then disc/track) — the producer does
-- NOT diversity-cap these; they're meant as whole-album discovery.
-- $1 user_id.
WITH affinity_artists AS (
SELECT artist_id FROM general_likes_artists WHERE user_id = $1
-- #421 New for you, tiered per project rule #131 (#1267). The old
-- single hard rule (added <30d AND direct affinity) had two failure
-- modes with one root: no notion of consumption — the album you
-- devoured in week one crowded the mix for three more weeks, and
-- after a quiet month nothing qualified at all.
--
-- "Consumed" = any track attempted for >=30s (the First Listens
-- threshold, #1268): once you've meaningfully engaged, the album is
-- no longer news. Applies to every tier — played albums leave the
-- mix the next build.
--
-- tier 1 the exact desire: unconsumed albums added <30d by
-- direct-affinity artists (liked, or >=3 attempted plays)
-- tier 2 step back a little: unconsumed affinity albums from the
-- wider 30-90d window — added while you weren't looking
-- tier 3 step back more: any unconsumed album added <90d
-- regardless of affinity, newest first
--
-- Album-coherent within tiers (newest album first, then disc/track);
-- the producer fills top-down and rotates within tier blocks. The
-- tier lands on playlist_tracks.pick_kind so metrics can price what
-- each step-back trades away. $1 user_id.
WITH attempted AS (
SELECT DISTINCT pe.track_id
FROM play_events pe
WHERE pe.user_id = $1 AND pe.duration_played_ms >= 30000
),
consumed_albums AS (
SELECT DISTINCT t.album_id
FROM attempted a JOIN tracks t ON t.id = a.track_id
),
affinity_artists AS (
SELECT gla.artist_id FROM general_likes_artists gla WHERE gla.user_id = $1
UNION
SELECT t.artist_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1 AND pe.was_skipped = false
FROM attempted a JOIN tracks t ON t.id = a.track_id
GROUP BY t.artist_id
HAVING COUNT(*) >= 3
),
albums_tiered AS (
SELECT al.id AS album_id, al.created_at,
CASE
WHEN al.created_at >= now() - interval '30 days'
AND al.artist_id IN (SELECT artist_id FROM affinity_artists) THEN 1
WHEN al.artist_id IN (SELECT artist_id FROM affinity_artists) THEN 2
ELSE 3
END AS tier
FROM albums al
WHERE al.created_at >= now() - interval '90 days'
AND al.id NOT IN (SELECT album_id FROM consumed_albums)
)
SELECT t.id, t.album_id, t.artist_id
SELECT t.id, t.album_id, t.artist_id, alt.tier::int AS tier
FROM tracks t
JOIN albums al ON al.id = t.album_id
JOIN affinity_artists aa ON aa.artist_id = al.artist_id
WHERE al.created_at >= now() - interval '30 days'
AND NOT EXISTS (
JOIN albums_tiered alt ON alt.album_id = t.album_id
WHERE NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
ORDER BY al.created_at DESC, t.disc_number NULLS FIRST, t.track_number NULLS FIRST
LIMIT 200;
ORDER BY alt.tier, alt.created_at DESC, t.album_id,
t.disc_number NULLS FIRST, t.track_number NULLS FIRST
LIMIT 300;
-- name: ListOnThisDayTracks :many
-- #422 On This Day: tracks the user played around this calendar date
@@ -115,13 +165,23 @@ SELECT t.id, t.album_id, t.artist_id
-- and window ±7→±10 so it surfaces on a months-old library instead
-- of needing a full year of history; still skips cleanly (no rows →
-- no playlist) when there's no qualifying history yet.
-- Day-of-year distance is circular (#1256): plain ABS made Dec 28 vs
-- Jan 3 read as 359 days apart, silently gutting the window for ~3
-- weeks around every New Year — precisely when holiday nostalgia is
-- the point. LEAST(d, 365-d) wraps the boundary; leap-year drift of
-- ±1 is absorbed by the ±10 window. The build date ($2, already here
-- for the md5 rotation) anchors "today" instead of now() so the
-- window is testable and consistent with the mix's daily determinism.
-- $1 user_id, $2 date string.
WITH windowed AS (
SELECT track_id, COUNT(*) AS c
FROM play_events
WHERE user_id = $1 AND was_skipped = false
AND started_at < now() - interval '30 days'
AND ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM now())) <= 10
AND LEAST(
ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM $2::date)),
365 - ABS(EXTRACT(DOY FROM started_at) - EXTRACT(DOY FROM $2::date))
) <= 10
GROUP BY track_id
)
SELECT t.id, t.album_id, t.artist_id
@@ -135,35 +195,65 @@ SELECT t.id, t.album_id, t.artist_id
LIMIT 200;
-- name: ListFirstListensTracks :many
-- #423 First Listens: albums the user has never played any track of.
-- Tiered: liked-artist albums first, then played-artist albums, then
-- the rest — album-coherent within each tier. Not diversity-capped
-- (whole-album discovery). $1 user_id.
WITH heard_albums AS (
SELECT DISTINCT t.album_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
-- #423 First Listens: songs the user has never heard or even
-- attempted. "Attempted" is track-level with a >=30s listen threshold
-- (#1268, operator decision 2026-07-03) — the old version disqualified
-- a whole album on ANY play_event, so a 2-second accidental brush
-- banished it permanently.
--
-- tier 1 the exact desire: albums with ZERO attempted tracks
-- tier 2 step back: barely-attempted albums (<=25% of tracks
-- reached 30s) — brushed, never explored. The already-
-- attempted tracks themselves are excluded; they're not
-- first listens.
--
-- Within each tier, affinity-ordered as before: liked artist, then
-- attempted-played artist (>=30s — skip-only contact isn't trust),
-- then the rest; album-coherent. The tier lands on
-- playlist_tracks.pick_kind via the producer. $1 user_id.
WITH attempted AS (
SELECT DISTINCT pe.track_id
FROM play_events pe
WHERE pe.user_id = $1 AND pe.duration_played_ms >= 30000
),
played_artists AS (
album_attempts AS (
SELECT t.album_id,
COUNT(a.track_id) AS attempted_count,
COUNT(*) AS track_count
FROM tracks t
LEFT JOIN attempted a ON a.track_id = t.id
GROUP BY t.album_id
),
attempted_artists AS (
SELECT DISTINCT t.artist_id
FROM play_events pe JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
FROM attempted a JOIN tracks t ON t.id = a.track_id
),
albums_tiered AS (
SELECT aa.album_id,
CASE
WHEN aa.attempted_count = 0 THEN 1
WHEN aa.attempted_count::float / GREATEST(aa.track_count, 1) <= 0.25 THEN 2
END AS tier
FROM album_attempts aa
)
SELECT t.id, t.album_id, t.artist_id
SELECT t.id, t.album_id, t.artist_id, alt.tier::int AS tier
FROM tracks t
JOIN albums al ON al.id = t.album_id
WHERE NOT EXISTS (SELECT 1 FROM heard_albums h WHERE h.album_id = al.id)
JOIN albums_tiered alt ON alt.album_id = al.id
WHERE alt.tier IS NOT NULL
AND NOT EXISTS (SELECT 1 FROM attempted a WHERE a.track_id = t.id)
AND NOT EXISTS (
SELECT 1 FROM lidarr_quarantine q
WHERE q.user_id = $1 AND q.track_id = t.id
)
ORDER BY
alt.tier,
(CASE
WHEN EXISTS (
SELECT 1 FROM general_likes_artists gla
WHERE gla.user_id = $1 AND gla.artist_id = al.artist_id
) THEN 0
WHEN al.artist_id IN (SELECT artist_id FROM played_artists) THEN 1
WHEN al.artist_id IN (SELECT artist_id FROM attempted_artists) THEN 1
ELSE 2
END),
al.id, t.disc_number NULLS FIRST, t.track_number NULLS FIRST
+71 -18
View File
@@ -47,14 +47,29 @@ UPDATE system_playlist_runs
UPDATE system_playlist_runs SET in_flight = false WHERE in_flight = true;
-- name: PickSeedArtists :many
-- Top-5 most-engaged distinct artist candidates in the user's last 7
-- days. The Go-side picker (pickSeedArtistsForDay) shuffles these
-- daily-deterministically and takes the first 3 so the set of
-- "Songs like X" mixes rotates day-to-day.
-- Score = unskipped-play count + 5 if user has liked the artist.
WITH plays AS (
-- Top-5 most-engaged distinct artist candidates, tiered so the
-- "Songs like X" mixes never silently vanish (#1255): the old hard
-- 7-day window emptied the seed pool after a quiet week, and the
-- daily atomic-replace build then deleted every existing mix until
-- the user played something again. Same fallback shape as
-- PickTopPlayedTracksForUser (For You's seeds):
-- tier 0 engagement in the last 7 days
-- tier 1 (only if tier 0 empty) last 30 days
-- tier 2 (only if tiers 0+1 empty) all-time
-- tier 3 (only if 0-2 empty) liked artists with no play history
-- All returned rows share one tier; produceSeedMixes maps it onto the
-- rule-#131 pick-kind ladder and stamps the built tracks, so metrics
-- can compare mixes seeded from fresh vs stale engagement.
-- Score = unskipped-play count + 5 if user has liked the artist. The
-- Go-side picker (pickSeedArtistsForDay) shuffles the 5
-- daily-deterministically and takes 3 so the mix set rotates.
WITH liked AS (
SELECT gla.artist_id FROM general_likes_artists gla WHERE gla.user_id = $1
),
recent7 AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
0 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
@@ -62,14 +77,52 @@ WITH plays AS (
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
liked AS (
SELECT artist_id FROM general_likes_artists WHERE user_id = $1
recent30 AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
1 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
AND pe.started_at > now() - INTERVAL '30 days'
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
alltime AS (
SELECT t.artist_id,
COUNT(*) FILTER (WHERE pe.was_skipped = false) AS play_count,
2 AS tier
FROM play_events pe
JOIN tracks t ON t.id = pe.track_id
WHERE pe.user_id = $1
AND t.artist_id IS NOT NULL
GROUP BY t.artist_id
),
likedonly AS (
SELECT l.artist_id, 0::bigint AS play_count, 3 AS tier
FROM liked l
),
chosen AS (
SELECT artist_id, play_count, tier FROM recent7
UNION ALL
SELECT artist_id, play_count, tier FROM recent30
WHERE NOT EXISTS (SELECT 1 FROM recent7)
UNION ALL
SELECT artist_id, play_count, tier FROM alltime
WHERE NOT EXISTS (SELECT 1 FROM recent7)
AND NOT EXISTS (SELECT 1 FROM recent30)
UNION ALL
SELECT artist_id, play_count, tier FROM likedonly
WHERE NOT EXISTS (SELECT 1 FROM recent7)
AND NOT EXISTS (SELECT 1 FROM recent30)
AND NOT EXISTS (SELECT 1 FROM alltime)
)
SELECT p.artist_id,
(p.play_count + CASE WHEN l.artist_id IS NOT NULL THEN 5 ELSE 0 END)::bigint AS score
FROM plays p
LEFT JOIN liked l ON l.artist_id = p.artist_id
ORDER BY score DESC, p.artist_id
SELECT c.artist_id,
(c.play_count + CASE WHEN l.artist_id IS NOT NULL THEN 5 ELSE 0 END)::bigint AS score,
c.tier::int AS tier
FROM chosen c
LEFT JOIN liked l ON l.artist_id = c.artist_id
ORDER BY score DESC, c.artist_id
LIMIT 5;
-- name: PickTopPlayedTracksForUser :many
@@ -78,10 +131,10 @@ SELECT p.artist_id,
-- tier 1 (only if tier 0 empty) all-time top non-skip plays
-- tier 2 (only if tiers 0+1 empty) liked tracks
-- Returns up to 5 ids; tie-break by track_id for determinism. The
-- Go-side picker (pickForYouSeedForDay) rotates one per day via
-- userIDHash. Widened from a hard 7-day window, which made For-You
-- disappear after a week of not listening and never recover on a
-- self-hosted library with sparse history.
-- Go-side picker (pickDailySeeds) draws the day's seeds from these.
-- Widened from a hard 7-day window, which made For-You disappear
-- after a week of not listening and never recover on a self-hosted
-- library with sparse history.
WITH recent AS (
SELECT t.id, COUNT(*) AS c, 0 AS tier
FROM play_events pe
+5
View File
@@ -62,6 +62,11 @@ var dataTables = []string{
// (never recreates the singleton, seeded once by 0018); ResetDB
// resets its counter via UPDATE below instead.
"cover_art_provider_settings",
// recsettings.New reconciles shipped defaults on every construction
// (#1250), so truncating gives each test pristine tuning values.
"recommendation_weight_profiles",
"taste_tuning",
"recommendation_tuning_audit",
"tracks",
"albums",
"artists",
+46
View File
@@ -97,6 +97,36 @@ var systemPlaylistSources = map[string]bool{
"songs_like_artist": true,
}
// lookupSystemPickKind resolves a system-playlist play's pick_kind
// (taste/fresh, Discover's bucket, or a rule-#131 tier) from the user's
// live snapshot of that variant at ingestion time. Attribution must be
// frozen now — snapshots rebuild daily, so it can't be reconstructed at
// metrics-read time (#1249, generalized in #1270). Source strings
// double as system_variant values (see systemPlaylistSources). Returns
// nil (unattributed) when the track isn't in the current snapshot —
// plays predating the feature, or an offline replay arriving after the
// track rotated out — and when the variant doesn't stamp yet. Real DB
// errors propagate and fail the caller's transaction.
func lookupSystemPickKind(
ctx context.Context,
q *dbq.Queries,
userID, trackID pgtype.UUID,
variant string,
) (*string, error) {
kind, err := q.GetSystemPickKindForTrack(ctx, dbq.GetSystemPickKindForTrackParams{
UserID: userID,
SystemVariant: &variant,
TrackID: trackID,
})
if errors.Is(err, pgx.ErrNoRows) {
return nil, nil
}
if err != nil {
return nil, err
}
return kind, nil
}
// RecordPlayStartedWithSource is RecordPlayStarted plus a `source`
// tag identifying which surface the play came from. When source is a
// known system-playlist kind the track is appended to that user's
@@ -127,6 +157,13 @@ func (w *Writer) RecordPlayStartedWithSource(
if source != "" {
sourcePtr = &source
}
var pickKind *string
if systemPlaylistSources[source] {
pickKind, err = lookupSystemPickKind(ctx, q, userID, trackID, source)
if err != nil {
return err
}
}
ev, err := q.InsertPlayEvent(ctx, dbq.InsertPlayEventParams{
UserID: userID,
TrackID: trackID,
@@ -134,6 +171,7 @@ func (w *Writer) RecordPlayStartedWithSource(
StartedAt: pgtype.Timestamptz{Time: at, Valid: true},
ClientID: clientIDPtr,
Source: sourcePtr,
PickKind: pickKind,
})
if err != nil {
return err
@@ -395,6 +433,13 @@ func (w *Writer) RecordOfflinePlay(
if source != "" {
sourcePtr = &source
}
var pickKind *string
if systemPlaylistSources[source] {
pickKind, err = lookupSystemPickKind(ctx, q, userID, trackID, source)
if err != nil {
return err
}
}
ev, err := q.InsertPlayEvent(ctx, dbq.InsertPlayEventParams{
UserID: userID,
TrackID: trackID,
@@ -402,6 +447,7 @@ func (w *Writer) RecordOfflinePlay(
StartedAt: pgtype.Timestamptz{Time: at, Valid: true},
ClientID: clientIDPtr,
Source: sourcePtr,
PickKind: pickKind,
})
if err != nil {
return err
+120
View File
@@ -488,3 +488,123 @@ func TestRecordPlayStarted_NoRotationWithoutSource(t *testing.T) {
t.Errorf("expected no rotation row for a source-less play")
}
}
// seedSystemSnapshot creates a system playlist of the given variant
// containing the fixture track with the given pick_kind, mimicking
// what the builder persists (#1249, generalized #1270).
func seedSystemSnapshot(t *testing.T, f fixture, variant, pickKind string) {
t.Helper()
pl, err := f.q.CreateSystemPlaylist(context.Background(), dbq.CreateSystemPlaylistParams{
UserID: f.user, Name: variant, SystemVariant: &variant,
})
if err != nil {
t.Fatalf("CreateSystemPlaylist: %v", err)
}
if _, err := f.q.AppendPlaylistTrack(context.Background(), dbq.AppendPlaylistTrackParams{
PlaylistID: pl.ID, TrackID: f.track, PickKind: &pickKind,
}); err != nil {
t.Fatalf("AppendPlaylistTrack: %v", err)
}
}
func TestRecordPlayStartedWithSource_StampsForYouPickKind(t *testing.T) {
f := newFixture(t, 200_000)
seedSystemSnapshot(t, f, "for_you", "fresh")
res, err := f.w.RecordPlayStartedWithSource(
context.Background(), f.user, f.track, "c", "for_you", time.Now().UTC(),
)
if err != nil {
t.Fatalf("RecordPlayStartedWithSource: %v", err)
}
got, err := f.q.GetPlayEventByID(context.Background(), res.PlayEventID)
if err != nil {
t.Fatalf("get: %v", err)
}
if got.PickKind == nil || *got.PickKind != "fresh" {
t.Errorf("pick_kind = %v, want fresh (frozen from live snapshot)", got.PickKind)
}
}
func TestRecordPlayStartedWithSource_NoSnapshotMatch_PickKindNull(t *testing.T) {
// A for_you play whose track isn't in the current snapshot (rotated
// out, or no snapshot exists) stays unattributed rather than guessing.
f := newFixture(t, 200_000)
res, err := f.w.RecordPlayStartedWithSource(
context.Background(), f.user, f.track, "c", "for_you", time.Now().UTC(),
)
if err != nil {
t.Fatalf("RecordPlayStartedWithSource: %v", err)
}
got, err := f.q.GetPlayEventByID(context.Background(), res.PlayEventID)
if err != nil {
t.Fatalf("get: %v", err)
}
if got.PickKind != nil {
t.Errorf("pick_kind = %q, want NULL (track not in snapshot)", *got.PickKind)
}
}
func TestRecordPlayStartedWithSource_CrossVariant_PickKindNull(t *testing.T) {
// Attribution is variant-scoped: even with a For You snapshot
// containing the track, a play from a different surface must not
// pick up For You's stamp — it reads its OWN variant's snapshot,
// which here doesn't exist.
f := newFixture(t, 200_000)
seedSystemSnapshot(t, f, "for_you", "taste")
res, err := f.w.RecordPlayStartedWithSource(
context.Background(), f.user, f.track, "c", "discover", time.Now().UTC(),
)
if err != nil {
t.Fatalf("RecordPlayStartedWithSource: %v", err)
}
got, err := f.q.GetPlayEventByID(context.Background(), res.PlayEventID)
if err != nil {
t.Fatalf("get: %v", err)
}
if got.PickKind != nil {
t.Errorf("pick_kind = %q, want NULL for a cross-variant play", *got.PickKind)
}
}
func TestRecordPlayStartedWithSource_StampsDiscoverBucket(t *testing.T) {
// Discover stamps its candidate bucket (#1270); a discover play whose
// track is in the live snapshot freezes that bucket onto the play.
f := newFixture(t, 200_000)
seedSystemSnapshot(t, f, "discover", "dormant")
res, err := f.w.RecordPlayStartedWithSource(
context.Background(), f.user, f.track, "c", "discover", time.Now().UTC(),
)
if err != nil {
t.Fatalf("RecordPlayStartedWithSource: %v", err)
}
got, err := f.q.GetPlayEventByID(context.Background(), res.PlayEventID)
if err != nil {
t.Fatalf("get: %v", err)
}
if got.PickKind == nil || *got.PickKind != "dormant" {
t.Errorf("pick_kind = %v, want dormant (frozen from live snapshot)", got.PickKind)
}
}
func TestRecordOfflinePlay_StampsForYouPickKind(t *testing.T) {
// The offline replay path threads source the same way; a replayed
// for_you play whose track is still in the snapshot gets attributed.
f := newFixture(t, 200_000)
seedSystemSnapshot(t, f, "for_you", "taste")
at := time.Now().UTC().Add(-time.Hour)
if err := f.w.RecordOfflinePlay(
context.Background(), f.user, f.track, "c", "for_you", at, 180_000,
); err != nil {
t.Fatalf("RecordOfflinePlay: %v", err)
}
var pickKind *string
if err := f.pool.QueryRow(context.Background(),
`SELECT pick_kind FROM play_events WHERE user_id = $1 AND track_id = $2`,
f.user, f.track,
).Scan(&pickKind); err != nil {
t.Fatalf("select: %v", err)
}
if pickKind == nil || *pickKind != "taste" {
t.Errorf("pick_kind = %v, want taste", pickKind)
}
}
+18 -5
View File
@@ -20,11 +20,15 @@ const (
// discoverTrack is the common shape used by the bucket allocator. The
// three sqlc-generated row types collapse into this internal struct so
// downstream functions don't need to be generic over them.
// downstream functions don't need to be generic over them. PickKind is
// the originating bucket (dormant/cross_user/random, #1270), stamped in
// the row adapters so it survives the interleave — bucket identity is
// what makes the 40/30/30 allocation measurable in the metrics.
type discoverTrack struct {
ID pgtype.UUID
AlbumID pgtype.UUID
ArtistID pgtype.UUID
PickKind string
}
// buildDiscoverCandidates assembles the Discover playlist track list.
@@ -91,7 +95,7 @@ func buildDiscoverCandidates(ctx context.Context, q *dbq.Queries, logger *slog.L
// already gave us the ranking.
ranked := make([]rankedCandidate, 0, len(out))
for _, t := range out {
ranked = append(ranked, rankedCandidate{TrackID: t.ID})
ranked = append(ranked, rankedCandidate{TrackID: t.ID, PickKind: t.PickKind})
}
return ranked, nil
}
@@ -99,7 +103,10 @@ func buildDiscoverCandidates(ctx context.Context, q *dbq.Queries, logger *slog.L
func dormantRowsToTracks(rows []dbq.ListDormantArtistTracksForDiscoverRow) []discoverTrack {
out := make([]discoverTrack, len(rows))
for i, r := range rows {
out[i] = discoverTrack{ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID}
out[i] = discoverTrack{
ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID,
PickKind: pickKindDormant,
}
}
return out
}
@@ -107,7 +114,10 @@ func dormantRowsToTracks(rows []dbq.ListDormantArtistTracksForDiscoverRow) []dis
func crossUserRowsToTracks(rows []dbq.ListCrossUserLikedTracksForDiscoverRow) []discoverTrack {
out := make([]discoverTrack, len(rows))
for i, r := range rows {
out[i] = discoverTrack{ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID}
out[i] = discoverTrack{
ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID,
PickKind: pickKindCrossUser,
}
}
return out
}
@@ -115,7 +125,10 @@ func crossUserRowsToTracks(rows []dbq.ListCrossUserLikedTracksForDiscoverRow) []
func randomRowsToTracks(rows []dbq.ListRandomUnheardTracksForDiscoverRow) []discoverTrack {
out := make([]discoverTrack, len(rows))
for i, r := range rows {
out[i] = discoverTrack{ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID}
out[i] = discoverTrack{
ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID,
PickKind: pickKindRandom,
}
}
return out
}
+24
View File
@@ -161,3 +161,27 @@ func TestInterleaveBuckets_DedupsAcrossBuckets(t *testing.T) {
}
}
}
func TestInterleaveBuckets_PreservesBucketPickKind(t *testing.T) {
// Bucket provenance (#1270) is stamped on discoverTrack before the
// interleave and must survive it — including dedup, where a track in
// two buckets keeps the stamp of the bucket it was taken from.
dormant := []discoverTrack{
{ID: uuidN(1), PickKind: pickKindDormant},
{ID: uuidN(2), PickKind: pickKindDormant},
}
random := []discoverTrack{
{ID: uuidN(1), PickKind: pickKindRandom}, // shared with dormant
{ID: uuidN(3), PickKind: pickKindRandom},
}
got := interleaveBuckets(dormant, nil, random)
if len(got) != 3 {
t.Fatalf("len = %d, want 3", len(got))
}
want := map[byte]string{1: pickKindDormant, 2: pickKindDormant, 3: pickKindRandom}
for _, tr := range got {
if w := want[tr.ID.Bytes[15]]; tr.PickKind != w {
t.Errorf("track %d pick_kind = %q, want %q", tr.ID.Bytes[15], tr.PickKind, w)
}
}
}
+195 -8
View File
@@ -99,7 +99,7 @@ func TestPickHeadAndTail_SmallPool(t *testing.T) {
makeCand(2, 11, 101, 0.9),
makeCand(3, 12, 102, 0.8),
}
got := pickHeadAndTail(in, testUserID, "2026-05-07", time.Now(), 5, 2)
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 5, 2)
if len(got) != 3 {
t.Errorf("len = %d, want 3 (pool too small for head/tail split)", len(got))
}
@@ -112,7 +112,7 @@ func TestPickHeadAndTail_ExactlyTotal(t *testing.T) {
for i := 0; i < 7; i++ {
in = append(in, makeCand(i+1, i+1, i+1, float64(7-i)))
}
got := pickHeadAndTail(in, testUserID, "2026-05-07", time.Now(), 5, 2)
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 5, 2)
if len(got) != 7 {
t.Errorf("len = %d, want 7 (pool == total)", len(got))
}
@@ -126,7 +126,7 @@ func TestPickHeadAndTail_HeadAndTailSplit(t *testing.T) {
for i := 0; i < 100; i++ {
in = append(in, makeCand(i+1, i+1, i+1, float64(100-i)))
}
got := pickHeadAndTail(in, testUserID, "2026-05-07", time.Now(), 20, 5)
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 20, 5)
if len(got) != 25 {
t.Errorf("len = %d, want 25 (20 head + 5 tail)", len(got))
}
@@ -139,8 +139,8 @@ func TestPickHeadAndTail_Determinism(t *testing.T) {
in = append(in, makeCand(i+1, i+1, i+1, float64(100-i)))
}
now := time.Now()
got1 := pickHeadAndTail(in, testUserID, "2026-05-07", now, 20, 5)
got2 := pickHeadAndTail(in, testUserID, "2026-05-07", now, 20, 5)
got1 := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", now, 20, 5)
got2 := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", now, 20, 5)
if len(got1) != len(got2) {
t.Fatalf("len mismatch: %d vs %d", len(got1), len(got2))
}
@@ -161,8 +161,8 @@ func TestPickHeadAndTail_HeadStable_TailVariesAcrossDays(t *testing.T) {
in = append(in, makeCand(i+1, i+1, i+1, float64(100-i)))
}
now := time.Now()
day1 := pickHeadAndTail(in, testUserID, "2026-05-07", now, 20, 5)
day2 := pickHeadAndTail(in, testUserID, "2026-05-08", now, 20, 5)
day1 := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", now, 20, 5)
day2 := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-08", now, 20, 5)
if len(day1) != 25 || len(day2) != 25 {
t.Fatalf("len mismatch: day1=%d day2=%d", len(day1), len(day2))
}
@@ -205,7 +205,7 @@ func TestPickHeadAndTail_TailFromBeyond2xHeadN(t *testing.T) {
sim := float64(50 - i)
in = append(in, makeCand(i+1, i+1, i+1, sim))
}
got := pickHeadAndTail(in, testUserID, "2026-05-07", time.Now(), 5, 3)
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 5, 3)
if len(got) != 8 {
t.Fatalf("len = %d, want 8 (5 head + 3 tail)", len(got))
}
@@ -243,3 +243,190 @@ func TestPickTopN_DiversityCap(t *testing.T) {
t.Errorf("len = %d, want 3 (artist 100 capped at 3)", len(got))
}
}
func TestPickHeadAndTail_MarksPickKinds(t *testing.T) {
// 100-deep pool with headN=20, tailN=5: the 20 head entries are the
// taste picks, the 5 tail entries the freshness injection (#1249).
in := make([]recommendation.Candidate, 0, 100)
for i := 0; i < 100; i++ {
in = append(in, makeCand(i+1, i+1, i+1, float64(100-i)))
}
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 20, 5)
if len(got) != 25 {
t.Fatalf("len = %d, want 25", len(got))
}
for i := 0; i < 20; i++ {
if got[i].PickKind != pickKindTaste {
t.Errorf("head[%d].PickKind = %q, want %q", i, got[i].PickKind, pickKindTaste)
}
}
for i := 20; i < 25; i++ {
if got[i].PickKind != pickKindFresh {
t.Errorf("tail[%d].PickKind = %q, want %q", i-20, got[i].PickKind, pickKindFresh)
}
}
}
func TestPickHeadAndTail_SmallPoolAllTaste(t *testing.T) {
// The small-pool fallback is pure top-N-by-score — that IS the taste
// mechanism, so nothing on this path is an exploration pick.
in := []recommendation.Candidate{
makeCand(1, 10, 100, 1.0),
makeCand(2, 11, 101, 0.9),
makeCand(3, 12, 102, 0.8),
}
got := pickHeadAndTail(in, nil, 1, testUserID, "2026-05-07", time.Now(), 5, 2)
for i, rc := range got {
if rc.PickKind != pickKindTaste {
t.Errorf("got[%d].PickKind = %q, want %q (fallback is all taste)",
i, rc.PickKind, pickKindTaste)
}
}
}
func TestPickTopN_NoPickKind(t *testing.T) {
// Songs-like-X and the discovery mixes don't split; their rows must
// persist pick_kind NULL (empty string here).
in := []recommendation.Candidate{makeCand(1, 10, 100, 1.0)}
got := pickTopN(in, testUserID, "2026-05-07", time.Now(), 25)
if len(got) != 1 || got[0].PickKind != "" {
t.Errorf("pickTopN PickKind = %q, want empty (persists as NULL)", got[0].PickKind)
}
}
func TestHeadQuotas(t *testing.T) {
cases := []struct {
seeds, headN int
want []int
}{
{1, 50, []int{50}},
{2, 50, []int{30, 20}},
{3, 50, []int{25, 15, 10}},
{3, 7, []int{4, 2, 1}}, // rounding remainder → primary seed
}
for _, c := range cases {
got := headQuotas(c.seeds, c.headN)
if len(got) != len(c.want) {
t.Fatalf("headQuotas(%d, %d) len = %d, want %d", c.seeds, c.headN, len(got), len(c.want))
}
sum := 0
for i := range got {
sum += got[i]
if got[i] != c.want[i] {
t.Errorf("headQuotas(%d, %d)[%d] = %d, want %d", c.seeds, c.headN, i, got[i], c.want[i])
}
}
if sum != c.headN {
t.Errorf("headQuotas(%d, %d) sums to %d, want %d", c.seeds, c.headN, sum, c.headN)
}
}
}
func TestPickQuotaHead_BlendsSeeds(t *testing.T) {
// Seed 0's neighborhood out-scores everything (#1269): without
// quotas the head would be all seed-0. With headN=10 and quotas
// [5,3,2], the walk takes the best 5 from seed 0, then seeds 1 and
// 2 get their guaranteed slots from deeper ranks.
in := make([]recommendation.Candidate, 0, 30)
seedOf := map[pgtype.UUID]int{}
for i := 0; i < 30; i++ {
c := makeCand(i+1, i+1, i+1, float64(30-i))
si := 0
if i >= 20 && i < 25 {
si = 1
} else if i >= 25 {
si = 2
}
seedOf[c.Track.ID] = si
in = append(in, c)
}
// Input is already score-descending and cap-free.
head := pickQuotaHead(in, seedOf, 3, 10)
if len(head) != 10 {
t.Fatalf("head len = %d, want 10", len(head))
}
want := []byte{1, 2, 3, 4, 5, 21, 22, 23, 26, 27}
for i, w := range want {
if head[i].Track.ID.Bytes[15] != w {
t.Errorf("head[%d] = track %d, want %d", i, head[i].Track.ID.Bytes[15], w)
}
}
}
func TestPickQuotaHead_ThinSeedSpillsBestFirst(t *testing.T) {
// Seed 1 has a single candidate; its unfilled quota goes to the
// best remaining candidates regardless of seed.
in := make([]recommendation.Candidate, 0, 12)
seedOf := map[pgtype.UUID]int{}
for i := 0; i < 12; i++ {
c := makeCand(i+1, i+1, i+1, float64(12-i))
si := 0
if i == 11 {
si = 1 // the worst-ranked candidate is seed 1's only one
}
seedOf[c.Track.ID] = si
in = append(in, c)
}
head := pickQuotaHead(in, seedOf, 2, 10)
if len(head) != 10 {
t.Fatalf("head len = %d, want 10", len(head))
}
// Quotas [6,4]: seed 0 fills 6 (tracks 1-6), seed 1 fills 1
// (track 12), spill tops up with tracks 7-9.
got := map[byte]bool{}
for _, c := range head {
got[c.Track.ID.Bytes[15]] = true
}
for _, w := range []byte{1, 2, 3, 4, 5, 6, 7, 8, 9, 12} {
if !got[w] {
t.Errorf("head missing track %d (spill should fill best-first)", w)
}
}
}
func TestPickWeightedTail_PrefersEarlyRanks(t *testing.T) {
// The fresh tail's sample is rank-weighted (#1269): across many
// days, the mean selected rank must sit well below the uniform
// draw's ~99.5 for a 200-deep pool.
pool := make([]recommendation.Candidate, 0, 200)
for i := 0; i < 200; i++ {
c := makeCand(1, i+1, i+1, 1.0)
// Distinct track IDs beyond one byte: bytes wrap at 256.
c.Track.ID.Bytes[14] = byte(i / 200)
c.Track.ID.Bytes[15] = byte(i % 200)
pool = append(pool, c)
}
rankOf := map[pgtype.UUID]int{}
for i, c := range pool {
rankOf[c.Track.ID] = i
}
totalRank, picks := 0, 0
for day := 1; day <= 28; day++ {
got := pickWeightedTail(pool, "2026-06-"+twoDigits(day), 10)
if len(got) != 10 {
t.Fatalf("day %d: len = %d, want 10", day, len(got))
}
for _, c := range got {
totalRank += rankOf[c.Track.ID]
picks++
}
}
mean := float64(totalRank) / float64(picks)
if mean > 80 {
t.Errorf("mean selected rank = %.1f; want well below uniform ~99.5", mean)
}
}
func TestPickWeightedTail_DeterministicWithinDay(t *testing.T) {
pool := make([]recommendation.Candidate, 0, 60)
for i := 0; i < 60; i++ {
pool = append(pool, makeCand(i+1, i+1, i+1, 1.0))
}
a := pickWeightedTail(pool, "2026-06-05", 8)
b := pickWeightedTail(pool, "2026-06-05", 8)
for i := range a {
if a[i].Track.ID != b[i].Track.ID {
t.Fatal("weighted tail must be deterministic within a day")
}
}
}
+1 -1
View File
@@ -236,7 +236,7 @@ func (s *Scheduler) runStartupCatchUp(ctx context.Context, users []dbq.ListActiv
// build can read a fresh profile (phase-2 consumption); both steps are
// best-effort and a failure in one is logged without blocking the other.
func (s *Scheduler) rebuildUserDaily(ctx context.Context, userID pgtype.UUID, now time.Time) {
if err := taste.BuildTasteProfile(ctx, s.pool, s.logger, userID, taste.DefaultConfig()); err != nil {
if err := taste.BuildTasteProfile(ctx, s.pool, s.logger, userID, currentTasteConfig()); err != nil {
s.logger.Warn("scheduler: taste profile rebuild failed",
"user_id", uuidStringPL(userID), "err", err)
}
+49 -24
View File
@@ -46,6 +46,26 @@ func TestPickSeedArtistsFromRows_Empty(t *testing.T) {
}
}
func TestPickKindForSeedTier(t *testing.T) {
// The seed query's fallback tiers map onto the rule-#131 ladder:
// fresh 7-day engagement is the exact desire, everything past the
// 30-day step-back collapses into the far tier (#1255).
cases := []struct {
tier int32
want string
}{
{0, pickKindTier1},
{1, pickKindTier2},
{2, pickKindTier3},
{3, pickKindTier3},
}
for _, c := range cases {
if got := pickKindForSeedTier(c.tier); got != c.want {
t.Errorf("pickKindForSeedTier(%d) = %q, want %q", c.tier, got, c.want)
}
}
}
// userIDHash is the per-user, per-day hash that drives the daily-
// determinism RNGs. Same family as tieBreakHash, just keyed on user
// ID instead of track ID.
@@ -76,60 +96,65 @@ func TestUserIDHash_DifferentUserChangesHash(t *testing.T) {
}
}
// pickForYouSeedForDay rotates the chosen For-You seed across the
// user's top-played candidates using userIDHash. Verifies the picker
// is deterministic within a day, varies across days, and degrades
// gracefully when fewer than 5 candidates exist.
// pickDailySeeds shuffles the candidate pool daily-deterministically
// and takes up to n — For-You uses n=forYouSeedCount for its
// multi-seed blend (#1269), Songs-like uses n=3 via the
// pickSeedArtistsForDay wrapper. Verifies determinism within a day,
// variation across days, and graceful degradation on small pools.
func TestPickForYouSeedForDay_DeterministicWithinDay(t *testing.T) {
func TestPickDailySeeds_DeterministicWithinDay(t *testing.T) {
u := pgtype.UUID{Bytes: [16]byte{1}, Valid: true}
seeds := []pgtype.UUID{
pool := []pgtype.UUID{
{Bytes: [16]byte{10}, Valid: true},
{Bytes: [16]byte{20}, Valid: true},
{Bytes: [16]byte{30}, Valid: true},
{Bytes: [16]byte{40}, Valid: true},
{Bytes: [16]byte{50}, Valid: true},
}
a := pickForYouSeedForDay(seeds, u, "2026-05-04")
b := pickForYouSeedForDay(seeds, u, "2026-05-04")
if a != b {
t.Fatalf("same day should pick same seed; got %v then %v", a, b)
a := pickDailySeeds(pool, u, "2026-05-04", 3)
b := pickDailySeeds(pool, u, "2026-05-04", 3)
if len(a) != 3 || len(b) != 3 {
t.Fatalf("lens = %d, %d; want 3, 3", len(a), len(b))
}
for i := range a {
if a[i] != b[i] {
t.Fatalf("same day should pick same seeds; got %v then %v", a, b)
}
}
}
func TestPickForYouSeedForDay_VariesAcrossDays(t *testing.T) {
func TestPickDailySeeds_VariesAcrossDays(t *testing.T) {
u := pgtype.UUID{Bytes: [16]byte{1}, Valid: true}
seeds := []pgtype.UUID{
pool := []pgtype.UUID{
{Bytes: [16]byte{10}, Valid: true},
{Bytes: [16]byte{20}, Valid: true},
{Bytes: [16]byte{30}, Valid: true},
{Bytes: [16]byte{40}, Valid: true},
{Bytes: [16]byte{50}, Valid: true},
}
picks := map[[16]byte]bool{}
firstPicks := map[[16]byte]bool{}
for i := 1; i <= 30; i++ {
date := "2026-05-" + twoDigits(i)
picks[pickForYouSeedForDay(seeds, u, date).Bytes] = true
firstPicks[pickDailySeeds(pool, u, date, 3)[0].Bytes] = true
}
if len(picks) < 2 {
t.Errorf("expected >=2 distinct seeds across 30 days; got %d", len(picks))
if len(firstPicks) < 2 {
t.Errorf("expected >=2 distinct lead seeds across 30 days; got %d", len(firstPicks))
}
}
func TestPickForYouSeedForDay_SingleCandidate(t *testing.T) {
func TestPickDailySeeds_SmallPool(t *testing.T) {
u := pgtype.UUID{Bytes: [16]byte{1}, Valid: true}
only := pgtype.UUID{Bytes: [16]byte{99}, Valid: true}
got := pickForYouSeedForDay([]pgtype.UUID{only}, u, "2026-05-04")
if got != only {
t.Errorf("single-seed pool should return that seed; got %v", got)
got := pickDailySeeds([]pgtype.UUID{only}, u, "2026-05-04", 3)
if len(got) != 1 || got[0] != only {
t.Errorf("single-entry pool should return just that entry; got %v", got)
}
}
func TestPickForYouSeedForDay_EmptyPool(t *testing.T) {
func TestPickDailySeeds_EmptyPool(t *testing.T) {
u := pgtype.UUID{Bytes: [16]byte{1}, Valid: true}
got := pickForYouSeedForDay(nil, u, "2026-05-04")
if got.Valid {
t.Errorf("empty pool should return zero UUID; got %v", got)
if got := pickDailySeeds(nil, u, "2026-05-04", 3); len(got) != 0 {
t.Errorf("empty pool should return nothing; got %v", got)
}
}
+315 -63
View File
@@ -12,8 +12,10 @@ import (
"errors"
"fmt"
"log/slog"
"math"
"math/rand"
"sort"
"sync"
"time"
"github.com/jackc/pgx/v5"
@@ -22,6 +24,7 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
"git.fabledsword.com/bvandeusen/minstrel/internal/recommendation"
"git.fabledsword.com/bvandeusen/minstrel/internal/taste"
)
// seedArtistRow mirrors the sqlc-generated PickSeedArtistsRow shape.
@@ -83,32 +86,15 @@ func userIDHash(userID pgtype.UUID, dateStr string) uint64 {
return binary.BigEndian.Uint64(sum[:8])
}
// pickForYouSeedForDay picks one of the user's top-played candidate
// tracks as today's For-You seed. With 5 candidates and SHA-256-based
// rotation, each candidate gets picked roughly 1 day in 5; the chosen
// seed drives the similarity candidate pool, so the resulting mix is
// substantially different across days for a user with diverse top-N
// plays.
//
// Degrades gracefully: 1 candidate → returns it; 0 → returns zero UUID
// (caller treats as "no seed available" and skips For-You).
func pickForYouSeedForDay(seeds []pgtype.UUID, userID pgtype.UUID, dateStr string) pgtype.UUID {
if len(seeds) == 0 {
return pgtype.UUID{}
}
idx := int(userIDHash(userID, dateStr) % uint64(len(seeds)))
return seeds[idx]
}
// pickSeedArtistsForDay takes a candidate pool of up to N artists and
// returns up to 3 of them, daily-deterministically shuffled. Each day
// gets a different ordering / selection, but within-day stability is
// preserved (same inputs always produce the same output).
// pickDailySeeds takes a candidate pool of UUIDs and returns up to n
// of them, daily-deterministically shuffled. Each day gets a different
// ordering / selection, but within-day stability is preserved (same
// inputs always produce the same output).
//
// Uses Fisher-Yates over a copy of the input, seeded by userIDHash.
// Degrades gracefully: <3 candidates returns whatever is available in
// Degrades gracefully: <n candidates returns whatever is available in
// shuffled order.
func pickSeedArtistsForDay(pool []pgtype.UUID, userID pgtype.UUID, dateStr string) []pgtype.UUID {
func pickDailySeeds(pool []pgtype.UUID, userID pgtype.UUID, dateStr string, n int) []pgtype.UUID {
if len(pool) == 0 {
return nil
}
@@ -118,29 +104,94 @@ func pickSeedArtistsForDay(pool []pgtype.UUID, userID pgtype.UUID, dateStr strin
rng.Shuffle(len(shuffled), func(i, j int) {
shuffled[i], shuffled[j] = shuffled[j], shuffled[i]
})
n := 3
if len(shuffled) < n {
n = len(shuffled)
}
return shuffled[:n]
}
// pickSeedArtistsForDay picks up to 3 seed artists for the Songs-like
// mixes; For-You uses pickDailySeeds directly with forYouSeedCount.
func pickSeedArtistsForDay(pool []pgtype.UUID, userID pgtype.UUID, dateStr string) []pgtype.UUID {
return pickDailySeeds(pool, userID, dateStr, 3)
}
// rankedCandidate is a (track_id, score) pair used during in-memory
// sorting before insert into playlist_tracks. T5 fills these from
// recommendation.Candidate scores.
// recommendation.Candidate scores. PickKind is the track's provenance
// within its mix — For-You's head/tail split (#1249) and Discover's
// buckets (#1270); empty persists as NULL (variant doesn't stamp yet).
type rankedCandidate struct {
TrackID pgtype.UUID
Score float64
PickKind string
}
// Pick kinds, persisted on playlist_tracks.pick_kind so plays can
// attribute skips to the population that sourced the track — For You's
// "taste engine missing" vs "freshness tax", Discover's three buckets
// (#1270). Values mirror the CHECK in migration 0039; the tier1-3
// values there land with the tiered-mix rebuilds.
const (
pickKindTaste = "taste"
pickKindFresh = "fresh"
pickKindDormant = "dormant"
pickKindCrossUser = "cross_user"
pickKindRandom = "random"
pickKindTier1 = "tier1"
pickKindTier2 = "tier2"
pickKindTier3 = "tier3"
)
// pickKindForSeedTier maps PickSeedArtists' seed-pool fallback tier
// onto the rule-#131 pick-kind ladder: fresh 7-day engagement pins the
// mix's exact desire (tier1), the 30-day window steps back a little
// (tier2), and all-time / liked-only seeds are the far fallback
// (tier3). Stamped mix-wide — seed staleness is a property of the
// whole build, and it's what the metrics breakdown attributes skips to.
func pickKindForSeedTier(tier int32) string {
switch tier {
case 0:
return pickKindTier1
case 1:
return pickKindTier2
default:
return pickKindTier3
}
}
// pickKindForMixTier maps a tiered mix query's 1-based tier column
// (numbered to match rule #131's ladder directly) onto the pick-kind
// vocabulary. Distinct from pickKindForSeedTier, whose 0-based tiers
// count fallback steps of the seed pool rather than eligibility rungs.
func pickKindForMixTier(tier int32) string {
switch tier {
case 1:
return pickKindTier1
case 2:
return pickKindTier2
default:
return pickKindTier3
}
}
const systemMixLength = 25
// systemMixWeights are the fixed scoring weights used by the cron worker.
// systemMixWeights are the scoring weights used by the daily builds.
// JitterMagnitude is small (0.1) and combined with a userIDHash-seeded
// RNG (see scoreAndSortCandidates) — same (user, day) produces same
// scores within a day, but near-tied candidates reshuffle across days
// so the playlist doesn't feel frozen.
var systemMixWeights = recommendation.ScoringWeights{
//
// DB-tunable since #1250: the recsettings service pushes the current
// daily_mix profile via SetSystemMixWeights at boot and on every admin
// change (coverart Configure() pattern — no signature threading, live
// effect without restart). The literal here is only the pre-push
// value; shipped defaults live in recsettings.ShippedDailyMixWeights,
// which must stay in sync with it.
var (
systemTuningMu sync.RWMutex
systemMixWeights = recommendation.ScoringWeights{
BaseWeight: 1.0,
LikeBoost: 2.0,
RecencyWeight: 1.0,
@@ -154,6 +205,37 @@ var systemMixWeights = recommendation.ScoringWeights{
// while passive avoidance (negative) gently demotes.
TasteWeight: 1.5,
}
systemTasteConfig = taste.DefaultConfig()
)
// SetSystemMixWeights installs the current daily_mix scoring weights.
// Called by the recsettings service at boot and on admin updates.
func SetSystemMixWeights(w recommendation.ScoringWeights) {
systemTuningMu.Lock()
defer systemTuningMu.Unlock()
systemMixWeights = w
}
// SetTasteConfig installs the taste-profile build configuration
// (half-life + engagement curve, #1250). Same push model as
// SetSystemMixWeights.
func SetTasteConfig(c taste.Config) {
systemTuningMu.Lock()
defer systemTuningMu.Unlock()
systemTasteConfig = c
}
func currentSystemMixWeights() recommendation.ScoringWeights {
systemTuningMu.RLock()
defer systemTuningMu.RUnlock()
return systemMixWeights
}
func currentTasteConfig() taste.Config {
systemTuningMu.RLock()
defer systemTuningMu.RUnlock()
return systemTasteConfig
}
// forYouHeadN is the number of top-scored tracks that anchor the For-You
// playlist; forYouTailN is the number sampled from positions 2*headN
@@ -175,6 +257,122 @@ const (
forYouTailN = 50
)
// forYouSeedCount is how many of the user's top-5 tracks seed each
// day's build (#1269). A single rotating seed made the mix bipolar
// day-to-day — swinging whole neighborhoods as the rotation moved
// between dissimilar anchors; blending 3 spans neighborhoods within
// one day and smooths the swing across days.
const forYouSeedCount = 3
// headQuotas fixes each seed's share of the For-You head so one
// neighborhood can't monopolize it even when its candidates out-score
// the others (#1269): 50/30/20 for three seeds, 60/40 for two, all of
// it for one. Rounding remainder goes to the primary seed; a thin
// seed's unfilled quota spills best-first in pickQuotaHead.
func headQuotas(numSeeds, headN int) []int {
if numSeeds <= 1 {
return []int{headN}
}
fractions := []float64{0.5, 0.3, 0.2}
if numSeeds == 2 {
fractions = []float64{0.6, 0.4}
}
if numSeeds > len(fractions) {
// Not reachable at forYouSeedCount = 3; even split keeps any
// future seed-count change from silently starving seeds.
fractions = make([]float64, numSeeds)
for i := range fractions {
fractions[i] = 1.0 / float64(numSeeds)
}
}
quotas := make([]int, numSeeds)
assigned := 0
for i := 0; i < numSeeds; i++ {
quotas[i] = int(float64(headN) * fractions[i])
assigned += quotas[i]
}
quotas[0] += headN - assigned
return quotas
}
// pickQuotaHead walks the score-sorted capped pool best-first, taking
// candidates whose originating seed still has head quota; slots a thin
// seed can't fill spill best-first regardless of seed in a second
// pass. A nil seedOf (single-seed path, tests) attributes everything
// to seed 0, which reduces to plain top-headN.
func pickQuotaHead(capped []recommendation.Candidate, seedOf map[pgtype.UUID]int, numSeeds, headN int) []recommendation.Candidate {
quotas := headQuotas(numSeeds, headN)
taken := make([]int, len(quotas))
inHead := make(map[pgtype.UUID]bool, headN)
head := make([]recommendation.Candidate, 0, headN)
for _, c := range capped {
if len(head) >= headN {
break
}
si := seedOf[c.Track.ID]
if si < 0 || si >= len(quotas) {
si = 0
}
if taken[si] >= quotas[si] {
continue
}
taken[si]++
head = append(head, c)
inHead[c.Track.ID] = true
}
for _, c := range capped {
if len(head) >= headN {
break
}
if inHead[c.Track.ID] {
continue
}
head = append(head, c)
inHead[c.Track.ID] = true
}
return head
}
// forYouTailHalfLifeRanks tunes the fresh tail's rank bias (#1269): a
// candidate's sampling weight halves every 50 ranks, so the 101st-best
// candidate is far likelier than the 401st-best instead of equal —
// freshness keeps its "you'll probably enjoy this" half.
const forYouTailHalfLifeRanks = 50.0
// tailSampleUniverse quantizes tieBreakHash into a uniform (0,1] draw
// for the weighted sample; 1e6 buckets is plenty of resolution for
// pools of a few hundred.
const tailSampleUniverse uint64 = 1_000_000
// pickWeightedTail samples up to tailN candidates from the rank-
// ordered tail pool with exponentially rank-decaying weights, using
// deterministic Efraimidis-Spirakis keys (u^(1/w), u derived from
// tieBreakHash): same (track, day) → same key, so the sample is stable
// within a day and rotates across days. Replaces the uniform daily
// draw where rank 380 had the same chance as rank 101.
func pickWeightedTail(tailPool []recommendation.Candidate, dateStr string, tailN int) []recommendation.Candidate {
if len(tailPool) <= tailN {
return tailPool
}
type keyed struct {
c recommendation.Candidate
key float64
}
keys := make([]keyed, len(tailPool))
for i, c := range tailPool {
u := (float64(tieBreakHash(c.Track.ID, dateStr)%tailSampleUniverse) + 1) /
float64(tailSampleUniverse+1)
w := math.Exp2(-float64(i) / forYouTailHalfLifeRanks)
keys[i] = keyed{c: c, key: math.Pow(u, 1/w)}
}
sort.SliceStable(keys, func(i, j int) bool { return keys[i].key > keys[j].key })
out := make([]recommendation.Candidate, tailN)
for i := 0; i < tailN; i++ {
out[i] = keys[i].c
}
return out
}
// scoreAndSortCandidates scores every candidate with recommendation.Score
// and returns a new slice sorted by score DESC (ties broken by
// tieBreakHash). The scoring RNG is seeded by userIDHash so jitter is
@@ -199,9 +397,10 @@ func scoreAndSortCandidates(cands []recommendation.Candidate, userID pgtype.UUID
sort.SliceStable(ordered, func(i, j int) bool {
return uuidLessPL(ordered[i].Track.ID, ordered[j].Track.ID)
})
weights := currentSystemMixWeights()
pairs := make([]scored, len(ordered))
for i, c := range ordered {
pairs[i] = scored{c: c, score: recommendation.Score(c.Inputs, systemMixWeights, now, rng.Float64)}
pairs[i] = scored{c: c, score: recommendation.Score(c.Inputs, weights, now, rng.Float64)}
}
sort.SliceStable(pairs, func(i, j int) bool {
if pairs[i].score != pairs[j].score {
@@ -336,10 +535,11 @@ func systemForYouSourceLimits() recommendation.CandidateSourceLimits {
}
}
// produceForYou: today's seed from the user's top-5 played tracks
// (rotates daily via userIDHash), similarity candidate pool, head+
// tail composition. The base seed query failing is fatal; a
// candidate-load failure is logged and yields no For-You.
// produceForYou: blend candidate pools from up to forYouSeedCount of
// the user's top-5 played tracks (rotating daily via pickDailySeeds),
// then head+tail composition with per-seed head quotas (#1269). The
// base seed query failing is fatal; a per-seed candidate-load failure
// is logged and that seed just contributes nothing.
func produceForYou(
ctx context.Context, q *dbq.Queries, logger *slog.Logger,
userID pgtype.UUID, dateStr string, now time.Time,
@@ -348,24 +548,40 @@ func produceForYou(
if err != nil {
return nil, fmt.Errorf("pick for-you seed candidates: %w", err)
}
forYouSeed := pickForYouSeedForDay(forYouSeeds, userID, dateStr)
if !forYouSeed.Valid {
seeds := pickDailySeeds(forYouSeeds, userID, dateStr, forYouSeedCount)
if len(seeds) == 0 {
return nil, nil
}
zeroVec := recommendation.SessionVector{Seed: true}
// Merge per-seed pools; first-seen wins on dedup, and seedOf
// remembers which seed sourced each track for the head quotas.
var merged []recommendation.Candidate
seedOf := map[pgtype.UUID]int{}
for i, seed := range seeds {
cands, cerr := recommendation.LoadCandidatesFromSimilarity(
ctx, q, userID, forYouSeed,
ctx, q, userID, seed,
1, // recentlyPlayedHours — small to avoid filtering the seed's recent neighbourhood
zeroVec,
[]pgtype.UUID{forYouSeed},
seeds,
systemForYouSourceLimits(),
)
if cerr != nil {
logger.Warn("system playlist: for-you candidates load failed; skipping",
"user_id", uuidStringPL(userID), "err", cerr)
logger.Warn("system playlist: for-you candidates load failed for seed; continuing",
"user_id", uuidStringPL(userID), "seed", uuidStringPL(seed), "err", cerr)
continue
}
for _, c := range cands {
if _, seen := seedOf[c.Track.ID]; seen {
continue
}
seedOf[c.Track.ID] = i
merged = append(merged, c)
}
}
if len(merged) == 0 {
return nil, nil
}
tracks := pickHeadAndTail(cands, userID, dateStr, now, forYouHeadN, forYouTailN)
tracks := pickHeadAndTail(merged, seedOf, len(seeds), userID, dateStr, now, forYouHeadN, forYouTailN)
if len(tracks) == 0 {
return nil, nil
}
@@ -373,8 +589,11 @@ func produceForYou(
}
// produceSeedMixes: up to 3 "Songs like {artist}" mixes. Seed
// artists rotate daily-deterministically. The base seed-artist
// query failing is fatal; per-artist failures are logged + skipped.
// artists rotate daily-deterministically; the seed query falls back
// through widening engagement windows (#1255) and every returned row
// shares the winning tier, stamped onto the built tracks as their
// pick_kind. The base seed-artist query failing is fatal; per-artist
// failures are logged + skipped.
func produceSeedMixes(
ctx context.Context, q *dbq.Queries, logger *slog.Logger,
userID pgtype.UUID, dateStr string, now time.Time,
@@ -383,6 +602,10 @@ func produceSeedMixes(
if err != nil {
return nil, fmt.Errorf("pick seed artists: %w", err)
}
seedTierKind := ""
if len(seedRows) > 0 {
seedTierKind = pickKindForSeedTier(seedRows[0].Tier)
}
seedRowsLocal := make([]seedArtistRow, 0, len(seedRows))
for _, r := range seedRows {
seedRowsLocal = append(seedRowsLocal, seedArtistRow{
@@ -427,6 +650,9 @@ func produceSeedMixes(
if len(tracks) == 0 {
continue
}
for i := range tracks {
tracks[i].PickKind = seedTierKind
}
out = append(out, builtPlaylist{
Name: fmt.Sprintf("Songs like %s", artistRow.Name),
Variant: "songs_like_artist",
@@ -601,20 +827,22 @@ func pickTopN(cands []recommendation.Candidate, userID pgtype.UUID, dateStr stri
capped = capped[:n]
}
rng := rand.New(rand.NewSource(int64(userIDHash(userID, dateStr))))
weights := currentSystemMixWeights()
out := make([]rankedCandidate, len(capped))
for i, c := range capped {
out[i] = rankedCandidate{
TrackID: c.Track.ID,
Score: recommendation.Score(c.Inputs, systemMixWeights, now, rng.Float64),
Score: recommendation.Score(c.Inputs, weights, now, rng.Float64),
}
}
return out
}
// pickHeadAndTail picks headN from the score-sorted head plus tailN from
// positions 2*headN onward (the tail), with the tail sampled
// daily-deterministically via tieBreakHash. Caps applied before the
// head/tail split. Used by For-You only.
// pickHeadAndTail picks headN taste anchors from the score-sorted pool
// (under per-seed quotas when seedOf/numSeeds describe a multi-seed
// blend, #1269) plus tailN freshness picks sampled rank-weighted from
// positions 2*headN onward. Caps applied before the head/tail split.
// Used by For-You only.
//
// The "tail" — candidates ranked beyond 2*headN — is still similarity-
// related (every candidate passed the similarity filter) but isn't among
@@ -624,14 +852,20 @@ func pickTopN(cands []recommendation.Candidate, userID pgtype.UUID, dateStr stri
// Falls back to standard pickTopN behavior when the candidate pool is too
// small to support a meaningful head/tail split (capped pool <=
// headN+tailN, or no candidates at or beyond position 2*headN).
func pickHeadAndTail(cands []recommendation.Candidate, userID pgtype.UUID, dateStr string, now time.Time, headN, tailN int) []rankedCandidate {
func pickHeadAndTail(
cands []recommendation.Candidate, seedOf map[pgtype.UUID]int, numSeeds int,
userID pgtype.UUID, dateStr string, now time.Time, headN, tailN int,
) []rankedCandidate {
sorted := scoreAndSortCandidates(cands, userID, dateStr, now)
capped := capCandidatesByAlbumAndArtist(sorted)
rng := rand.New(rand.NewSource(int64(userIDHash(userID, dateStr))))
weights := currentSystemMixWeights()
total := headN + tailN
if len(capped) <= total {
// Pool too small for a head/tail split — return up to total entries.
// All marked taste: top-N-by-score IS the taste mechanism; no
// exploration sampling happens on this path.
if len(capped) < total {
total = len(capped)
}
@@ -639,43 +873,55 @@ func pickHeadAndTail(cands []recommendation.Candidate, userID pgtype.UUID, dateS
for i := 0; i < total; i++ {
out[i] = rankedCandidate{
TrackID: capped[i].Track.ID,
Score: recommendation.Score(capped[i].Inputs, systemMixWeights, now, rng.Float64),
Score: recommendation.Score(capped[i].Inputs, weights, now, rng.Float64),
PickKind: pickKindTaste,
}
}
return out
}
head := capped[:headN]
head := pickQuotaHead(capped, seedOf, numSeeds, headN)
inHead := make(map[pgtype.UUID]bool, len(head))
for _, c := range head {
inHead[c.Track.ID] = true
}
tailStart := 2 * headN
if tailStart >= len(capped) {
tailStart = headN
}
// Defensive copy so that sorting the tail pool does not mutate capped.
tailPool := append([]recommendation.Candidate{}, capped[tailStart:]...)
// Sort tail pool by tieBreakHash (daily-deterministic), take tailN.
// Sample is stable across requests within a day but varies across days.
sort.SliceStable(tailPool, func(i, j int) bool {
return tieBreakHash(tailPool[i].Track.ID, dateStr) < tieBreakHash(tailPool[j].Track.ID, dateStr)
})
tail := tailPool
if len(tail) > tailN {
tail = tail[:tailN]
// The tail pool keeps its rank order (position drives the sampling
// weight); head members are excluded — a quota walk can reach past
// tailStart when a seed's candidates rank deep.
tailPool := make([]recommendation.Candidate, 0, len(capped)-tailStart)
for _, c := range capped[tailStart:] {
if inHead[c.Track.ID] {
continue
}
tailPool = append(tailPool, c)
}
tail := pickWeightedTail(tailPool, dateStr, tailN)
// Combine: head order preserved (score-sorted), tail in tieBreakHash
// order. "First similar, then surprise" reads naturally in playback.
// Combine: head first (score-sorted under quotas), then the fresh
// sample. "First similar, then surprise" reads naturally in
// playback. Head entries are the taste picks; tail entries are the
// freshness injection (#1249) — the split the metrics page
// attributes skips to.
combined := make([]recommendation.Candidate, 0, len(head)+len(tail))
combined = append(combined, head...)
combined = append(combined, tail...)
out := make([]rankedCandidate, len(combined))
for i, c := range combined {
kind := pickKindTaste
if i >= len(head) {
kind = pickKindFresh
}
out[i] = rankedCandidate{
TrackID: c.Track.ID,
Score: recommendation.Score(c.Inputs, systemMixWeights, now, rng.Float64),
Score: recommendation.Score(c.Inputs, weights, now, rng.Float64),
PickKind: kind,
}
}
return out
@@ -709,9 +955,15 @@ func insertSystemPlaylist(ctx context.Context, qtx *dbq.Queries, userID pgtype.U
}
for _, t := range tracks {
var pickKind *string
if t.PickKind != "" {
k := t.PickKind
pickKind = &k
}
if _, err := qtx.AppendPlaylistTrack(ctx, dbq.AppendPlaylistTrackParams{
PlaylistID: p.ID,
TrackID: t.TrackID,
PickKind: pickKind,
}); err != nil {
// Track may have been deleted between candidate-load and insert;
// skip silently rather than failing the whole build.
+64 -13
View File
@@ -40,6 +40,17 @@ import (
// Discover so shuffle-on-play has a varied pool within a day.
const discoveryMixLen = 100
// Minimum viable mix sizes (issue #1246): below the floor the variant
// is withheld entirely so Home renders its "listen more to unlock"
// placeholder — a playlist with a couple of songs reads as a build
// bug, not a mix. Album-coherent mixes (NewForYou / FirstListens) get
// a lower floor because one legitimate new album (~5+ tracks) is a
// useful mix on its own; the scattered mixes need more to feel real.
const (
discoveryMixMinLen = 15
discoveryMixMinLenAlbum = 5
)
// discoveryMixSpec describes one discovery mix. The unified producer
// reads the spec and runs a single code path for all variants.
type discoveryMixSpec struct {
@@ -55,6 +66,11 @@ type discoveryMixSpec struct {
// or when day-over-day stability is the intended UX (NewForYou).
dailyRotate bool
// minLen is the minimum viable mix size: a finished pool below it
// is withheld (no playlist row) so the client renders the locked
// placeholder instead of a mix that looks built-wrong (#1246).
minLen int
// fetch returns the raw ranked rows. dateStr is supplied for
// queries that accept it (passed as the second positional arg
// historically); queries that don't accept it ignore the param.
@@ -79,25 +95,49 @@ func produceDiscoveryMix(spec discoveryMixSpec) systemPlaylistProducer {
if spec.dailyRotate {
pool = rotateForDay(pool, userID, dateStr)
}
return emit(spec.name, spec.variant, finishMix(pool, spec.diversify)), nil
tracks := finishMix(pool, spec.diversify)
if len(tracks) < spec.minLen {
logger.Info("system playlist: "+spec.variant+" below minimum viable size; withholding",
"user_id", uuidStringPL(userID), "pool", len(tracks), "min", spec.minLen)
return nil, nil
}
return emit(spec.name, spec.variant, tracks), nil
}
}
// rotateForDay rotates pool left by a daily-deterministic offset so
// each day's downstream truncate-to-N surfaces a different slice of
// the pool while contiguous-block ordering inside the slice is
// preserved. Empty / single-element pools pass through unchanged.
// rotateForDay rotates the pool left by a daily-deterministic offset
// so each day's downstream truncate-to-N surfaces a different slice
// while contiguous-block ordering inside the slice is preserved.
//
// Rotation happens WITHIN each contiguous same-pick-kind block
// (#1267): tiered mixes arrive tier-ordered, and a whole-pool
// rotation would hoist tier-3 filler above tier-1's exact fits.
// Untiered pools are a single block, which reduces to the original
// whole-pool rotation (same seed, same first draw). Empty /
// single-element pools pass through unchanged.
func rotateForDay(pool []discoverTrack, userID pgtype.UUID, dateStr string) []discoverTrack {
n := len(pool)
if n <= 1 {
return pool
}
rng := rand.New(rand.NewSource(int64(userIDHash(userID, dateStr))))
offset := rng.Intn(n)
rotated := make([]discoverTrack, 0, n)
rotated = append(rotated, pool[offset:]...)
rotated = append(rotated, pool[:offset]...)
return rotated
out := make([]discoverTrack, 0, n)
for start := 0; start < n; {
end := start + 1
for end < n && pool[end].PickKind == pool[start].PickKind {
end++
}
block := pool[start:end]
if len(block) > 1 {
offset := rng.Intn(len(block))
out = append(out, block[offset:]...)
out = append(out, block[:offset]...)
} else {
out = append(out, block...)
}
start = end
}
return out
}
// discoveryMixSpecs is the concrete spec list used by the registry in
@@ -109,6 +149,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
{
name: "Deep Cuts", variant: "deep_cuts",
diversify: true, dailyRotate: false, // SQL day-keys via md5(id||$2)
minLen: discoveryMixMinLen,
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, ds string) ([]discoverTrack, error) {
rows, err := q.ListDeepCutsTracks(ctx, dbq.ListDeepCutsTracksParams{
UserID: uid, Column2: ds,
@@ -126,6 +167,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
{
name: "Rediscover", variant: "rediscover",
diversify: true, dailyRotate: true, // SQL has no date arg
minLen: discoveryMixMinLen,
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, _ string) ([]discoverTrack, error) {
rows, err := q.ListRediscoverTracks(ctx, uid)
if err != nil {
@@ -141,6 +183,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
{
name: "New for you", variant: "new_for_you",
diversify: true, dailyRotate: true, // operator wants daily rotation on all deterministic mixes
minLen: discoveryMixMinLenAlbum,
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, _ string) ([]discoverTrack, error) {
rows, err := q.ListNewForYouTracks(ctx, uid)
if err != nil {
@@ -148,7 +191,10 @@ var discoveryMixSpecs = []discoveryMixSpec{
}
out := make([]discoverTrack, len(rows))
for i, r := range rows {
out[i] = discoverTrack{ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID}
out[i] = discoverTrack{
ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID,
PickKind: pickKindForMixTier(r.Tier),
}
}
return out, nil
},
@@ -156,6 +202,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
{
name: "On this day", variant: "on_this_day",
diversify: true, dailyRotate: false, // SQL day-keys via md5(id||$2)
minLen: discoveryMixMinLen,
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, ds string) ([]discoverTrack, error) {
rows, err := q.ListOnThisDayTracks(ctx, dbq.ListOnThisDayTracksParams{
UserID: uid, Column2: ds,
@@ -173,6 +220,7 @@ var discoveryMixSpecs = []discoveryMixSpec{
{
name: "First listens", variant: "first_listens",
diversify: true, dailyRotate: true, // SQL has no date arg; daily rotate + diversity top-up
minLen: discoveryMixMinLenAlbum,
fetch: func(ctx context.Context, q *dbq.Queries, uid pgtype.UUID, _ string) ([]discoverTrack, error) {
rows, err := q.ListFirstListensTracks(ctx, uid)
if err != nil {
@@ -180,7 +228,10 @@ var discoveryMixSpecs = []discoveryMixSpec{
}
out := make([]discoverTrack, len(rows))
for i, r := range rows {
out[i] = discoverTrack{ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID}
out[i] = discoverTrack{
ID: r.ID, AlbumID: r.AlbumID, ArtistID: r.ArtistID,
PickKind: pickKindForMixTier(r.Tier),
}
}
return out, nil
},
@@ -217,7 +268,7 @@ func finishMix(rows []discoverTrack, diversify bool) []rankedCandidate {
}
tracks := make([]rankedCandidate, len(pool))
for i, t := range pool {
tracks[i] = rankedCandidate{TrackID: t.ID}
tracks[i] = rankedCandidate{TrackID: t.ID, PickKind: t.PickKind}
}
return tracks
}
+342
View File
@@ -0,0 +1,342 @@
package playlists_test
import (
"context"
"fmt"
"path/filepath"
"testing"
"time"
"github.com/jackc/pgx/v5/pgtype"
"github.com/jackc/pgx/v5/pgxpool"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
)
// seedTrackForArtist creates another track under an existing artist +
// album pair. seedTrack's mbid-less upsert never dedupes artists, and
// Deep Cuts' affinity rule needs several tracks sharing one artist_id.
func seedTrackForArtist(t *testing.T, pool *pgxpool.Pool, title string, albumID, artistID pgtype.UUID) dbq.Track {
t.Helper()
track, err := dbq.New(pool).UpsertTrack(context.Background(), dbq.UpsertTrackParams{
Title: title, AlbumID: albumID, ArtistID: artistID,
DurationMs: 1000,
FilePath: filepath.Join(t.TempDir(), fmt.Sprintf("%s.mp3", title)),
FileSize: 100, FileFormat: "mp3",
})
if err != nil {
t.Fatalf("seed track for artist: %v", err)
}
return track
}
// seedSourcedPlayEvent is seedPlayEvent plus a source tag — needed by
// the Rediscover cooldown, which keys off rediscover-sourced skips.
func seedSourcedPlayEvent(t *testing.T, pool *pgxpool.Pool, userID, trackID pgtype.UUID, startedAt time.Time, wasSkipped bool, source string) {
t.Helper()
_, err := pool.Exec(context.Background(), `
WITH s AS (
INSERT INTO play_sessions (user_id, started_at, last_event_at)
VALUES ($1, $3, $3)
RETURNING id
)
INSERT INTO play_events (user_id, track_id, session_id, started_at, was_skipped, source)
SELECT $1, $2, s.id, $3, $4, $5 FROM s
`, userID, trackID, startedAt, wasSkipped, source)
if err != nil {
t.Fatalf("seed sourced play_event: %v", err)
}
}
// idSet collapses candidate rows into a lookup by track id.
func idSet[T any](rows []T, id func(T) pgtype.UUID) map[pgtype.UUID]bool {
out := make(map[pgtype.UUID]bool, len(rows))
for _, r := range rows {
out[id(r)] = true
}
return out
}
// seedPlayWithDuration inserts a completed play with an explicit
// duration_played_ms — the "attempted >= 30s" queries key off it.
func seedPlayWithDuration(t *testing.T, pool *pgxpool.Pool, userID, trackID pgtype.UUID, startedAt time.Time, durationMs int32) {
t.Helper()
_, err := pool.Exec(context.Background(), `
WITH s AS (
INSERT INTO play_sessions (user_id, started_at, last_event_at)
VALUES ($1, $3, $3)
RETURNING id
)
INSERT INTO play_events (user_id, track_id, session_id, started_at, duration_played_ms)
SELECT $1, $2, s.id, $3, $4 FROM s
`, userID, trackID, startedAt, durationMs)
if err != nil {
t.Fatalf("seed play with duration: %v", err)
}
}
// seedAlbumForArtist creates a fresh album under an existing artist —
// the tiered album queries need several albums sharing one artist_id.
func seedAlbumForArtist(t *testing.T, pool *pgxpool.Pool, title string, artistID pgtype.UUID) pgtype.UUID {
t.Helper()
al, err := dbq.New(pool).UpsertAlbum(context.Background(), dbq.UpsertAlbumParams{
Title: title, SortTitle: title, ArtistID: artistID,
})
if err != nil {
t.Fatalf("seed album for artist: %v", err)
}
return al.ID
}
func backdateAlbum(t *testing.T, pool *pgxpool.Pool, albumID pgtype.UUID, to time.Time) {
t.Helper()
if _, err := pool.Exec(context.Background(),
`UPDATE albums SET created_at = $2 WHERE id = $1`, albumID, to); err != nil {
t.Fatalf("backdate album: %v", err)
}
}
func likeArtist(t *testing.T, pool *pgxpool.Pool, userID, artistID pgtype.UUID) {
t.Helper()
if _, err := pool.Exec(context.Background(),
`INSERT INTO general_likes_artists (user_id, artist_id) VALUES ($1, $2)`,
userID, artistID); err != nil {
t.Fatalf("like artist: %v", err)
}
}
func TestListDeepCutsTracks_SkippedTracksExcluded(t *testing.T) {
// #1257: eligibility used to count only unskipped plays, so a track
// skipped repeatedly (and never finished) read as "barely heard".
pool := newPool(t)
q := dbq.New(pool)
u := seedUser(t, pool, "dc1")
now := time.Now().UTC()
// Affinity artist: 5 non-skip plays on the heavy track.
heavy := seedTrack(t, pool, "dc1-heavy", "dc1-artist")
for p := 0; p < 5; p++ {
seedPlayEvent(t, pool, u.ID, heavy.ID, now.Add(-time.Duration(p+1)*time.Hour), false)
}
// Actively rejected: two skips, zero completed plays.
rejected := seedTrackForArtist(t, pool, "dc1-rejected", heavy.AlbumID, heavy.ArtistID)
seedPlayEvent(t, pool, u.ID, rejected.ID, now.Add(-2*time.Hour), true)
seedPlayEvent(t, pool, u.ID, rejected.ID, now.Add(-3*time.Hour), true)
// Genuinely unexplored.
unheard := seedTrackForArtist(t, pool, "dc1-unheard", heavy.AlbumID, heavy.ArtistID)
// One accidental skip stays below the threshold.
oneSkip := seedTrackForArtist(t, pool, "dc1-oneskip", heavy.AlbumID, heavy.ArtistID)
seedPlayEvent(t, pool, u.ID, oneSkip.ID, now.Add(-4*time.Hour), true)
rows, err := q.ListDeepCutsTracks(context.Background(), dbq.ListDeepCutsTracksParams{
UserID: u.ID, Column2: "2026-07-03",
})
if err != nil {
t.Fatalf("ListDeepCutsTracks: %v", err)
}
got := idSet(rows, func(r dbq.ListDeepCutsTracksRow) pgtype.UUID { return r.ID })
if got[rejected.ID] {
t.Error("twice-skipped track qualified as a deep cut; skips must disqualify")
}
if !got[unheard.ID] {
t.Error("unexplored track missing from deep cuts")
}
if !got[oneSkip.ID] {
t.Error("single accidental skip must not banish a track")
}
if got[heavy.ID] {
t.Error("heavily-played track must not be a deep cut")
}
}
func TestListOnThisDayTracks_WrapsYearBoundary(t *testing.T) {
// #1256: plain ABS on day-of-year made Dec 28 vs Jan 3 read as 359
// days apart. The build date parameter anchors "today" so the wrap
// is testable at any run date.
pool := newPool(t)
q := dbq.New(pool)
u := seedUser(t, pool, "otd1")
// Played Dec 28, 2025 — 6 circular days from the Jan 3 build date.
winter := seedTrack(t, pool, "otd1-winter", "otd1-artist")
seedPlayEvent(t, pool, u.ID, winter.ID,
time.Date(2025, 12, 28, 12, 0, 0, 0, time.UTC), false)
// Played mid-March — far outside the ±10-day window.
spring := seedTrack(t, pool, "otd1-spring", "otd1-artist")
seedPlayEvent(t, pool, u.ID, spring.ID,
time.Date(2025, 3, 15, 12, 0, 0, 0, time.UTC), false)
rows, err := q.ListOnThisDayTracks(context.Background(), dbq.ListOnThisDayTracksParams{
UserID: u.ID, Column2: "2026-01-03",
})
if err != nil {
t.Fatalf("ListOnThisDayTracks: %v", err)
}
got := idSet(rows, func(r dbq.ListOnThisDayTracksRow) pgtype.UUID { return r.ID })
if !got[winter.ID] {
t.Error("play 6 calendar days across the year boundary must qualify (circular DOY)")
}
if got[spring.ID] {
t.Error("play ~70 days away must not qualify")
}
}
func TestListNewForYouTracks_TiersAndConsumption(t *testing.T) {
// #1267: tiered eligibility with consume-on-play. Tier 1 = fresh
// affinity albums; tier 2 = affinity albums from the wider 90-day
// window; tier 3 = fresh additions without affinity; any attempted
// (>=30s) track consumes the album out of the mix entirely.
pool := newPool(t)
q := dbq.New(pool)
u := seedUser(t, pool, "nfy1")
now := time.Now().UTC()
// Affinity artist via a like; anchor track establishes the artist.
anchor := seedTrack(t, pool, "nfy1-anchor", "nfy1-artistX")
likeArtist(t, pool, u.ID, anchor.ArtistID)
backdateAlbum(t, pool, anchor.AlbumID, now.Add(-200*24*time.Hour)) // out of every window
// Fresh affinity album → tier 1.
freshAlbum := seedAlbumForArtist(t, pool, "nfy1-fresh", anchor.ArtistID)
freshTrack := seedTrackForArtist(t, pool, "nfy1-fresh-1", freshAlbum, anchor.ArtistID)
// Affinity album added 60 days ago → tier 2.
olderAlbum := seedAlbumForArtist(t, pool, "nfy1-older", anchor.ArtistID)
olderTrack := seedTrackForArtist(t, pool, "nfy1-older-1", olderAlbum, anchor.ArtistID)
backdateAlbum(t, pool, olderAlbum, now.Add(-60*24*time.Hour))
// Fresh album by an unrelated artist → tier 3.
otherTrack := seedTrack(t, pool, "nfy1-other-1", "nfy1-artistY")
// Fresh affinity album with one attempted track → consumed, absent.
consumedAlbum := seedAlbumForArtist(t, pool, "nfy1-consumed", anchor.ArtistID)
consumedTrack := seedTrackForArtist(t, pool, "nfy1-consumed-1", consumedAlbum, anchor.ArtistID)
seedPlayWithDuration(t, pool, u.ID, consumedTrack.ID, now.Add(-time.Hour), 35_000)
// Affinity album added 120 days ago → outside every window, absent.
ancientAlbum := seedAlbumForArtist(t, pool, "nfy1-ancient", anchor.ArtistID)
ancientTrack := seedTrackForArtist(t, pool, "nfy1-ancient-1", ancientAlbum, anchor.ArtistID)
backdateAlbum(t, pool, ancientAlbum, now.Add(-120*24*time.Hour))
rows, err := q.ListNewForYouTracks(context.Background(), u.ID)
if err != nil {
t.Fatalf("ListNewForYouTracks: %v", err)
}
tiers := map[pgtype.UUID]int32{}
for _, r := range rows {
tiers[r.ID] = r.Tier
}
if got := tiers[freshTrack.ID]; got != 1 {
t.Errorf("fresh affinity album tier = %d, want 1", got)
}
if got := tiers[olderTrack.ID]; got != 2 {
t.Errorf("60-day affinity album tier = %d, want 2", got)
}
if got := tiers[otherTrack.ID]; got != 3 {
t.Errorf("fresh non-affinity album tier = %d, want 3", got)
}
if _, in := tiers[consumedTrack.ID]; in {
t.Error("album with an attempted track must be consumed out of the mix")
}
if _, in := tiers[ancientTrack.ID]; in {
t.Error("album added 120 days ago must be outside every window")
}
}
func TestListFirstListensTracks_AttemptThreshold(t *testing.T) {
// #1268: "attempted" is track-level at >=30s. A 2-second brush no
// longer disqualifies an album; barely-attempted albums step back
// to tier 2 with the attempted tracks themselves excluded.
pool := newPool(t)
q := dbq.New(pool)
u := seedUser(t, pool, "fl1")
now := time.Now().UTC()
// Never touched → tier 1.
untouched := seedTrack(t, pool, "fl1-untouched-1", "fl1-artistA")
// One 2-second accidental brush → still tier 1 (the headline bug).
brushed := seedTrack(t, pool, "fl1-brushed-1", "fl1-artistB")
seedPlayWithDuration(t, pool, u.ID, brushed.ID, now.Add(-time.Hour), 2_000)
// 4-track album with one attempted (25%) → tier 2, attempted track
// excluded, siblings present.
barely1 := seedTrack(t, pool, "fl1-barely-1", "fl1-artistC")
barely2 := seedTrackForArtist(t, pool, "fl1-barely-2", barely1.AlbumID, barely1.ArtistID)
barely3 := seedTrackForArtist(t, pool, "fl1-barely-3", barely1.AlbumID, barely1.ArtistID)
barely4 := seedTrackForArtist(t, pool, "fl1-barely-4", barely1.AlbumID, barely1.ArtistID)
seedPlayWithDuration(t, pool, u.ID, barely1.ID, now.Add(-time.Hour), 40_000)
// 2-track album with one attempted (50%) → past the brush
// threshold, absent entirely.
explored1 := seedTrack(t, pool, "fl1-explored-1", "fl1-artistD")
explored2 := seedTrackForArtist(t, pool, "fl1-explored-2", explored1.AlbumID, explored1.ArtistID)
seedPlayWithDuration(t, pool, u.ID, explored1.ID, now.Add(-time.Hour), 40_000)
rows, err := q.ListFirstListensTracks(context.Background(), u.ID)
if err != nil {
t.Fatalf("ListFirstListensTracks: %v", err)
}
tiers := map[pgtype.UUID]int32{}
for _, r := range rows {
tiers[r.ID] = r.Tier
}
if got := tiers[untouched.ID]; got != 1 {
t.Errorf("untouched album tier = %d, want 1", got)
}
if got := tiers[brushed.ID]; got != 1 {
t.Errorf("2-second brush must not disqualify: tier = %d, want 1", got)
}
if _, in := tiers[barely1.ID]; in {
t.Error("the attempted track itself is not a first listen")
}
for _, tr := range []dbq.Track{barely2, barely3, barely4} {
if got := tiers[tr.ID]; got != 2 {
t.Errorf("sibling of attempted track tier = %d, want 2", got)
}
}
if _, in := tiers[explored1.ID]; in {
t.Error("half-attempted album must be excluded (attempted track)")
}
if _, in := tiers[explored2.ID]; in {
t.Error("half-attempted album must be excluded (sibling)")
}
}
func TestListRediscoverTracks_SkippedResurfacingCoolsDown(t *testing.T) {
// #1258: a skip on a rediscover-sourced play is a declined
// invitation — the track sits out ~90 days instead of re-qualifying
// the next day. Skips from other surfaces don't cool down, and the
// cooldown expires.
pool := newPool(t)
q := dbq.New(pool)
u := seedUser(t, pool, "rd1")
now := time.Now().UTC()
eligible := func(name string, playsAge time.Duration) dbq.Track {
tk := seedTrack(t, pool, name, name+"-artist")
for p := 0; p < 3; p++ {
seedPlayEvent(t, pool, u.ID, tk.ID,
now.Add(-playsAge-time.Duration(p)*time.Hour), false)
}
return tk
}
// Declined 5 days ago from Rediscover itself → cooling down.
declined := eligible("rd1-declined", 60*24*time.Hour)
seedSourcedPlayEvent(t, pool, u.ID, declined.ID, now.Add(-5*24*time.Hour), true, "rediscover")
// Skipped 5 days ago, but from a different surface → still eligible.
otherSkip := eligible("rd1-otherskip", 60*24*time.Hour)
seedSourcedPlayEvent(t, pool, u.ID, otherSkip.ID, now.Add(-5*24*time.Hour), true, "radio")
// Declined 100 days ago → cooldown expired, eligible again.
lapsed := eligible("rd1-lapsed", 120*24*time.Hour)
seedSourcedPlayEvent(t, pool, u.ID, lapsed.ID, now.Add(-100*24*time.Hour), true, "rediscover")
rows, err := q.ListRediscoverTracks(context.Background(), u.ID)
if err != nil {
t.Fatalf("ListRediscoverTracks: %v", err)
}
got := idSet(rows, func(r dbq.ListRediscoverTracksRow) pgtype.UUID { return r.ID })
if got[declined.ID] {
t.Error("track declined from rediscover 5 days ago must be cooling down")
}
if !got[otherSkip.ID] {
t.Error("skip from a non-rediscover surface must not cool the track down")
}
if !got[lapsed.ID] {
t.Error("cooldown must expire after ~90 days")
}
}
+104
View File
@@ -0,0 +1,104 @@
package playlists
import (
"testing"
"github.com/jackc/pgx/v5/pgtype"
)
func mixTrack(b byte, kind string) discoverTrack {
return discoverTrack{ID: pgtype.UUID{Bytes: [16]byte{b}, Valid: true}, PickKind: kind}
}
func idsOf(pool []discoverTrack) []byte {
out := make([]byte, len(pool))
for i, t := range pool {
out[i] = t.ID.Bytes[0]
}
return out
}
func TestRotateForDay_RotatesWithinTierBlocks(t *testing.T) {
// Tiered pools arrive tier-ordered; rotation must vary the daily
// slice WITHIN each tier without hoisting tier-3 filler above
// tier-1's exact fits (#1267).
pool := []discoverTrack{
mixTrack(1, pickKindTier1), mixTrack(2, pickKindTier1), mixTrack(3, pickKindTier1),
mixTrack(4, pickKindTier2), mixTrack(5, pickKindTier2),
mixTrack(6, pickKindTier3),
}
u := pgtype.UUID{Bytes: [16]byte{42}, Valid: true}
got := rotateForDay(pool, u, "2026-07-03")
if len(got) != len(pool) {
t.Fatalf("len = %d, want %d", len(got), len(pool))
}
wantKinds := []string{
pickKindTier1, pickKindTier1, pickKindTier1,
pickKindTier2, pickKindTier2, pickKindTier3,
}
for i, k := range wantKinds {
if got[i].PickKind != k {
t.Fatalf("pos %d kind = %q, want %q (tier order broken: %v)",
i, got[i].PickKind, k, idsOf(got))
}
}
// Each block is a rotation of its input: contiguity check for the
// 3-element tier-1 block (successor relation preserved cyclically).
wantOrder := map[byte]byte{1: 2, 2: 3, 3: 1}
for i := 0; i < 2; i++ {
if got[i+1].ID.Bytes[0] != wantOrder[got[i].ID.Bytes[0]] {
t.Errorf("tier1 block is not a rotation: %v", idsOf(got[:3]))
break
}
}
// Determinism within a day.
again := rotateForDay(pool, u, "2026-07-03")
for i := range got {
if got[i].ID != again[i].ID {
t.Fatal("rotation must be deterministic for the same (user, day)")
}
}
}
func TestRotateForDay_UntieredPoolIsWholeRotation(t *testing.T) {
// Untiered pools (all PickKind "") are one block — the original
// whole-pool rotation semantics.
pool := []discoverTrack{mixTrack(1, ""), mixTrack(2, ""), mixTrack(3, ""), mixTrack(4, "")}
got := rotateForDay(pool, pgtype.UUID{Bytes: [16]byte{7}, Valid: true}, "2026-07-03")
if len(got) != 4 {
t.Fatalf("len = %d, want 4", len(got))
}
wantNext := map[byte]byte{1: 2, 2: 3, 3: 4, 4: 1}
for i := 0; i < 3; i++ {
cur, next := got[i].ID.Bytes[0], got[i+1].ID.Bytes[0]
if wantNext[cur] != next {
t.Errorf("not a rotation of the input: %v", idsOf(got))
break
}
}
}
func TestFinishMix_PropagatesPickKind(t *testing.T) {
// Tier stamps ride discoverTrack through diversify/truncate into
// the rankedCandidates that insertSystemPlaylist persists (#1267).
pool := []discoverTrack{mixTrack(1, pickKindTier1), mixTrack(2, pickKindTier2)}
got := finishMix(pool, false)
if len(got) != 2 {
t.Fatalf("len = %d, want 2", len(got))
}
if got[0].PickKind != pickKindTier1 || got[1].PickKind != pickKindTier2 {
t.Errorf("pick kinds = %q, %q — want tier1, tier2", got[0].PickKind, got[1].PickKind)
}
}
func TestPickKindForMixTier(t *testing.T) {
cases := []struct {
tier int32
want string
}{{1, pickKindTier1}, {2, pickKindTier2}, {3, pickKindTier3}, {9, pickKindTier3}}
for _, c := range cases {
if got := pickKindForMixTier(c.tier); got != c.want {
t.Errorf("pickKindForMixTier(%d) = %q, want %q", c.tier, got, c.want)
}
}
}
+53
View File
@@ -135,6 +135,59 @@ func TestBuildSystemPlaylists_SufficientActivity(t *testing.T) {
}
}
func TestBuildSystemPlaylists_StaleActivity_SongsLikeSurvives(t *testing.T) {
// #1255: the seed-artist query's old hard 7-day window emptied the
// pool after a quiet week, and the atomic-replace build then deleted
// every "Songs like X" mix. With the tiered fallback, plays that are
// ~20 days old (outside 7d, inside 30d) must still seed the mixes —
// and the built tracks carry the tier2 stamp so metrics can compare
// stale-seeded mixes against fresh ones.
pool := newPool(t)
logger := discardLogger()
u := seedUser(t, pool, "stale1")
old := time.Now().UTC().Add(-20 * 24 * time.Hour)
for a := 0; a < 4; a++ {
for k := 0; k < 3; k++ {
tk := seedTrack(t, pool,
"stale1-track-"+string(rune('A'+a))+string(rune('0'+k)),
"stale1-artist-"+string(rune('A'+a)))
for p := 0; p < 3; p++ {
seedPlayEvent(t, pool, u.ID, tk.ID,
old.Add(-time.Duration(a*10+k+p)*time.Hour), false)
}
}
}
if err := playlists.BuildSystemPlaylists(context.Background(), pool, logger, u.ID, time.Now().UTC(), t.TempDir()); err != nil {
t.Fatalf("build: %v", err)
}
rows, err := pool.Query(context.Background(), `
SELECT DISTINCT COALESCE(pt.pick_kind, '<null>')
FROM playlist_tracks pt
JOIN playlists p ON p.id = pt.playlist_id
WHERE p.user_id = $1 AND p.system_variant = 'songs_like_artist'
`, u.ID)
if err != nil {
t.Fatalf("query pick_kinds: %v", err)
}
defer rows.Close()
var kinds []string
for rows.Next() {
var k string
if err := rows.Scan(&k); err != nil {
t.Fatalf("scan: %v", err)
}
kinds = append(kinds, k)
}
if len(kinds) == 0 {
t.Fatal("expected songs_like_artist tracks from the 30-day fallback tier; got none (the vanish bug)")
}
if len(kinds) != 1 || kinds[0] != "tier2" {
t.Errorf("pick_kinds = %v, want exactly [tier2] (30-day fallback seeds)", kinds)
}
}
func TestBuildSystemPlaylists_QuarantineExcluded(t *testing.T) {
pool := newPool(t)
logger := discardLogger()
+8 -3
View File
@@ -83,10 +83,14 @@ func buildYouMightLike(
"user_id", uuidStringPL(userID), "err", err)
return youMightLikeResult{built: false}
}
seed := pickForYouSeedForDay(seeds, userID, dateStr)
if !seed.Valid {
// One rotating seed is right here (unlike For-You's multi-seed
// blend, #1269): the row is a short shelf, not a mix, and a single
// neighborhood per day keeps it coherent.
daily := pickDailySeeds(seeds, userID, dateStr, 1)
if len(daily) == 0 {
return youMightLikeResult{built: true}
}
seed := daily[0]
zeroVec := recommendation.SessionVector{Seed: true}
// You-might-like surfaces in-library artists the user does NOT actively
@@ -123,11 +127,12 @@ func rollUpCandidates(
cands []recommendation.Candidate, userID pgtype.UUID, dateStr string, now time.Time,
) (albumIDs, artistIDs []pgtype.UUID) {
rng := rand.New(rand.NewSource(int64(userIDHash(userID, dateStr))))
weights := currentSystemMixWeights()
albumScores := map[pgtype.UUID][]float64{}
artistScores := map[pgtype.UUID][]float64{}
albumArtist := map[pgtype.UUID]pgtype.UUID{}
for _, c := range cands {
s := recommendation.Score(c.Inputs, systemMixWeights, now, rng.Float64)
s := recommendation.Score(c.Inputs, weights, now, rng.Float64)
if c.Track.AlbumID.Valid {
albumScores[c.Track.AlbumID] = append(albumScores[c.Track.AlbumID], s)
albumArtist[c.Track.AlbumID] = c.Track.ArtistID
+178
View File
@@ -0,0 +1,178 @@
// patch.go — field-name mapping + validation for the tuning patches.
// Wire field names are the snake_case column names; the admin API and
// web card use them verbatim.
package recsettings
import (
"errors"
"fmt"
"git.fabledsword.com/bvandeusen/minstrel/internal/recommendation"
)
var (
ErrUnknownScope = errors.New("unknown tuning scope")
ErrUnknownField = errors.New("unknown tuning field")
ErrOutOfRange = errors.New("tuning value out of range")
)
// weightBound caps every scoring weight's magnitude. The scoring
// terms are all in [-1, 1] before weighting, so ±10 is far past any
// useful setting — the bound exists to catch typos (e.g. 100 for
// 1.00), not to constrain exploration.
const weightBound = 10.0
// weightField describes one patchable ScoringWeights field.
type weightField struct {
get func(recommendation.ScoringWeights) float64
set func(*recommendation.ScoringWeights, float64)
// nonNegative marks fields where a negative value is meaningless
// (a negative jitter magnitude or skip penalty inverts intent in a
// way the score formula already expresses through its sign).
nonNegative bool
}
var weightFields = map[string]weightField{
"base_weight": {
get: func(w recommendation.ScoringWeights) float64 { return w.BaseWeight },
set: func(w *recommendation.ScoringWeights, v float64) { w.BaseWeight = v },
},
"like_boost": {
get: func(w recommendation.ScoringWeights) float64 { return w.LikeBoost },
set: func(w *recommendation.ScoringWeights, v float64) { w.LikeBoost = v },
},
"recency_weight": {
get: func(w recommendation.ScoringWeights) float64 { return w.RecencyWeight },
set: func(w *recommendation.ScoringWeights, v float64) { w.RecencyWeight = v },
},
"skip_penalty": {
get: func(w recommendation.ScoringWeights) float64 { return w.SkipPenalty },
set: func(w *recommendation.ScoringWeights, v float64) { w.SkipPenalty = v },
nonNegative: true,
},
"jitter_magnitude": {
get: func(w recommendation.ScoringWeights) float64 { return w.JitterMagnitude },
set: func(w *recommendation.ScoringWeights, v float64) { w.JitterMagnitude = v },
nonNegative: true,
},
"context_weight": {
get: func(w recommendation.ScoringWeights) float64 { return w.ContextWeight },
set: func(w *recommendation.ScoringWeights, v float64) { w.ContextWeight = v },
},
"similarity_weight": {
get: func(w recommendation.ScoringWeights) float64 { return w.SimilarityWeight },
set: func(w *recommendation.ScoringWeights, v float64) { w.SimilarityWeight = v },
},
"taste_weight": {
get: func(w recommendation.ScoringWeights) float64 { return w.TasteWeight },
set: func(w *recommendation.ScoringWeights, v float64) { w.TasteWeight = v },
},
}
// applyWeightPatch validates and applies a partial update, returning
// the new weights and the list of actual changes (values equal to the
// current setting are dropped, so a re-submitted form is a no-op).
func applyWeightPatch(
current recommendation.ScoringWeights, patch map[string]float64,
) (recommendation.ScoringWeights, []fieldChange, error) {
next := current
var changes []fieldChange
for field, v := range patch {
f, ok := weightFields[field]
if !ok {
return current, nil, fmt.Errorf("%w: %q", ErrUnknownField, field)
}
if v < -weightBound || v > weightBound {
return current, nil, fmt.Errorf("%w: %s = %v (|v| must be <= %v)",
ErrOutOfRange, field, v, weightBound)
}
if f.nonNegative && v < 0 {
return current, nil, fmt.Errorf("%w: %s = %v (must be >= 0)",
ErrOutOfRange, field, v)
}
old := f.get(next)
if old == v {
continue
}
f.set(&next, v)
changes = append(changes, fieldChange{Field: field, Old: old, New: v})
}
return next, changes, nil
}
// Taste tuning bounds. The half-life window is generous — from "taste
// is last week" to "taste is a decade" — and the curve points must
// stay ordered inside [0, 1] or the engagement ramps degenerate.
const (
tasteHalfLifeMin = 1.0
tasteHalfLifeMax = 3650.0
)
// applyTastePatch validates and applies a partial taste update. The
// curve-ordering invariant (hard_skip < neutral < full) is checked on
// the PATCHED result, so a patch may move several points at once.
func applyTastePatch(current TasteTuning, patch map[string]float64) (TasteTuning, []fieldChange, error) {
next := current
var changes []fieldChange
for field, v := range patch {
var target *float64
switch field {
case "half_life_days":
if v < tasteHalfLifeMin || v > tasteHalfLifeMax {
return current, nil, fmt.Errorf("%w: %s = %v (must be in [%v, %v])",
ErrOutOfRange, field, v, tasteHalfLifeMin, tasteHalfLifeMax)
}
target = &next.HalfLifeDays
case "engagement_hard_skip":
target = &next.EngagementHardSkip
case "engagement_neutral":
target = &next.EngagementNeutral
case "engagement_full":
target = &next.EngagementFull
default:
return current, nil, fmt.Errorf("%w: %q", ErrUnknownField, field)
}
if field != "half_life_days" && (v < 0 || v > 1) {
return current, nil, fmt.Errorf("%w: %s = %v (must be in [0, 1])",
ErrOutOfRange, field, v)
}
if *target == v {
continue
}
changes = append(changes, fieldChange{Field: field, Old: *target, New: v})
*target = v
}
if !(next.EngagementHardSkip < next.EngagementNeutral &&
next.EngagementNeutral < next.EngagementFull) {
return current, nil, fmt.Errorf(
"%w: engagement curve must satisfy hard_skip < neutral < full (got %v < %v < %v)",
ErrOutOfRange, next.EngagementHardSkip, next.EngagementNeutral, next.EngagementFull)
}
return next, changes, nil
}
// diffWeights returns per-field changes from a to b (empty when equal).
func diffWeights(a, b recommendation.ScoringWeights) []fieldChange {
var out []fieldChange
for field, f := range weightFields {
if f.get(a) != f.get(b) {
out = append(out, fieldChange{Field: field, Old: f.get(a), New: f.get(b)})
}
}
return out
}
// diffTaste returns per-field changes from a to b (empty when equal).
func diffTaste(a, b TasteTuning) []fieldChange {
var out []fieldChange
add := func(field string, oldV, newV float64) {
if oldV != newV {
out = append(out, fieldChange{Field: field, Old: oldV, New: newV})
}
}
add("half_life_days", a.HalfLifeDays, b.HalfLifeDays)
add("engagement_hard_skip", a.EngagementHardSkip, b.EngagementHardSkip)
add("engagement_neutral", a.EngagementNeutral, b.EngagementNeutral)
add("engagement_full", a.EngagementFull, b.EngagementFull)
return out
}
+374
View File
@@ -0,0 +1,374 @@
// Package recsettings is the DB-backed home of the recommendation
// tuning knobs (#1250): the two scoring-weight profiles (radio /
// daily_mix) and the taste-profile build settings (engagement
// half-life + completion curve). It follows the coverart
// SettingsService pattern — boot reconcile seeds shipped defaults for
// missing rows, values are cached under a RWMutex, and every change
// takes effect live (the daily_mix profile + taste config are pushed
// into package playlists; radio reads the cache per request).
//
// Framing (decision #1247): this is the defaults-discovery lab. The
// operator turns knobs to FIND good values; found-good values get
// baked back into the Shipped* functions below as new shipped
// defaults. End users and other operators should never need the card.
package recsettings
import (
"context"
"encoding/json"
"fmt"
"log/slog"
"sort"
"sync"
"github.com/jackc/pgx/v5/pgxpool"
"git.fabledsword.com/bvandeusen/minstrel/internal/db/dbq"
"git.fabledsword.com/bvandeusen/minstrel/internal/playlists"
"git.fabledsword.com/bvandeusen/minstrel/internal/recommendation"
"git.fabledsword.com/bvandeusen/minstrel/internal/taste"
)
// Scope names — the three tunable groups. radio + daily_mix are weight
// profiles; taste is the profile-build settings singleton.
const (
ScopeRadio = "radio"
ScopeDailyMix = "daily_mix"
ScopeTaste = "taste"
)
// TasteTuning is the tunable subset of taste.Config: the engagement
// half-life and the completion→engagement curve points.
type TasteTuning struct {
HalfLifeDays float64
EngagementHardSkip float64
EngagementNeutral float64
EngagementFull float64
}
// ShippedRadioWeights are the shipped radio-profile defaults (moved
// here from config.RecommendationConfig — YAML is bootstrap-only,
// rule: config in UI). Radio is seed-directed (the user picked a
// direction), so taste is a lighter nudge than in the daily mixes.
func ShippedRadioWeights() recommendation.ScoringWeights {
return recommendation.ScoringWeights{
BaseWeight: 1.0,
LikeBoost: 2.0,
RecencyWeight: 1.0,
SkipPenalty: 1.0,
JitterMagnitude: 0.1,
ContextWeight: 2.0,
SimilarityWeight: 2.0,
TasteWeight: 1.0,
}
}
// ShippedDailyMixWeights are the shipped daily_mix-profile defaults.
// Must stay in sync with the pre-push literal in playlists/system.go.
func ShippedDailyMixWeights() recommendation.ScoringWeights {
return recommendation.ScoringWeights{
BaseWeight: 1.0,
LikeBoost: 2.0,
RecencyWeight: 1.0,
SkipPenalty: 2.0,
JitterMagnitude: 0.1,
ContextWeight: 0.5,
SimilarityWeight: 1.5,
TasteWeight: 1.5,
}
}
// ShippedTasteTuning mirrors taste.DefaultConfig's tunable subset.
func ShippedTasteTuning() TasteTuning {
d := taste.DefaultConfig()
return TasteTuning{
HalfLifeDays: d.HalfLifeDays,
EngagementHardSkip: d.Engagement.HardSkip,
EngagementNeutral: d.Engagement.NeutralCompletion,
EngagementFull: d.Engagement.FullCompletion,
}
}
// Service caches the tuning values and owns their DB persistence +
// audit trail. Construct with New at boot.
type Service struct {
pool *pgxpool.Pool
logger *slog.Logger
mu sync.RWMutex
profiles map[string]recommendation.ScoringWeights
taste TasteTuning
}
// New boots the service: seeds shipped defaults for missing rows,
// loads the current values, and pushes the daily_mix weights + taste
// config into package playlists so the daily builds pick them up.
func New(ctx context.Context, pool *pgxpool.Pool, logger *slog.Logger) (*Service, error) {
s := &Service{
pool: pool,
logger: logger,
profiles: map[string]recommendation.ScoringWeights{},
}
if err := s.reconcile(ctx); err != nil {
return nil, fmt.Errorf("recsettings boot: %w", err)
}
return s, nil
}
// reconcile seeds missing rows with shipped defaults, reads everything
// back into the cache, and pushes the playlist-side values.
func (s *Service) reconcile(ctx context.Context) error {
q := dbq.New(s.pool)
for profile, w := range map[string]recommendation.ScoringWeights{
ScopeRadio: ShippedRadioWeights(),
ScopeDailyMix: ShippedDailyMixWeights(),
} {
if err := q.UpsertWeightProfileDefaults(ctx, upsertParams(profile, w)); err != nil {
return fmt.Errorf("seed profile %q: %w", profile, err)
}
}
st := ShippedTasteTuning()
if err := q.UpsertTasteTuningDefaults(ctx, dbq.UpsertTasteTuningDefaultsParams{
HalfLifeDays: st.HalfLifeDays,
EngagementHardSkip: st.EngagementHardSkip,
EngagementNeutral: st.EngagementNeutral,
EngagementFull: st.EngagementFull,
}); err != nil {
return fmt.Errorf("seed taste tuning: %w", err)
}
rows, err := q.ListWeightProfiles(ctx)
if err != nil {
return fmt.Errorf("list weight profiles: %w", err)
}
tt, err := q.GetTasteTuning(ctx)
if err != nil {
return fmt.Errorf("get taste tuning: %w", err)
}
s.mu.Lock()
s.profiles = map[string]recommendation.ScoringWeights{}
for _, r := range rows {
s.profiles[r.Profile] = weightsFromRow(r)
}
s.taste = TasteTuning{
HalfLifeDays: tt.HalfLifeDays,
EngagementHardSkip: tt.EngagementHardSkip,
EngagementNeutral: tt.EngagementNeutral,
EngagementFull: tt.EngagementFull,
}
s.mu.Unlock()
s.push()
return nil
}
// push installs the daily-build values into package playlists (the
// coverart Configure() pattern). Radio needs no push — its handler
// reads Weights(ScopeRadio) per request.
func (s *Service) push() {
playlists.SetSystemMixWeights(s.Weights(ScopeDailyMix))
playlists.SetTasteConfig(s.TasteConfig())
}
// Weights returns the cached weights for a profile scope. Unknown
// scopes return the shipped radio defaults (defensive; callers pass
// the Scope* constants).
func (s *Service) Weights(profile string) recommendation.ScoringWeights {
s.mu.RLock()
defer s.mu.RUnlock()
if w, ok := s.profiles[profile]; ok {
return w
}
return ShippedRadioWeights()
}
// Taste returns the cached taste-tuning values.
func (s *Service) Taste() TasteTuning {
s.mu.RLock()
defer s.mu.RUnlock()
return s.taste
}
// TasteConfig assembles the full taste.Config the profile builder
// consumes: shipped non-tunable knobs (like bonuses, floors, caps)
// plus the tuned half-life and curve. WindowDays scales with the
// half-life at the shipped ratio (270/75 = 3.6 half-lives) — the
// window is a query-cost bound, not an independent knob.
func (s *Service) TasteConfig() taste.Config {
t := s.Taste()
cfg := taste.DefaultConfig()
cfg.HalfLifeDays = t.HalfLifeDays
cfg.WindowDays = t.HalfLifeDays * 3.6
cfg.Engagement = taste.EngagementParams{
HardSkip: t.EngagementHardSkip,
NeutralCompletion: t.EngagementNeutral,
FullCompletion: t.EngagementFull,
}
return cfg
}
// fieldChange is one entry of an audit row's changes array.
type fieldChange struct {
Field string `json:"field"`
Old float64 `json:"old"`
New float64 `json:"new"`
}
// UpdateProfile applies a partial update to one weight profile.
// Unknown fields and out-of-range values reject the whole patch. A
// no-op patch (all values equal to current) writes no audit row.
func (s *Service) UpdateProfile(ctx context.Context, profile string, patch map[string]float64) error {
if profile != ScopeRadio && profile != ScopeDailyMix {
return fmt.Errorf("%w: %q", ErrUnknownScope, profile)
}
current := s.Weights(profile)
next, changes, err := applyWeightPatch(current, patch)
if err != nil {
return err
}
if len(changes) == 0 {
return nil
}
return s.persistProfile(ctx, profile, next, "update", changes)
}
// UpdateTaste applies a partial update to the taste tuning singleton.
func (s *Service) UpdateTaste(ctx context.Context, patch map[string]float64) error {
current := s.Taste()
next, changes, err := applyTastePatch(current, patch)
if err != nil {
return err
}
if len(changes) == 0 {
return nil
}
return s.persistTaste(ctx, next, "update", changes)
}
// Reset restores a scope to its shipped defaults, with one audit row
// carrying the full diff. A scope already at defaults is a no-op.
func (s *Service) Reset(ctx context.Context, scope string) error {
switch scope {
case ScopeRadio, ScopeDailyMix:
shipped := ShippedRadioWeights()
if scope == ScopeDailyMix {
shipped = ShippedDailyMixWeights()
}
changes := diffWeights(s.Weights(scope), shipped)
if len(changes) == 0 {
return nil
}
return s.persistProfile(ctx, scope, shipped, "reset", changes)
case ScopeTaste:
shipped := ShippedTasteTuning()
changes := diffTaste(s.Taste(), shipped)
if len(changes) == 0 {
return nil
}
return s.persistTaste(ctx, shipped, "reset", changes)
default:
return fmt.Errorf("%w: %q", ErrUnknownScope, scope)
}
}
// persistProfile writes the profile row + audit entry, refreshes the
// cache, and pushes daily-build values.
func (s *Service) persistProfile(
ctx context.Context, profile string, w recommendation.ScoringWeights,
action string, changes []fieldChange,
) error {
q := dbq.New(s.pool)
if _, err := q.UpdateWeightProfile(ctx, updateParams(profile, w)); err != nil {
return fmt.Errorf("update profile %q: %w", profile, err)
}
if err := s.audit(ctx, q, profile, action, changes); err != nil {
return err
}
s.mu.Lock()
s.profiles[profile] = w
s.mu.Unlock()
s.push()
return nil
}
func (s *Service) persistTaste(
ctx context.Context, t TasteTuning, action string, changes []fieldChange,
) error {
q := dbq.New(s.pool)
if _, err := q.UpdateTasteTuning(ctx, dbq.UpdateTasteTuningParams{
HalfLifeDays: t.HalfLifeDays,
EngagementHardSkip: t.EngagementHardSkip,
EngagementNeutral: t.EngagementNeutral,
EngagementFull: t.EngagementFull,
}); err != nil {
return fmt.Errorf("update taste tuning: %w", err)
}
if err := s.audit(ctx, q, ScopeTaste, action, changes); err != nil {
return err
}
s.mu.Lock()
s.taste = t
s.mu.Unlock()
s.push()
return nil
}
// audit writes one recommendation_tuning_audit row. Changes are
// sorted by field so rows are deterministic and diff-friendly.
func (s *Service) audit(
ctx context.Context, q *dbq.Queries, scope, action string, changes []fieldChange,
) error {
sort.Slice(changes, func(i, j int) bool { return changes[i].Field < changes[j].Field })
payload, err := json.Marshal(changes)
if err != nil {
return fmt.Errorf("marshal audit changes: %w", err)
}
if err := q.InsertTuningAudit(ctx, dbq.InsertTuningAuditParams{
Scope: scope, Action: action, Changes: payload,
}); err != nil {
return fmt.Errorf("insert audit row: %w", err)
}
return nil
}
func upsertParams(profile string, w recommendation.ScoringWeights) dbq.UpsertWeightProfileDefaultsParams {
return dbq.UpsertWeightProfileDefaultsParams{
Profile: profile,
BaseWeight: w.BaseWeight,
LikeBoost: w.LikeBoost,
RecencyWeight: w.RecencyWeight,
SkipPenalty: w.SkipPenalty,
JitterMagnitude: w.JitterMagnitude,
ContextWeight: w.ContextWeight,
SimilarityWeight: w.SimilarityWeight,
TasteWeight: w.TasteWeight,
}
}
func updateParams(profile string, w recommendation.ScoringWeights) dbq.UpdateWeightProfileParams {
return dbq.UpdateWeightProfileParams{
Profile: profile,
BaseWeight: w.BaseWeight,
LikeBoost: w.LikeBoost,
RecencyWeight: w.RecencyWeight,
SkipPenalty: w.SkipPenalty,
JitterMagnitude: w.JitterMagnitude,
ContextWeight: w.ContextWeight,
SimilarityWeight: w.SimilarityWeight,
TasteWeight: w.TasteWeight,
}
}
func weightsFromRow(r dbq.RecommendationWeightProfile) recommendation.ScoringWeights {
return recommendation.ScoringWeights{
BaseWeight: r.BaseWeight,
LikeBoost: r.LikeBoost,
RecencyWeight: r.RecencyWeight,
SkipPenalty: r.SkipPenalty,
JitterMagnitude: r.JitterMagnitude,
ContextWeight: r.ContextWeight,
SimilarityWeight: r.SimilarityWeight,
TasteWeight: r.TasteWeight,
}
}
+222
View File
@@ -0,0 +1,222 @@
package recsettings
import (
"context"
"encoding/json"
"errors"
"io"
"log/slog"
"os"
"testing"
"github.com/jackc/pgx/v5/pgxpool"
"git.fabledsword.com/bvandeusen/minstrel/internal/db"
"git.fabledsword.com/bvandeusen/minstrel/internal/dbtest"
)
func newPool(t *testing.T) *pgxpool.Pool {
t.Helper()
if testing.Short() {
t.Skip("skipping integration test in -short mode")
}
dsn := os.Getenv("MINSTREL_TEST_DATABASE_URL")
if dsn == "" {
t.Skip("MINSTREL_TEST_DATABASE_URL not set")
}
logger := slog.New(slog.NewTextHandler(io.Discard, nil))
if err := db.Migrate(dsn, logger); err != nil {
t.Fatalf("migrate: %v", err)
}
pool, err := pgxpool.New(context.Background(), dsn)
if err != nil {
t.Fatalf("pool: %v", err)
}
t.Cleanup(pool.Close)
dbtest.ResetDB(t, pool)
return pool
}
func newService(t *testing.T, pool *pgxpool.Pool) *Service {
t.Helper()
s, err := New(context.Background(), pool, slog.New(slog.NewTextHandler(io.Discard, nil)))
if err != nil {
t.Fatalf("recsettings.New: %v", err)
}
return s
}
func auditRows(t *testing.T, pool *pgxpool.Pool) []struct {
Scope, Action string
Changes []fieldChange
} {
t.Helper()
rows, err := pool.Query(context.Background(),
`SELECT scope, action, changes FROM recommendation_tuning_audit ORDER BY id`)
if err != nil {
t.Fatalf("query audit: %v", err)
}
defer rows.Close()
var out []struct {
Scope, Action string
Changes []fieldChange
}
for rows.Next() {
var scope, action string
var raw []byte
if err := rows.Scan(&scope, &action, &raw); err != nil {
t.Fatalf("scan audit: %v", err)
}
var changes []fieldChange
if err := json.Unmarshal(raw, &changes); err != nil {
t.Fatalf("unmarshal audit changes: %v", err)
}
out = append(out, struct {
Scope, Action string
Changes []fieldChange
}{scope, action, changes})
}
return out
}
func TestNew_SeedsShippedDefaults(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
if got := s.Weights(ScopeRadio); got != ShippedRadioWeights() {
t.Errorf("radio weights = %+v, want shipped defaults", got)
}
if got := s.Weights(ScopeDailyMix); got != ShippedDailyMixWeights() {
t.Errorf("daily_mix weights = %+v, want shipped defaults", got)
}
if got := s.Taste(); got != ShippedTasteTuning() {
t.Errorf("taste tuning = %+v, want shipped defaults", got)
}
// Seeding must not write audit rows — nothing changed.
if rows := auditRows(t, pool); len(rows) != 0 {
t.Errorf("boot reconcile wrote %d audit rows, want 0", len(rows))
}
}
func TestUpdateProfile_PersistsAndAudits(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
if err := s.UpdateProfile(context.Background(), ScopeRadio,
map[string]float64{"taste_weight": 2.5, "skip_penalty": 3.0}); err != nil {
t.Fatalf("UpdateProfile: %v", err)
}
got := s.Weights(ScopeRadio)
if got.TasteWeight != 2.5 || got.SkipPenalty != 3.0 {
t.Errorf("weights = %+v, want taste 2.5 / skip 3.0", got)
}
// The other profile is untouched.
if s.Weights(ScopeDailyMix) != ShippedDailyMixWeights() {
t.Error("daily_mix must be unaffected by a radio update")
}
// Values survive a fresh boot (persisted, not just cached), and the
// reconcile's ON CONFLICT DO NOTHING doesn't clobber tuned rows.
s2 := newService(t, pool)
if got := s2.Weights(ScopeRadio); got.TasteWeight != 2.5 {
t.Errorf("rebooted taste_weight = %v, want 2.5", got.TasteWeight)
}
rows := auditRows(t, pool)
if len(rows) != 1 {
t.Fatalf("audit rows = %d, want 1", len(rows))
}
if rows[0].Scope != ScopeRadio || rows[0].Action != "update" || len(rows[0].Changes) != 2 {
t.Errorf("audit row = %+v, want radio/update with 2 changes", rows[0])
}
// Changes are field-sorted: skip_penalty before taste_weight.
if rows[0].Changes[0].Field != "skip_penalty" || rows[0].Changes[0].New != 3.0 {
t.Errorf("changes[0] = %+v, want skip_penalty → 3.0", rows[0].Changes[0])
}
}
func TestUpdateProfile_Validation(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
cases := []struct {
name string
scope string
patch map[string]float64
want error
}{
{"unknown scope", "banana", map[string]float64{"taste_weight": 1}, ErrUnknownScope},
{"unknown field", ScopeRadio, map[string]float64{"vibes": 1}, ErrUnknownField},
{"over bound", ScopeRadio, map[string]float64{"taste_weight": 11}, ErrOutOfRange},
{"negative jitter", ScopeRadio, map[string]float64{"jitter_magnitude": -0.1}, ErrOutOfRange},
}
for _, c := range cases {
if err := s.UpdateProfile(context.Background(), c.scope, c.patch); !errors.Is(err, c.want) {
t.Errorf("%s: err = %v, want %v", c.name, err, c.want)
}
}
if s.Weights(ScopeRadio) != ShippedRadioWeights() {
t.Error("rejected patches must not partially apply")
}
if rows := auditRows(t, pool); len(rows) != 0 {
t.Errorf("rejected patches wrote %d audit rows, want 0", len(rows))
}
}
func TestUpdateTaste_CurveOrderingEnforced(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
// Moving neutral above full must reject on the PATCHED result.
err := s.UpdateTaste(context.Background(), map[string]float64{"engagement_neutral": 0.95})
if !errors.Is(err, ErrOutOfRange) {
t.Fatalf("err = %v, want ErrOutOfRange (curve ordering)", err)
}
// A coherent multi-point move is fine.
if err := s.UpdateTaste(context.Background(), map[string]float64{
"engagement_neutral": 0.40, "engagement_full": 0.95, "half_life_days": 30,
}); err != nil {
t.Fatalf("UpdateTaste: %v", err)
}
cfg := s.TasteConfig()
if cfg.HalfLifeDays != 30 || cfg.Engagement.NeutralCompletion != 0.40 {
t.Errorf("taste config = %+v, want tuned values", cfg)
}
// WindowDays scales with the half-life at the shipped ratio.
if cfg.WindowDays != 30*3.6 {
t.Errorf("WindowDays = %v, want %v", cfg.WindowDays, 30*3.6)
}
}
func TestReset_RestoresShippedAndAudits(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
// Reset with nothing changed is a no-op (no audit row).
if err := s.Reset(context.Background(), ScopeDailyMix); err != nil {
t.Fatalf("no-op reset: %v", err)
}
if rows := auditRows(t, pool); len(rows) != 0 {
t.Fatalf("no-op reset wrote audit rows")
}
if err := s.UpdateProfile(context.Background(), ScopeDailyMix,
map[string]float64{"similarity_weight": 4}); err != nil {
t.Fatalf("UpdateProfile: %v", err)
}
if err := s.Reset(context.Background(), ScopeDailyMix); err != nil {
t.Fatalf("Reset: %v", err)
}
if s.Weights(ScopeDailyMix) != ShippedDailyMixWeights() {
t.Error("reset must restore shipped defaults")
}
rows := auditRows(t, pool)
if len(rows) != 2 || rows[1].Action != "reset" {
t.Fatalf("audit rows = %+v, want update then reset", rows)
}
}
func TestUpdate_NoOpWritesNoAudit(t *testing.T) {
pool := newPool(t)
s := newService(t, pool)
shipped := ShippedRadioWeights()
if err := s.UpdateProfile(context.Background(), ScopeRadio,
map[string]float64{"taste_weight": shipped.TasteWeight}); err != nil {
t.Fatalf("UpdateProfile: %v", err)
}
if rows := auditRows(t, pool); len(rows) != 0 {
t.Errorf("no-op update wrote %d audit rows, want 0", len(rows))
}
}
+20 -1
View File
@@ -27,6 +27,7 @@ import (
"git.fabledsword.com/bvandeusen/minstrel/internal/mailer"
"git.fabledsword.com/bvandeusen/minstrel/internal/playevents"
"git.fabledsword.com/bvandeusen/minstrel/internal/playlists"
"git.fabledsword.com/bvandeusen/minstrel/internal/recsettings"
"git.fabledsword.com/bvandeusen/minstrel/internal/subsonic"
"git.fabledsword.com/bvandeusen/minstrel/internal/tracks"
"git.fabledsword.com/bvandeusen/minstrel/web"
@@ -88,6 +89,13 @@ type Server struct {
// PUT /api/me/timezone and POST /api/auth/register can call
// Refresh synchronously.
PlaylistScheduler *playlists.Scheduler
// RecSettings is the DB-backed recommendation tuning lab (#1250):
// scoring-weight profiles + taste-build knobs. Constructed in
// cmd/minstrel/main.go (it pushes daily-mix weights into package
// playlists at boot); the API layer reads radio weights per request
// and serves the admin tuning endpoints from it. Router() constructs
// a fallback when nil (tests).
RecSettings *recsettings.Service
// StreamSecret is the HMAC key used by /api/cast/stream-token to
// mint signed UPnP / Sonos stream URLs and by /api/tracks/{id}/stream
// to verify them. Sourced from config.Config.StreamSecret. Tests that
@@ -143,7 +151,18 @@ func (s *Server) Router() http.Handler {
if bus == nil {
bus = eventbus.New()
}
api.Mount(r, s.Pool, s.Logger, writer, s.RecommendationCfg, lidarrCfg, lidarrReqs, lidarrQuar, tracksSvc, playlistsSvc, s.CoverEnricher, s.CoverSettings, s.LibraryScanner, s.ScanCfg, s.DataDir, smtpSender, bus, s.PlaylistScheduler, s.StreamSecret)
recSettings := s.RecSettings
if recSettings == nil {
// Test contexts construct Server directly without main.go's
// boot wiring; reconcile here so radio + the admin tuning
// endpoints work against the same pool.
var err error
recSettings, err = recsettings.New(context.Background(), s.Pool, s.Logger)
if err != nil {
s.Logger.Error("server: recsettings boot failed", "err", err)
}
}
api.Mount(r, s.Pool, s.Logger, writer, s.RecommendationCfg, recSettings, lidarrCfg, lidarrReqs, lidarrQuar, tracksSvc, playlistsSvc, s.CoverEnricher, s.CoverSettings, s.LibraryScanner, s.ScanCfg, s.DataDir, smtpSender, bus, s.PlaylistScheduler, s.StreamSecret)
// /api/admin/scan is the only admin route owned by the server package
// (it needs the Scanner). Register it as a single inline-middleware
// route — using r.Route("/api/admin", ...) here would create a second
+22 -20
View File
@@ -1,18 +1,36 @@
import { createQuery } from '@tanstack/svelte-query';
import { api } from './client';
// Mirrors internal/api/me_recommendation_metrics.go.
export type RecommendationMetric = {
source: string;
// Mirrors internal/api/me_recommendation_metrics.go: raw play sources are
// bucketed server-side into stable surface families, grouped by intent, and
// anchored by the manual-plays baseline (milestone #127).
export type SurfaceMetric = {
key: string;
label: string;
plays: number;
skips: number;
skip_rate: number;
avg_completion: number;
low_confidence: boolean;
// Present when the surface's builder stamps pick-kind provenance and
// the window holds attributed plays (#1249, generalized #1270): For
// You's taste/fresh split, Discover's candidate buckets, the tiered
// mixes' tier1-3 — plus pre-attribution plays.
breakdown?: SurfaceMetric[];
};
export type SurfaceIntent = 'go_to' | 'discovery' | 'direct';
export type SurfaceGroup = {
intent: SurfaceIntent;
label: string;
surfaces: SurfaceMetric[];
};
export type RecommendationMetrics = {
window_days: number;
sources: RecommendationMetric[];
baseline: SurfaceMetric | null;
groups: SurfaceGroup[];
};
export function getRecommendationMetrics(): Promise<RecommendationMetrics> {
@@ -28,19 +46,3 @@ export function createRecommendationMetricsQuery() {
staleTime: 60_000
});
}
// Friendly labels for the system-playlist source keys (play_events.source).
const SOURCE_LABELS: Record<string, string> = {
for_you: 'For You',
discover: 'Discover',
deep_cuts: 'Deep cuts',
rediscover: 'Rediscover',
new_for_you: 'New for you',
on_this_day: 'On this day',
first_listens: 'First listens',
songs_like_artist: 'Songs like…'
};
export function sourceLabel(source: string): string {
return SOURCE_LABELS[source] ?? source;
}
+86
View File
@@ -0,0 +1,86 @@
import { api } from './client';
// Mirrors internal/api/admin_recommendation_tuning.go (#1250): the
// recommendation tuning lab. Field names are the server's snake_case
// keys, used verbatim in PATCH bodies.
export type WeightProfile = {
base_weight: number;
like_boost: number;
recency_weight: number;
skip_penalty: number;
jitter_magnitude: number;
context_weight: number;
similarity_weight: number;
taste_weight: number;
};
export type TasteTuning = {
half_life_days: number;
engagement_hard_skip: number;
engagement_neutral: number;
engagement_full: number;
};
export type TuningScope = 'radio' | 'daily_mix' | 'taste';
export type TuningSnapshot = {
profiles: Record<'radio' | 'daily_mix', WeightProfile>;
taste: TasteTuning;
shipped: {
profiles: Record<'radio' | 'daily_mix', WeightProfile>;
taste: TasteTuning;
};
};
export function getTuning(): Promise<TuningSnapshot> {
return api.get<TuningSnapshot>('/api/admin/recommendation-tuning');
}
export function patchTuning(
scope: TuningScope,
values: Record<string, number>
): Promise<TuningSnapshot> {
return api.patch<TuningSnapshot>(`/api/admin/recommendation-tuning/${scope}`, { values });
}
export function resetTuning(scope: TuningScope): Promise<TuningSnapshot> {
return api.post<TuningSnapshot>(`/api/admin/recommendation-tuning/${scope}/reset`, {});
}
// Trends (#1251): weekly per-surface outcome series with knob-turn
// markers. Mirrors internal/api/admin_recommendation_trends.go.
export type TrendPoint = {
week_start: string;
plays: number;
skips: number;
skip_rate: number;
avg_completion: number;
taste_hit_rate: number;
};
export type TrendSeries = {
key: string;
label: string;
intent: string;
plays: number;
points: TrendPoint[];
};
export type TrendMarker = {
changed_at: string;
scope: string;
action: string;
changes: { field: string; old: number; new: number }[];
};
export type TrendsResponse = {
weeks: number;
series: TrendSeries[];
markers: TrendMarker[];
};
export function getTrends(weeks = 12): Promise<TrendsResponse> {
return api.get<TrendsResponse>(`/api/admin/recommendation-trends?weeks=${weeks}`);
}
+1
View File
@@ -10,6 +10,7 @@
{ href: '/admin/quarantine', label: 'Quarantine' },
{ href: '/admin/playback-errors', label: 'Playback errors' },
{ href: '/admin/diagnostics', label: 'Diagnostics' },
{ href: '/admin/tuning', label: 'Tuning' },
{ href: '/admin/users', label: 'Users' }
];
+2 -1
View File
@@ -52,7 +52,7 @@ describe('AdminTabs', () => {
);
});
test('renders all seven tabs in order', () => {
test('renders all eight tabs in order', () => {
state.pageUrl = new URL('http://localhost/admin');
render(AdminTabs);
const links = screen.getAllByRole('link');
@@ -63,6 +63,7 @@ describe('AdminTabs', () => {
'Quarantine',
'Playback errors',
'Diagnostics',
'Tuning',
'Users'
]);
});
+434
View File
@@ -0,0 +1,434 @@
<script lang="ts">
import { pageTitle } from '$lib/branding';
import {
getTuning,
patchTuning,
resetTuning,
getTrends,
type TuningScope,
type TuningSnapshot,
type WeightProfile,
type TasteTuning,
type TrendsResponse,
type TrendSeries,
type TrendMarker
} from '$lib/api/tuning';
import { errMessage } from '$lib/api/errors';
import { pushToast } from '$lib/stores/toast.svelte';
// The defaults-discovery lab (#1250): DB-backed scoring-weight
// profiles + taste-build knobs with live effect. This card exists to
// FIND good values — found-good values get baked into shipped
// defaults, so end users and other operators never need it.
const weightFields: { key: keyof WeightProfile; label: string; hint: string }[] = [
{ key: 'base_weight', label: 'Base weight', hint: 'Floor score every candidate starts from.' },
{ key: 'like_boost', label: 'Like boost', hint: 'Added when the track is liked.' },
{ key: 'recency_weight', label: 'Recency weight', hint: 'Rewards tracks not played recently (030d ramp).' },
{ key: 'skip_penalty', label: 'Skip penalty', hint: 'Subtracts skips/plays ratio.' },
{ key: 'jitter_magnitude', label: 'Jitter', hint: 'Random reshuffle magnitude for near-ties.' },
{ key: 'context_weight', label: 'Context weight', hint: 'Session-vector similarity contribution.' },
{ key: 'similarity_weight', label: 'Similarity weight', hint: 'Seed-similarity contribution.' },
{ key: 'taste_weight', label: 'Taste weight', hint: 'Learned taste-profile fit, in [-1, +1].' }
];
const tasteFields: { key: keyof TasteTuning; label: string; hint: string }[] = [
{ key: 'half_life_days', label: 'Half-life (days)', hint: "A play's influence halves every this-many days." },
{ key: 'engagement_hard_skip', label: 'Hard-skip point', hint: 'Completion at/below which a play reads 1.' },
{ key: 'engagement_neutral', label: 'Neutral point', hint: 'Completion at which a play reads 0.' },
{ key: 'engagement_full', label: 'Full point', hint: 'Completion at/above which a play reads +1.' }
];
const profileScopes: { scope: 'radio' | 'daily_mix'; label: string; blurb: string }[] = [
{ scope: 'radio', label: 'Radio', blurb: 'Seed-directed listening — the user picked a direction.' },
{ scope: 'daily_mix', label: 'Daily mixes', blurb: 'For You, Songs like…, and the discovery mixes.' }
];
let snapshot = $state<TuningSnapshot | null>(null);
let loadFailed = $state(false);
// Editable copies, string-typed for the inputs; parsed on save.
let form = $state<Record<string, Record<string, string>>>({});
let saving = $state<TuningScope | null>(null);
function fillForm(snap: TuningSnapshot) {
const f: Record<string, Record<string, string>> = { radio: {}, daily_mix: {}, taste: {} };
for (const p of ['radio', 'daily_mix'] as const) {
for (const { key } of weightFields) f[p][key] = String(snap.profiles[p][key]);
}
for (const { key } of tasteFields) f.taste[key] = String(snap.taste[key]);
form = f;
}
$effect(() => {
getTuning()
.then((snap) => {
fillForm(snap);
snapshot = snap;
})
.catch(() => {
loadFailed = true;
});
});
// A knob deviates when its CURRENT SAVED value differs from shipped;
// the dot marks where this install has drifted from defaults.
function deviates(scope: 'radio' | 'daily_mix' | 'taste', key: string): boolean {
if (!snapshot) return false;
if (scope === 'taste') {
return snapshot.taste[key as keyof TasteTuning] !== snapshot.shipped.taste[key as keyof TasteTuning];
}
return (
snapshot.profiles[scope][key as keyof WeightProfile] !==
snapshot.shipped.profiles[scope][key as keyof WeightProfile]
);
}
function currentValue(scope: TuningScope, key: string): number {
if (!snapshot) return 0;
if (scope === 'taste') return snapshot.taste[key as keyof TasteTuning];
return snapshot.profiles[scope][key as keyof WeightProfile];
}
async function save(scope: TuningScope) {
if (!snapshot) return;
const values: Record<string, number> = {};
for (const [key, raw] of Object.entries(form[scope] ?? {})) {
const v = Number(raw);
if (!Number.isFinite(v)) {
pushToast(`${key} is not a number.`, 'error');
return;
}
if (v !== currentValue(scope, key)) values[key] = v;
}
if (Object.keys(values).length === 0) {
pushToast('Nothing changed.');
return;
}
saving = scope;
try {
snapshot = await patchTuning(scope, values);
fillForm(snapshot);
pushToast('Saved — takes effect on the next scoring pass.');
} catch (e: unknown) {
pushToast(`Save failed: ${errMessage(e)}`, 'error');
} finally {
saving = null;
}
}
async function reset(scope: TuningScope) {
saving = scope;
try {
snapshot = await resetTuning(scope);
fillForm(snapshot);
pushToast('Reset to shipped defaults.');
} catch (e: unknown) {
pushToast(`Reset failed: ${errMessage(e)}`, 'error');
} finally {
saving = null;
}
}
// Trends (#1251): weekly skip-rate sparklines per surface with
// knob-turn markers — the verify half of the tune→verify loop.
let trends = $state<TrendsResponse | null>(null);
let trendsFailed = $state(false);
$effect(() => {
getTrends()
.then((t) => {
trends = t;
})
.catch(() => {
trendsFailed = true;
});
});
const SPARK_W = 220;
const SPARK_H = 36;
const TREND_LOW_VOLUME = 20;
// The week axis is the sorted union of every series' buckets, so all
// sparklines and markers share one x scale.
const weekAxis = $derived.by(() => {
if (!trends) return [] as string[];
const set = new Set<string>();
for (const s of trends.series) for (const p of s.points) set.add(p.week_start);
return [...set].sort();
});
function xFor(week: string): number {
const i = weekAxis.indexOf(week);
if (i < 0 || weekAxis.length < 2) return 0;
return (i / (weekAxis.length - 1)) * SPARK_W;
}
// Polyline of the series' weekly skip rate on a fixed [0,1] y-scale
// (higher = worse, drawn upward) so rows are visually comparable.
function sparkPoints(s: TrendSeries): string {
return s.points
.map((p) => `${xFor(p.week_start).toFixed(1)},${(SPARK_H - p.skip_rate * SPARK_H).toFixed(1)}`)
.join(' ');
}
// A marker lands on the latest axis week that starts at/before it.
function markerX(m: TrendMarker): number {
const day = m.changed_at.slice(0, 10);
let idx = -1;
for (let i = 0; i < weekAxis.length; i++) {
if (weekAxis[i] <= day) idx = i;
}
if (idx < 0 || weekAxis.length < 2) return 0;
return (idx / (weekAxis.length - 1)) * SPARK_W;
}
function windowTasteHitRate(s: TrendSeries): number {
let plays = 0;
let hits = 0;
for (const p of s.points) {
plays += p.plays;
hits += p.taste_hit_rate * p.plays;
}
return plays > 0 ? hits / plays : 0;
}
function latest(s: TrendSeries): { skip: number; completion: number } {
const last = s.points[s.points.length - 1];
return last ? { skip: last.skip_rate, completion: last.avg_completion } : { skip: 0, completion: 0 };
}
function pct(v: number): string {
return `${(v * 100).toFixed(0)}%`;
}
function markerSummary(m: TrendMarker): string {
const when = m.changed_at.slice(0, 10);
if (m.action === 'reset') return `${when}${m.scope} reset to defaults`;
const fields = (m.changes ?? []).map((c) => `${c.field} ${c.old}${c.new}`).join(', ');
return `${when}${m.scope}: ${fields}`;
}
</script>
<svelte:head>
<title>{pageTitle('Tuning')}</title>
</svelte:head>
<div class="space-y-6 p-4">
<div>
<h1 class="text-xl font-semibold">Recommendation tuning</h1>
<p class="mt-1 max-w-3xl text-sm text-text-secondary">
The defaults-discovery lab. Changes apply live — radio on the next request, daily mixes on
the next rebuild — and every change is recorded so the metrics page can tie outcome shifts
to knob turns. Found-good values get baked into shipped defaults; a dot marks knobs that
currently deviate from them.
</p>
</div>
{#if loadFailed}
<p class="text-sm text-text-secondary">Couldn't load tuning settings.</p>
{:else if !snapshot}
<p class="text-sm text-text-secondary">Loading…</p>
{:else}
<section class="grid gap-4 lg:grid-cols-2">
{#each profileScopes as p (p.scope)}
<div class="space-y-3 rounded border border-border bg-surface p-4">
<div>
<h2 class="text-lg font-semibold">{p.label}</h2>
<p class="text-xs text-text-secondary">{p.blurb}</p>
</div>
<div class="space-y-2">
{#each weightFields as f (f.key)}
<div class="grid grid-cols-[1fr_7rem] items-center gap-2">
<label class="text-sm" for="{p.scope}-{f.key}" title={f.hint}>
{f.label}
{#if deviates(p.scope, f.key)}
<span
class="ml-1 inline-block h-1.5 w-1.5 rounded-full bg-accent align-middle"
title="Deviates from the shipped default ({snapshot.shipped.profiles[p.scope][f.key]})"
></span>
{/if}
</label>
<input
id="{p.scope}-{f.key}"
type="number"
step="0.1"
bind:value={form[p.scope][f.key]}
class="w-full rounded border border-border bg-background px-2 py-1 text-right text-sm tabular-nums outline-none focus:border-accent"
/>
</div>
{/each}
</div>
<div class="flex gap-2">
<button
type="button"
disabled={saving !== null}
onclick={() => save(p.scope)}
class="rounded bg-accent px-3 py-1.5 text-sm text-white disabled:opacity-50"
>
Save {p.label.toLowerCase()}
</button>
<button
type="button"
disabled={saving !== null}
onclick={() => reset(p.scope)}
class="rounded border border-border px-3 py-1.5 text-sm text-text-secondary hover:text-text-primary disabled:opacity-50"
>
Reset to defaults
</button>
</div>
</div>
{/each}
</section>
<section class="space-y-3 rounded border border-border bg-surface p-4 lg:max-w-xl">
<div>
<h2 class="text-lg font-semibold">Taste profile build</h2>
<p class="text-xs text-text-secondary">
How plays become the learned taste profile: the influence half-life and the
completion→engagement curve (must stay ordered: hard-skip &lt; neutral &lt; full).
</p>
</div>
<div class="space-y-2">
{#each tasteFields as f (f.key)}
<div class="grid grid-cols-[1fr_7rem] items-center gap-2">
<label class="text-sm" for="taste-{f.key}" title={f.hint}>
{f.label}
{#if deviates('taste', f.key)}
<span
class="ml-1 inline-block h-1.5 w-1.5 rounded-full bg-accent align-middle"
title="Deviates from the shipped default ({snapshot.shipped.taste[f.key]})"
></span>
{/if}
</label>
<input
id="taste-{f.key}"
type="number"
step={f.key === 'half_life_days' ? '1' : '0.05'}
bind:value={form.taste[f.key]}
class="w-full rounded border border-border bg-background px-2 py-1 text-right text-sm tabular-nums outline-none focus:border-accent"
/>
</div>
{/each}
</div>
<div class="flex gap-2">
<button
type="button"
disabled={saving !== null}
onclick={() => save('taste')}
class="rounded bg-accent px-3 py-1.5 text-sm text-white disabled:opacity-50"
>
Save taste
</button>
<button
type="button"
disabled={saving !== null}
onclick={() => reset('taste')}
class="rounded border border-border px-3 py-1.5 text-sm text-text-secondary hover:text-text-primary disabled:opacity-50"
>
Reset to defaults
</button>
</div>
</section>
{/if}
<!-- Weekly trends (#1251): the verify half. Sparklines share one
week axis; dashed ticks mark knob turns so cause→effect reads
off the chart. -->
<section class="space-y-3 rounded border border-border bg-surface p-4">
<div>
<h2 class="text-lg font-semibold">Weekly trends</h2>
<p class="text-xs text-text-secondary">
Skip rate per surface over the last {trends?.weeks ?? 12} weeks (lower is better; all
users aggregated, rates only). Dashed ticks mark tuning changes. Taste hit is the share
of plays whose artist fits the current taste profile.
</p>
</div>
{#if trendsFailed}
<p class="text-sm text-text-secondary">Couldn't load trends.</p>
{:else if !trends}
<p class="text-sm text-text-secondary">Loading…</p>
{:else if trends.series.length === 0}
<p class="text-sm text-text-secondary">
No plays recorded yet — trends appear once listening accumulates.
</p>
{:else}
<table class="w-full text-sm">
<thead>
<tr class="text-left text-text-secondary">
<th class="py-1 font-medium">Surface</th>
<th class="py-1 font-medium">Skip rate by week</th>
<th class="py-1 text-right font-medium">Plays</th>
<th class="py-1 text-right font-medium">Latest skip</th>
<th class="py-1 text-right font-medium">Latest completion</th>
<th class="py-1 text-right font-medium">Taste hit</th>
</tr>
</thead>
<tbody>
{#each trends.series as s (s.key)}
<tr
class="border-t border-border"
class:opacity-60={s.plays < TREND_LOW_VOLUME}
title={s.plays < TREND_LOW_VOLUME
? 'Fewer than 20 plays in the window — treat as anecdote, not signal.'
: undefined}
>
<td class="py-1.5 pr-2">{s.label}</td>
<td class="py-1.5">
<svg
width={SPARK_W}
height={SPARK_H}
viewBox="0 0 {SPARK_W} {SPARK_H}"
role="img"
aria-label="{s.label} weekly skip rate"
data-testid="sparkline-{s.key}"
>
<line x1="0" y1={SPARK_H - 0.5} x2={SPARK_W} y2={SPARK_H - 0.5}
stroke="currentColor" opacity="0.15" />
{#each trends.markers as m, i (i)}
<line
x1={markerX(m)} y1="0" x2={markerX(m)} y2={SPARK_H}
stroke="currentColor" opacity="0.35" stroke-dasharray="2,2"
>
<title>{markerSummary(m)}</title>
</line>
{/each}
{#if s.points.length > 1}
<polyline
class="text-accent"
points={sparkPoints(s)}
fill="none"
stroke="currentColor"
stroke-width="1.5"
/>
{:else if s.points.length === 1}
<circle
class="text-accent"
cx={xFor(s.points[0].week_start)}
cy={SPARK_H - s.points[0].skip_rate * SPARK_H}
r="2" fill="currentColor"
/>
{/if}
</svg>
</td>
<td class="py-1.5 text-right tabular-nums">{s.plays}</td>
<td class="py-1.5 text-right tabular-nums">{pct(latest(s).skip)}</td>
<td class="py-1.5 text-right tabular-nums">{pct(latest(s).completion)}</td>
<td class="py-1.5 text-right tabular-nums">{pct(windowTasteHitRate(s))}</td>
</tr>
{/each}
</tbody>
</table>
{#if trends.markers.length > 0}
<div class="space-y-1">
<h3 class="text-sm font-medium">Tuning changes in this window</h3>
<ul class="space-y-0.5 text-xs text-text-secondary">
{#each trends.markers as m, i (i)}
<li>{markerSummary(m)}</li>
{/each}
</ul>
</div>
{/if}
{/if}
</section>
</div>
+197
View File
@@ -0,0 +1,197 @@
import { describe, expect, test, vi, beforeEach } from 'vitest';
import { render, screen, fireEvent, waitFor } from '@testing-library/svelte';
import type { TuningSnapshot, TrendsResponse } from '$lib/api/tuning';
vi.mock('$lib/api/tuning', async (orig) => {
const actual = (await orig()) as Record<string, unknown>;
return {
...actual,
getTuning: vi.fn(),
patchTuning: vi.fn(),
resetTuning: vi.fn(),
getTrends: vi.fn()
};
});
const pushToast = vi.fn();
vi.mock('$lib/stores/toast.svelte', () => ({ pushToast: (...a: unknown[]) => pushToast(...a) }));
import TuningPage from './+page.svelte';
import { getTuning, patchTuning, resetTuning, getTrends } from '$lib/api/tuning';
const weights = (over: Partial<Record<string, number>> = {}) => ({
base_weight: 1,
like_boost: 2,
recency_weight: 1,
skip_penalty: 2,
jitter_magnitude: 0.1,
context_weight: 0.5,
similarity_weight: 1.5,
taste_weight: 1.5,
...over
});
const taste = (over: Partial<Record<string, number>> = {}) => ({
half_life_days: 75,
engagement_hard_skip: 0.05,
engagement_neutral: 0.3,
engagement_full: 0.9,
...over
});
function snapshot(over: Partial<TuningSnapshot> = {}): TuningSnapshot {
return {
profiles: { radio: weights({ taste_weight: 1 }), daily_mix: weights() },
taste: taste(),
shipped: {
profiles: { radio: weights({ taste_weight: 1 }), daily_mix: weights() },
taste: taste()
},
...over
} as TuningSnapshot;
}
const emptyTrends: TrendsResponse = { weeks: 12, series: [], markers: [] };
beforeEach(() => {
vi.clearAllMocks();
(getTrends as ReturnType<typeof vi.fn>).mockResolvedValue(emptyTrends);
});
describe('Admin tuning page', () => {
test('renders both profiles and the taste card with current values', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
render(TuningPage);
await waitFor(() => expect(screen.getByText('Radio')).toBeInTheDocument());
expect(screen.getByText('Daily mixes')).toBeInTheDocument();
expect(screen.getByText('Taste profile build')).toBeInTheDocument();
const radioTaste = screen.getByLabelText(/taste weight/i, {
selector: '#radio-taste_weight'
}) as HTMLInputElement;
expect(radioTaste.value).toBe('1');
const halfLife = document.getElementById('taste-half_life_days') as HTMLInputElement;
expect(halfLife.value).toBe('75');
});
test('save sends only the changed fields for the scope', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
(patchTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
render(TuningPage);
await waitFor(() => expect(screen.getByText('Radio')).toBeInTheDocument());
const input = document.getElementById('radio-taste_weight') as HTMLInputElement;
await fireEvent.input(input, { target: { value: '2.5' } });
await fireEvent.click(screen.getByRole('button', { name: /save radio/i }));
await waitFor(() =>
expect(patchTuning).toHaveBeenCalledWith('radio', { taste_weight: 2.5 })
);
});
test('save with no changes calls nothing and says so', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
render(TuningPage);
await waitFor(() => expect(screen.getByText('Radio')).toBeInTheDocument());
await fireEvent.click(screen.getByRole('button', { name: /save radio/i }));
expect(patchTuning).not.toHaveBeenCalled();
await waitFor(() => expect(pushToast).toHaveBeenCalledWith('Nothing changed.'));
});
test('reset calls resetTuning for the scope', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
(resetTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
render(TuningPage);
await waitFor(() => expect(screen.getByText('Taste profile build')).toBeInTheDocument());
const resetButtons = screen.getAllByRole('button', { name: /reset to defaults/i });
await fireEvent.click(resetButtons[resetButtons.length - 1]);
await waitFor(() => expect(resetTuning).toHaveBeenCalledWith('taste'));
});
test('marks knobs that deviate from shipped defaults', async () => {
const snap = snapshot();
snap.profiles.daily_mix = weights({ similarity_weight: 4 });
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snap);
render(TuningPage);
await waitFor(() => expect(screen.getByText('Daily mixes')).toBeInTheDocument());
expect(
screen.getByTitle(/deviates from the shipped default \(1\.5\)/i)
).toBeInTheDocument();
});
test('renders weekly trend rows with sparklines and knob-turn markers (#1251)', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
(getTrends as ReturnType<typeof vi.fn>).mockResolvedValue({
weeks: 12,
series: [
{
key: 'radio',
label: 'Radio',
intent: 'go_to',
plays: 40,
points: [
{
week_start: '2026-06-22',
plays: 25,
skips: 5,
skip_rate: 0.2,
avg_completion: 0.8,
taste_hit_rate: 0.6
},
{
week_start: '2026-06-29',
plays: 15,
skips: 6,
skip_rate: 0.4,
avg_completion: 0.7,
taste_hit_rate: 0.5
}
]
},
{
key: 'discover',
label: 'Discover',
intent: 'discovery',
plays: 5,
points: [
{
week_start: '2026-06-29',
plays: 5,
skips: 3,
skip_rate: 0.6,
avg_completion: 0.4,
taste_hit_rate: 0.2
}
]
}
],
markers: [
{
changed_at: '2026-06-30T10:00:00Z',
scope: 'radio',
action: 'update',
changes: [{ field: 'taste_weight', old: 1, new: 2 }]
}
]
} satisfies TrendsResponse);
render(TuningPage);
await waitFor(() => expect(screen.getByText('Weekly trends')).toBeInTheDocument());
expect(screen.getByTestId('sparkline-radio')).toBeInTheDocument();
expect(screen.getByTestId('sparkline-discover')).toBeInTheDocument();
// Latest skip rate column for radio = 40% (also discover's latest
// completion, hence getAllBy).
expect(screen.getAllByText('40%').length).toBeGreaterThan(0);
// The knob turn is listed under the chart AND tooltipped on each
// sparkline's marker tick, hence getAllBy.
expect(
screen.getAllByText(/2026-06-30 — radio: taste_weight 1→2/).length
).toBeGreaterThan(0);
// Low-volume series (5 plays) is dimmed as anecdote.
expect(screen.getByTitle(/fewer than 20 plays in the window/i)).toBeInTheDocument();
});
test('empty trends show the accumulation hint', async () => {
(getTuning as ReturnType<typeof vi.fn>).mockResolvedValue(snapshot());
render(TuningPage);
await waitFor(() =>
expect(screen.getByText(/trends appear once listening accumulates/i)).toBeInTheDocument()
);
});
});
+146 -11
View File
@@ -10,8 +10,9 @@
} from '$lib/api/listenbrainz';
import {
createRecommendationMetricsQuery,
sourceLabel,
type RecommendationMetrics
type RecommendationMetrics,
type SurfaceIntent,
type SurfaceMetric
} from '$lib/api/metrics';
import { theme, setTheme, type ThemePreference } from '$lib/stores/theme.svelte';
import { player, setCrossfade } from '$lib/player/store.svelte';
@@ -30,6 +31,46 @@
const status = createLBStatusQuery() as CreateQueryResult<LBStatus>;
const metrics = createRecommendationMetricsQuery() as CreateQueryResult<RecommendationMetrics>;
// Per-intent expectation copy: each surface band is judged against its
// job — discovery mixes are supposed to run hotter skip rates.
const intentHints: Record<SurfaceIntent, string> = {
go_to: 'Everyday surfaces — these should beat your own picking.',
discovery: 'Exploration surfaces — higher skip rates here are expected and healthy.',
direct: 'Music you chose yourself — naturally close to the baseline.'
};
function pct(v: number): string {
return `${(v * 100).toFixed(0)}%`;
}
// Delta in percentage points vs the baseline, signed ("+12" / "5").
function deltaPts(value: number, baseline: number): string {
const pts = Math.round((value - baseline) * 100);
return pts > 0 ? `+${pts}` : `${pts}`;
}
// A surface's skip delta is "worse" when it skips more than the
// baseline; completion delta is "worse" when it completes less.
function skipDeltaClass(m: SurfaceMetric, baseline: SurfaceMetric): string {
return m.skip_rate > baseline.skip_rate ? 'text-danger' : 'text-text-secondary';
}
function completionDeltaClass(m: SurfaceMetric, baseline: SurfaceMetric): string {
return m.avg_completion < baseline.avg_completion ? 'text-danger' : 'text-text-secondary';
}
// Pick-kind breakdowns are collapsed by default (#1270): with every
// system mix stamping provenance, always-open sub-rows would triple
// the table height and bury the surface-level comparison.
let expandedBreakdowns = $state(new Set<string>());
function toggleBreakdown(key: string) {
const next = new Set(expandedBreakdowns);
if (next.has(key)) next.delete(key);
else next.add(key);
expandedBreakdowns = next;
}
const tokenMutation = createTokenMutation(queryClient);
const enabledMutation = createEnabledMutation(queryClient);
@@ -278,14 +319,32 @@
<h2 class="text-lg font-semibold">Recommendation metrics</h2>
<p class="text-sm text-text-secondary">
How plays launched from each recommendation surface land, over the last
{$metrics.data?.window_days ?? 30} days. Lower skip rate and higher average
completion mean the surface is hitting.
{$metrics.data?.window_days ?? 30} days — compared against the baseline of
music you picked yourself. Deltas are percentage points vs that baseline.
</p>
{#if $metrics.isPending}
<p class="text-sm text-text-secondary">Loading…</p>
{:else if $metrics.isError}
<p class="text-sm text-text-secondary">Couldn't load metrics.</p>
{:else if $metrics.data && $metrics.data.sources.length > 0}
{:else if $metrics.data && ($metrics.data.groups.length > 0 || $metrics.data.baseline)}
{@const baseline = $metrics.data.baseline}
{#if baseline}
<p class="rounded bg-background px-3 py-2 text-sm">
<span class="font-medium">Baseline — manual plays:</span>
<span class="text-text-secondary">
{baseline.plays} plays · {pct(baseline.skip_rate)} skip ·
{pct(baseline.avg_completion)} completion
</span>
</p>
{:else}
<p class="text-sm text-text-secondary">
No manual plays in this window yet, so deltas are hidden — raw rates only.
</p>
{/if}
{#each $metrics.data.groups as group (group.intent)}
<div class="space-y-1">
<h3 class="text-sm font-medium">{group.label}</h3>
<p class="text-xs text-text-secondary">{intentHints[group.intent]}</p>
<table class="w-full text-sm">
<thead>
<tr class="text-left text-text-secondary">
@@ -296,19 +355,95 @@
</tr>
</thead>
<tbody>
{#each $metrics.data.sources as m (m.source)}
<tr class="border-t border-border">
<td class="py-1">{sourceLabel(m.source)}</td>
{#each group.surfaces as m (m.key)}
<tr
class="border-t border-border"
class:opacity-60={m.low_confidence}
title={m.low_confidence
? 'Fewer than 20 plays — treat these rates as anecdote, not signal.'
: undefined}
>
<td class="py-1">
{#if (m.breakdown ?? []).length > 0}
<button
type="button"
class="inline-flex items-center gap-1 rounded outline-none focus-visible:ring-1 focus-visible:ring-accent"
aria-expanded={expandedBreakdowns.has(m.key)}
onclick={() => toggleBreakdown(m.key)}
>
<span class="text-xs text-text-secondary" aria-hidden="true">
{expandedBreakdowns.has(m.key) ? '▾' : '▸'}
</span>
{m.label}
</button>
{:else}
{m.label}
{/if}{#if m.low_confidence}<span class="ml-1 text-xs text-text-secondary">· low data</span>{/if}
</td>
<td class="py-1 text-right tabular-nums">{m.plays}</td>
<td class="py-1 text-right tabular-nums">{(m.skip_rate * 100).toFixed(0)}%</td>
<td class="py-1 text-right tabular-nums">{(m.avg_completion * 100).toFixed(0)}%</td>
<td class="py-1 text-right tabular-nums">
{pct(m.skip_rate)}
{#if baseline}
<span class="ml-1 text-xs {skipDeltaClass(m, baseline)}">
{deltaPts(m.skip_rate, baseline.skip_rate)}
</span>
{/if}
</td>
<td class="py-1 text-right tabular-nums">
{pct(m.avg_completion)}
{#if baseline}
<span class="ml-1 text-xs {completionDeltaClass(m, baseline)}">
{deltaPts(m.avg_completion, baseline.avg_completion)}
</span>
{/if}
</td>
</tr>
<!-- Pick-kind provenance sub-rows (#1249/#1270): each
surface splits into the populations its builder
stamped — For You's taste vs freshness injection,
Discover's buckets, the tiered mixes' tiers — the
numbers that say WHERE a surface's skips come from.
Toggled per surface to keep the default view flat. -->
{#if expandedBreakdowns.has(m.key)}
{#each m.breakdown ?? [] as b (b.key)}
<tr
class="border-t border-border/50 text-text-secondary"
class:opacity-60={b.low_confidence}
title={b.low_confidence
? 'Fewer than 20 plays — treat these rates as anecdote, not signal.'
: undefined}
>
<td class="py-1 pl-4 text-xs">
{b.label}{#if b.low_confidence}<span class="ml-1">· low data</span>{/if}
</td>
<td class="py-1 text-right text-xs tabular-nums">{b.plays}</td>
<td class="py-1 text-right text-xs tabular-nums">
{pct(b.skip_rate)}
{#if baseline}
<span class="ml-1 {skipDeltaClass(b, baseline)}">
{deltaPts(b.skip_rate, baseline.skip_rate)}
</span>
{/if}
</td>
<td class="py-1 text-right text-xs tabular-nums">
{pct(b.avg_completion)}
{#if baseline}
<span class="ml-1 {completionDeltaClass(b, baseline)}">
{deltaPts(b.avg_completion, baseline.avg_completion)}
</span>
{/if}
</td>
</tr>
{/each}
{/if}
{/each}
</tbody>
</table>
</div>
{/each}
{:else}
<p class="text-sm text-text-secondary">
No recommendation plays yet. Play something from For You, Discover, or a mix.
No plays recorded yet. Play something from For You, Discover, or a mix.
</p>
{/if}
</section>
+2 -3
View File
@@ -28,11 +28,10 @@ vi.mock('$lib/api/listenbrainz', () => {
vi.mock('$lib/api/metrics', () => ({
createRecommendationMetricsQuery: () => ({
subscribe: (run: (v: unknown) => void) => {
run({ isPending: false, isError: false, data: { window_days: 30, sources: [] } });
run({ isPending: false, isError: false, data: { window_days: 30, baseline: null, groups: [] } });
return () => {};
}
}),
sourceLabel: (s: string) => s
})
}));
beforeEach(() => {
+87 -4
View File
@@ -20,14 +20,19 @@ vi.mock('$lib/api/me', () => ({
regenerateAPIToken: vi.fn()
}));
// Mutable holder so individual tests can inject populated metrics;
// vi.mock is hoisted, hence vi.hoisted for the shared reference.
const metricsMock = vi.hoisted(() => ({
data: { window_days: 30, baseline: null, groups: [] } as unknown
}));
vi.mock('$lib/api/metrics', () => ({
createRecommendationMetricsQuery: () => ({
subscribe: (run: (v: unknown) => void) => {
run({ isPending: false, isError: false, data: { window_days: 30, sources: [] } });
run({ isPending: false, isError: false, data: metricsMock.data });
return () => {};
}
}),
sourceLabel: (s: string) => s
})
}));
import SettingsPage from './+page.svelte';
@@ -53,7 +58,10 @@ function mockMutationStore(mutateFn: (vars: unknown) => void = vi.fn()) {
return readable({ isPending: false, mutate: mutateFn, mutateAsync: vi.fn() });
}
afterEach(() => vi.clearAllMocks());
afterEach(() => {
vi.clearAllMocks();
metricsMock.data = { window_days: 30, baseline: null, groups: [] };
});
describe('Settings page — ListenBrainz', () => {
test('token-not-set state renders input + Save button', async () => {
@@ -178,6 +186,81 @@ describe('Settings page — Password card', () => {
});
});
describe('Settings page — Recommendation metrics card', () => {
const metric = (key: string, label: string, over: Record<string, unknown> = {}) => ({
key,
label,
plays: 30,
skips: 3,
skip_rate: 0.1,
avg_completion: 0.9,
low_confidence: false,
...over
});
test('breakdown rows are collapsed by default and expand on toggle (#1249/#1270)', async () => {
setupPage();
metricsMock.data = {
window_days: 30,
baseline: metric('manual', 'Manual library plays', { plays: 100 }),
groups: [
{
intent: 'go_to',
label: 'Go-to surfaces',
surfaces: [
metric('for_you', 'For You', {
plays: 45,
breakdown: [
metric('for_you_taste', 'Taste picks'),
metric('for_you_fresh', 'Fresh picks', {
plays: 10,
skip_rate: 0.4,
low_confidence: true
}),
metric('for_you_unattributed', 'Earlier plays', { plays: 5 })
]
})
]
}
]
};
render(SettingsPage);
const toggle = await screen.findByRole('button', { name: /for you/i });
// Collapsed by default: every stamping mix now carries a breakdown,
// so always-open sub-rows would swamp the surface-level view.
expect(toggle).toHaveAttribute('aria-expanded', 'false');
expect(screen.queryByText(/Taste picks/)).not.toBeInTheDocument();
await fireEvent.click(toggle);
expect(toggle).toHaveAttribute('aria-expanded', 'true');
expect(screen.getByText(/Taste picks/)).toBeInTheDocument();
expect(screen.getByText(/Fresh picks/)).toBeInTheDocument();
expect(screen.getByText(/Earlier plays/)).toBeInTheDocument();
await fireEvent.click(toggle);
expect(screen.queryByText(/Taste picks/)).not.toBeInTheDocument();
});
test('surfaces without a breakdown render no toggle and no sub-rows', async () => {
setupPage();
metricsMock.data = {
window_days: 30,
baseline: null,
groups: [
{
intent: 'go_to',
label: 'Go-to surfaces',
surfaces: [metric('radio', 'Radio')]
}
]
};
render(SettingsPage);
await waitFor(() => expect(screen.getByText('Radio')).toBeInTheDocument());
expect(screen.queryByRole('button', { name: /radio/i })).not.toBeInTheDocument();
expect(screen.queryByText(/Taste picks/)).not.toBeInTheDocument();
});
});
describe('Settings page — API Token card', () => {
test('first click on Regenerate shows "Click again to confirm"', async () => {
setupPage();