diff --git a/cmd/minstrel/main.go b/cmd/minstrel/main.go
index 4e887257..b9e9d00c 100644
--- a/cmd/minstrel/main.go
+++ b/cmd/minstrel/main.go
@@ -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,
diff --git a/internal/api/admin_recommendation_trends.go b/internal/api/admin_recommendation_trends.go
new file mode 100644
index 00000000..d9e8922c
--- /dev/null
+++ b/internal/api/admin_recommendation_trends.go
@@ -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
+}
diff --git a/internal/api/admin_recommendation_trends_test.go b/internal/api/admin_recommendation_trends_test.go
new file mode 100644
index 00000000..278f9cdb
--- /dev/null
+++ b/internal/api/admin_recommendation_trends_test.go
@@ -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)
+	}
+}
diff --git a/internal/api/admin_recommendation_tuning.go b/internal/api/admin_recommendation_tuning.go
new file mode 100644
index 00000000..1a556d9d
--- /dev/null
+++ b/internal/api/admin_recommendation_tuning.go
@@ -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))
+	}
+}
diff --git a/internal/api/admin_recommendation_tuning_test.go b/internal/api/admin_recommendation_tuning_test.go
new file mode 100644
index 00000000..4c6bba88
--- /dev/null
+++ b/internal/api/admin_recommendation_tuning_test.go
@@ -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)
+		}
+	}
+}
diff --git a/internal/api/api.go b/internal/api/api.go
index 5a7e4bfe..a8649fb7 100644
--- a/internal/api/api.go
+++ b/internal/api/api.go
@@ -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
diff --git a/internal/api/auth_test.go b/internal/api/auth_test.go
index 333256bf..ae564b9c 100644
--- a/internal/api/auth_test.go
+++ b/internal/api/auth_test.go
@@ -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
 }
 
diff --git a/internal/api/events.go b/internal/api/events.go
index c7e49f16..150f534f 100644
--- a/internal/api/events.go
+++ b/internal/api/events.go
@@ -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 {
diff --git a/internal/api/library_test.go b/internal/api/library_test.go
index 09d36cf7..370da98e 100644
--- a/internal/api/library_test.go
+++ b/internal/api/library_test.go
@@ -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",
diff --git a/internal/api/me_recommendation_metrics.go b/internal/api/me_recommendation_metrics.go
index 05a15a75..75761b05 100644
--- a/internal/api/me_recommendation_metrics.go
+++ b/internal/api/me_recommendation_metrics.go
@@ -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"`
+	WindowDays int `json:"window_days"`
+	// 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)
 	}
-	writeJSON(w, http.StatusOK, out)
+
+	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)
+	}
+	return resp
 }
 
 // parseMetricsDays reads the `days` query param (default 30, capped at 365).
diff --git a/internal/api/me_recommendation_metrics_test.go b/internal/api/me_recommendation_metrics_test.go
index 2a1af4c3..de717b79 100644
--- a/internal/api/me_recommendation_metrics_test.go
+++ b/internal/api/me_recommendation_metrics_test.go
@@ -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)
 	}
 }
diff --git a/internal/api/radio.go b/internal/api/radio.go
index c99a26f2..5aa68f6c 100644
--- a/internal/api/radio.go
+++ b/internal/api/radio.go
@@ -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)
diff --git a/internal/config/config.go b/internal/config/config.go
index dfd42189..dd2e4ac8 100644
--- a/internal/config/config.go
+++ b/internal/config/config.go
@@ -88,20 +88,14 @@ 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"`
+	RecentlyPlayedHours int `yaml:"recently_played_hours"`
+	RadioSize           int `yaml:"radio_size"`
+	RadioSizeMax        int `yaml:"radio_size_max"`
 }
 
 func Default() Config {
@@ -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,
diff --git a/internal/db/dbq/events.sql.go b/internal/db/dbq/events.sql.go
index 59f08d04..6db1800c 100644
--- a/internal/db/dbq/events.sql.go
+++ b/internal/db/dbq/events.sql.go
@@ -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
 }
diff --git a/internal/db/dbq/models.go b/internal/db/dbq/models.go
index bdf64e72..29528972 100644
--- a/internal/db/dbq/models.go
+++ b/internal/db/dbq/models.go
@@ -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
diff --git a/internal/db/dbq/playlists.sql.go b/internal/db/dbq/playlists.sql.go
index 71825f80..e4a9855e 100644
--- a/internal/db/dbq/playlists.sql.go
+++ b/internal/db/dbq/playlists.sql.go
@@ -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,
diff --git a/internal/db/dbq/recommendation_metrics.sql.go b/internal/db/dbq/recommendation_metrics.sql.go
index db1f6e77..daee78a4 100644
--- a/internal/db/dbq/recommendation_metrics.sql.go
+++ b/internal/db/dbq/recommendation_metrics.sql.go
@@ -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
+}
diff --git a/internal/db/dbq/recommendation_tuning.sql.go b/internal/db/dbq/recommendation_tuning.sql.go
new file mode 100644
index 00000000..d41c0f19
--- /dev/null
+++ b/internal/db/dbq/recommendation_tuning.sql.go
@@ -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
+}
diff --git a/internal/db/dbq/system_mixes.sql.go b/internal/db/dbq/system_mixes.sql.go
index 3ea0e5b9..ce98bd19 100644
--- a/internal/db/dbq/system_mixes.sql.go
+++ b/internal/db/dbq/system_mixes.sql.go
@@ -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
-      FROM tracks t
-      JOIN stats s ON s.track_id = t.id
-     WHERE s.c >= 5
-       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
-       )
 )
-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
+SELECT t.id, t.album_id, t.artist_id
+  FROM tracks t
+  JOIN stats s ON s.track_id = t.id
+ 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'
+   )
+ 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 {
diff --git a/internal/db/dbq/system_playlists.sql.go b/internal/db/dbq/system_playlists.sql.go
index aa1c41cb..4f8d1a86 100644
--- a/internal/db/dbq/system_playlists.sql.go
+++ b/internal/db/dbq/system_playlists.sql.go
@@ -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 {
diff --git a/internal/db/migrations/0038_for_you_pick_kind.down.sql b/internal/db/migrations/0038_for_you_pick_kind.down.sql
new file mode 100644
index 00000000..7158bd2f
--- /dev/null
+++ b/internal/db/migrations/0038_for_you_pick_kind.down.sql
@@ -0,0 +1,2 @@
+ALTER TABLE play_events DROP COLUMN pick_kind;
+ALTER TABLE playlist_tracks DROP COLUMN pick_kind;
diff --git a/internal/db/migrations/0038_for_you_pick_kind.up.sql b/internal/db/migrations/0038_for_you_pick_kind.up.sql
new file mode 100644
index 00000000..fdf86a39
--- /dev/null
+++ b/internal/db/migrations/0038_for_you_pick_kind.up.sql
@@ -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'));
diff --git a/internal/db/migrations/0039_pick_kind_vocabulary.down.sql b/internal/db/migrations/0039_pick_kind_vocabulary.down.sql
new file mode 100644
index 00000000..eb447d38
--- /dev/null
+++ b/internal/db/migrations/0039_pick_kind_vocabulary.down.sql
@@ -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'));
diff --git a/internal/db/migrations/0039_pick_kind_vocabulary.up.sql b/internal/db/migrations/0039_pick_kind_vocabulary.up.sql
new file mode 100644
index 00000000..36f619d2
--- /dev/null
+++ b/internal/db/migrations/0039_pick_kind_vocabulary.up.sql
@@ -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'
+        ));
diff --git a/internal/db/migrations/0040_recommendation_tuning.down.sql b/internal/db/migrations/0040_recommendation_tuning.down.sql
new file mode 100644
index 00000000..9575d0eb
--- /dev/null
+++ b/internal/db/migrations/0040_recommendation_tuning.down.sql
@@ -0,0 +1,3 @@
+DROP TABLE recommendation_tuning_audit;
+DROP TABLE taste_tuning;
+DROP TABLE recommendation_weight_profiles;
diff --git a/internal/db/migrations/0040_recommendation_tuning.up.sql b/internal/db/migrations/0040_recommendation_tuning.up.sql
new file mode 100644
index 00000000..7aba7c30
--- /dev/null
+++ b/internal/db/migrations/0040_recommendation_tuning.up.sql
@@ -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);
diff --git a/internal/db/queries/events.sql b/internal/db/queries/events.sql
index db0d4829..565b510c 100644
--- a/internal/db/queries/events.sql
+++ b/internal/db/queries/events.sql
@@ -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
diff --git a/internal/db/queries/playlists.sql b/internal/db/queries/playlists.sql
index 349e1940..058a1ffa 100644
--- a/internal/db/queries/playlists.sql
+++ b/internal/db/queries/playlists.sql
@@ -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
diff --git a/internal/db/queries/recommendation_metrics.sql b/internal/db/queries/recommendation_metrics.sql
index c7d04159..48d66565 100644
--- a/internal/db/queries/recommendation_metrics.sql
+++ b/internal/db/queries/recommendation_metrics.sql
@@ -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;
diff --git a/internal/db/queries/recommendation_tuning.sql b/internal/db/queries/recommendation_tuning.sql
new file mode 100644
index 00000000..760ea716
--- /dev/null
+++ b/internal/db/queries/recommendation_tuning.sql
@@ -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;
diff --git a/internal/db/queries/system_mixes.sql b/internal/db/queries/system_mixes.sql
index b483fad0..7503a122 100644
--- a/internal/db/queries/system_mixes.sql
+++ b/internal/db/queries/system_mixes.sql
@@ -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
@@ -37,76 +50,113 @@ 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
-      FROM tracks t
-      JOIN stats s ON s.track_id = t.id
-     WHERE s.c >= 5
-       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
-       )
-)
-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
- 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
-    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
-     GROUP BY t.artist_id
-    HAVING COUNT(*) >= 3
 )
 SELECT t.id, t.album_id, t.artist_id
   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'
+  JOIN stats s ON s.track_id = t.id
+ 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
    )
- ORDER BY al.created_at DESC, t.disc_number NULLS FIRST, t.track_number NULLS FIRST
+   -- 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'
+   )
+ 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, 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 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, alt.tier::int AS tier
+  FROM tracks t
+  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 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
 -- (±10 day-of-year) in the past, excluding the very recent (>30 days
@@ -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
diff --git a/internal/db/queries/system_playlists.sql b/internal/db/queries/system_playlists.sql
index 305c8b75..4f8a7696 100644
--- a/internal/db/queries/system_playlists.sql
+++ b/internal/db/queries/system_playlists.sql
@@ -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
diff --git a/internal/dbtest/reset.go b/internal/dbtest/reset.go
index eba016c9..b8c1465a 100644
--- a/internal/dbtest/reset.go
+++ b/internal/dbtest/reset.go
@@ -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",
diff --git a/internal/playevents/writer.go b/internal/playevents/writer.go
index a65ba169..22999945 100644
--- a/internal/playevents/writer.go
+++ b/internal/playevents/writer.go
@@ -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
diff --git a/internal/playevents/writer_test.go b/internal/playevents/writer_test.go
index a0158c11..b6dc1bd6 100644
--- a/internal/playevents/writer_test.go
+++ b/internal/playevents/writer_test.go
@@ -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)
+	}
+}
diff --git a/internal/playlists/discover.go b/internal/playlists/discover.go
index 65f41d48..809e33fb 100644
--- a/internal/playlists/discover.go
+++ b/internal/playlists/discover.go
@@ -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
 }
diff --git a/internal/playlists/discover_test.go b/internal/playlists/discover_test.go
index e299cb11..fc1d54a8 100644
--- a/internal/playlists/discover_test.go
+++ b/internal/playlists/discover_test.go
@@ -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)
+		}
+	}
+}
diff --git a/internal/playlists/foryou_test.go b/internal/playlists/foryou_test.go
index 0d69a7ae..2a102f9e 100644
--- a/internal/playlists/foryou_test.go
+++ b/internal/playlists/foryou_test.go
@@ -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")
+		}
+	}
+}
diff --git a/internal/playlists/scheduler.go b/internal/playlists/scheduler.go
index 97ab098f..c68d2da7 100644
--- a/internal/playlists/scheduler.go
+++ b/internal/playlists/scheduler.go
@@ -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)
 	}
diff --git a/internal/playlists/seed_selection_test.go b/internal/playlists/seed_selection_test.go
index c3e5ec43..e3399f3b 100644
--- a/internal/playlists/seed_selection_test.go
+++ b/internal/playlists/seed_selection_test.go
@@ -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)
 	}
 }
 
diff --git a/internal/playlists/system.go b/internal/playlists/system.go
index b618d9d7..5103c14f 100644
--- a/internal/playlists/system.go
+++ b/internal/playlists/system.go
@@ -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
 	}
@@ -118,41 +104,137 @@ 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
+	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{
-	BaseWeight:       1.0,
-	LikeBoost:        2.0,
-	RecencyWeight:    1.0,
-	SkipPenalty:      2.0,
-	JitterMagnitude:  0.1,
-	ContextWeight:    0.5,
-	SimilarityWeight: 1.5,
-	// Taste profile (#796 phase 2): the daily mixes are the primary
-	// taste-driven surface, so they lean on it. TasteMatchScore is in
-	// [-1,+1], so 1.5 makes a strong taste fit comparable to a like boost
-	// while passive avoidance (negative) gently demotes.
-	TasteWeight: 1.5,
+//
+// 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,
+		SkipPenalty:      2.0,
+		JitterMagnitude:  0.1,
+		ContextWeight:    0.5,
+		SimilarityWeight: 1.5,
+		// Taste profile (#796 phase 2): the daily mixes are the primary
+		// taste-driven surface, so they lean on it. TasteMatchScore is in
+		// [-1,+1], so 1.5 makes a strong taste fit comparable to a like boost
+		// 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
@@ -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}
-	cands, cerr := recommendation.LoadCandidatesFromSimilarity(
-		ctx, q, userID, forYouSeed,
-		1, // recentlyPlayedHours — small to avoid filtering the seed's recent neighbourhood
-		zeroVec,
-		[]pgtype.UUID{forYouSeed},
-		systemForYouSourceLimits(),
-	)
-	if cerr != nil {
-		logger.Warn("system playlist: for-you candidates load failed; skipping",
-			"user_id", uuidStringPL(userID), "err", cerr)
+	// 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, seed,
+			1, // recentlyPlayedHours — small to avoid filtering the seed's recent neighbourhood
+			zeroVec,
+			seeds,
+			systemForYouSourceLimits(),
+		)
+		if cerr != nil {
+			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
@@ -624,58 +852,76 @@ 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)
 		}
 		out := make([]rankedCandidate, total)
 		for i := 0; i < total; i++ {
 			out[i] = rankedCandidate{
-				TrackID: capped[i].Track.ID,
-				Score:   recommendation.Score(capped[i].Inputs, systemMixWeights, now, rng.Float64),
+				TrackID:  capped[i].Track.ID,
+				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),
+			TrackID:  c.Track.ID,
+			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.
diff --git a/internal/playlists/system_mixes.go b/internal/playlists/system_mixes.go
index 4bd862e4..2ca6ef78 100644
--- a/internal/playlists/system_mixes.go
+++ b/internal/playlists/system_mixes.go
@@ -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
 }
diff --git a/internal/playlists/system_mixes_db_test.go b/internal/playlists/system_mixes_db_test.go
new file mode 100644
index 00000000..a2a65c09
--- /dev/null
+++ b/internal/playlists/system_mixes_db_test.go
@@ -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")
+	}
+}
diff --git a/internal/playlists/system_mixes_test.go b/internal/playlists/system_mixes_test.go
new file mode 100644
index 00000000..64e863d0
--- /dev/null
+++ b/internal/playlists/system_mixes_test.go
@@ -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)
+		}
+	}
+}
diff --git a/internal/playlists/system_test.go b/internal/playlists/system_test.go
index 2e5fb5cf..e0757758 100644
--- a/internal/playlists/system_test.go
+++ b/internal/playlists/system_test.go
@@ -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()
diff --git a/internal/playlists/you_might_like.go b/internal/playlists/you_might_like.go
index 3d40e919..27ccaf58 100644
--- a/internal/playlists/you_might_like.go
+++ b/internal/playlists/you_might_like.go
@@ -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
diff --git a/internal/recsettings/patch.go b/internal/recsettings/patch.go
new file mode 100644
index 00000000..e3c78ab6
--- /dev/null
+++ b/internal/recsettings/patch.go
@@ -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
+}
diff --git a/internal/recsettings/service.go b/internal/recsettings/service.go
new file mode 100644
index 00000000..4a1a0b0b
--- /dev/null
+++ b/internal/recsettings/service.go
@@ -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,
+	}
+}
diff --git a/internal/recsettings/service_test.go b/internal/recsettings/service_test.go
new file mode 100644
index 00000000..65019ff3
--- /dev/null
+++ b/internal/recsettings/service_test.go
@@ -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))
+	}
+}
diff --git a/internal/server/server.go b/internal/server/server.go
index 248340c1..05b3cb65 100644
--- a/internal/server/server.go
+++ b/internal/server/server.go
@@ -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
diff --git a/web/src/lib/api/metrics.ts b/web/src/lib/api/metrics.ts
index be1fb87d..3f467d30 100644
--- a/web/src/lib/api/metrics.ts
+++ b/web/src/lib/api/metrics.ts
@@ -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;
-}
diff --git a/web/src/lib/api/tuning.ts b/web/src/lib/api/tuning.ts
new file mode 100644
index 00000000..8ffd0218
--- /dev/null
+++ b/web/src/lib/api/tuning.ts
@@ -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}`);
+}
diff --git a/web/src/lib/components/AdminTabs.svelte b/web/src/lib/components/AdminTabs.svelte
index f7acbabd..08a21264 100644
--- a/web/src/lib/components/AdminTabs.svelte
+++ b/web/src/lib/components/AdminTabs.svelte
@@ -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'           }
   ];
 
diff --git a/web/src/lib/components/AdminTabs.test.ts b/web/src/lib/components/AdminTabs.test.ts
index 2163c178..171b4d3e 100644
--- a/web/src/lib/components/AdminTabs.test.ts
+++ b/web/src/lib/components/AdminTabs.test.ts
@@ -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'
     ]);
   });
diff --git a/web/src/routes/admin/tuning/+page.svelte b/web/src/routes/admin/tuning/+page.svelte
new file mode 100644
index 00000000..ef7a9d7a
--- /dev/null
+++ b/web/src/routes/admin/tuning/+page.svelte
@@ -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 (0–30d 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>
diff --git a/web/src/routes/admin/tuning/tuning.test.ts b/web/src/routes/admin/tuning/tuning.test.ts
new file mode 100644
index 00000000..d953d164
--- /dev/null
+++ b/web/src/routes/admin/tuning/tuning.test.ts
@@ -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()
+    );
+  });
+});
diff --git a/web/src/routes/settings/+page.svelte b/web/src/routes/settings/+page.svelte
index eef29ee9..3c6c0d23 100644
--- a/web/src/routes/settings/+page.svelte
+++ b/web/src/routes/settings/+page.svelte
@@ -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);
 
@@ -278,37 +319,131 @@
     <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}
-      <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 text-right font-medium">Plays</th>
-            <th class="py-1 text-right font-medium">Skip rate</th>
-            <th class="py-1 text-right font-medium">Avg completion</th>
-          </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>
-              <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>
-            </tr>
-          {/each}
-        </tbody>
-      </table>
+    {: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">
+                <th class="py-1 font-medium">Surface</th>
+                <th class="py-1 text-right font-medium">Plays</th>
+                <th class="py-1 text-right font-medium">Skip rate</th>
+                <th class="py-1 text-right font-medium">Avg completion</th>
+              </tr>
+            </thead>
+            <tbody>
+              {#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">
+                    {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>
diff --git a/web/src/routes/settings/Appearance.test.ts b/web/src/routes/settings/Appearance.test.ts
index d254e97b..67407e35 100644
--- a/web/src/routes/settings/Appearance.test.ts
+++ b/web/src/routes/settings/Appearance.test.ts
@@ -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(() => {
diff --git a/web/src/routes/settings/settings.test.ts b/web/src/routes/settings/settings.test.ts
index ae913c07..fcd21fe5 100644
--- a/web/src/routes/settings/settings.test.ts
+++ b/web/src/routes/settings/settings.test.ts
@@ -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();