Dzarlax-AI · dzarlax · Jun 15, 2026 · Jun 15, 2026
diff --git a/internal/adminapi/adminapi_test.go b/internal/adminapi/adminapi_test.go
@@ -418,12 +418,21 @@ func TestTGAdminRecommendedModelsAppendFreeCandidatesUnrecommended(t *testing.T)
 	for _, model := range payload.Models {
 		if model.Recommended && !model.Free {
 			sawRecommended = true
+			if model.PrimaryReason != "Pareto frontier for simple" {
+				t.Fatalf("recommended primary reason = %q, want role-specific Pareto explanation: %+v", model.PrimaryReason, model)
+			}
+			if strings.EqualFold(model.PolicyLabel, "recommended") {
+				t.Fatalf("recommended policy label should be suppressed, got %+v", model)
+			}
 		}
 		if model.Free {
 			sawFreeCandidate = true
 			if model.Recommended || model.Policy != "free_unverified" || model.Source != "free" {
 				t.Fatalf("unexpected free candidate flags: %+v", model)
 			}
+			if model.StatusLabel != "Untested" || model.PolicyLabel != "Free unverified" {
+				t.Fatalf("unexpected free candidate display labels: %+v", model)
+			}
 		}
 	}
 	if !sawRecommended || !sawFreeCandidate {

diff --git a/internal/adminapi/handlers.go b/internal/adminapi/handlers.go
@@ -31,33 +31,37 @@ type uiSlot struct {
 }
 
 type uiModel struct {
-	ID              string
-	PromptPrice     float64
-	CompletionPrice float64
-	ContextLength   int
-	Vision          bool
-	Tools           bool
-	Reasoning       bool
-	Free            bool
-	Score           float64 // AA Intelligence Index
-	CodingIndex     float64 // AA Coding Index
-	AgenticIndex    float64 // AA Agentic Index
-	SpeedTPS        float64 // median output tokens/sec
-	TTFT            float64 // median time-to-first-token, seconds
-	ThinkTime       float64 // TTFA - TTFT — time spent thinking before answer starts (reasoners only)
-	AAPriceBlended  float64 // AA's reference blended 3:1 input/output price (USD / 1M)
-	MarkupPct       float64 // (OR blended - AA blended) / AA blended × 100 — positive means OR charges more
-	EffectivePrompt float64 // 0.9 × prompt + 0.1 × multimodal_slot.prompt for non-vision candidates under roles that route images elsewhere
-	ValuePerDollar  float64 // quality / prompt price (role-specific in preset path; agent/$ in browse path)
-	Recommended     bool
-	Source          string
-	Policy          string
-	Section         string
-	OverrideNote    string
-	Reasons         []string
-	Warnings        []string
-	Telemetry       modelTelemetry
-	Market          llm.OpenRouterMarketSignal
+	ID               string
+	PromptPrice      float64
+	CompletionPrice  float64
+	ContextLength    int
+	Vision           bool
+	Tools            bool
+	Reasoning        bool
+	Free             bool
+	Score            float64 // AA Intelligence Index
+	CodingIndex      float64 // AA Coding Index
+	AgenticIndex     float64 // AA Agentic Index
+	SpeedTPS         float64 // median output tokens/sec
+	TTFT             float64 // median time-to-first-token, seconds
+	ThinkTime        float64 // TTFA - TTFT — time spent thinking before answer starts (reasoners only)
+	AAPriceBlended   float64 // AA's reference blended 3:1 input/output price (USD / 1M)
+	MarkupPct        float64 // (OR blended - AA blended) / AA blended × 100 — positive means OR charges more
+	EffectivePrompt  float64 // 0.9 × prompt + 0.1 × multimodal_slot.prompt for non-vision candidates under roles that route images elsewhere
+	ValuePerDollar   float64 // quality / prompt price (role-specific in preset path; agent/$ in browse path)
+	Recommended      bool
+	Source           string
+	Policy           string
+	Section          string
+	OverrideNote     string
+	Reasons          []string
+	Warnings         []string
+	Telemetry        modelTelemetry
+	Market           llm.OpenRouterMarketSignal
+	StatusLabel      string
+	PolicyLabel      string
+	PrimaryReason    string
+	SecondaryReasons []string
 }
 
 type modelTelemetry struct {
@@ -405,6 +409,7 @@ func (s *Server) buildIndexData(r *http.Request) indexData {
 		models = filterModelOverrides(models, catalogProv, overrides, true)
 		attachMarketSignals(models, marketSignals)
 		attachModelTelemetry(models, s.loadModelTelemetry(ctx5, catalogProv))
+		decorateModelDisplay(models, preset)
 		sections := buildModelSections(models, true)
 		filters := uiFilters{
 			ActivePreset:      preset,
@@ -516,6 +521,7 @@ func (s *Server) buildIndexData(r *http.Request) indexData {
 	models = filterModelOverrides(models, catalogProv, overrides, f.Search == "")
 	attachMarketSignals(models, marketSignals)
 	attachModelTelemetry(models, s.loadModelTelemetry(ctx5, catalogProv))
+	decorateModelDisplay(models, "")
 	asc := sortDir == "asc"
 	sort.Slice(models, func(i, j int) bool {
 		var less bool

diff --git a/internal/adminapi/model_display.go b/internal/adminapi/model_display.go
@@ -0,0 +1,127 @@
+package adminapi
+
+import "strings"
+
+type modelDisplay struct {
+	StatusLabel      string
+	PolicyLabel      string
+	PrimaryReason    string
+	SecondaryReasons []string
+}
+
+func decorateModelDisplay(models []uiModel, role string) {
+	for i := range models {
+		applyModelDisplay(&models[i], role, false)
+	}
+}
+
+func applyModelDisplay(m *uiModel, role string, current bool) {
+	d := modelDisplayFor(*m, role, current)
+	m.StatusLabel = d.StatusLabel
+	m.PolicyLabel = d.PolicyLabel
+	m.PrimaryReason = d.PrimaryReason
+	m.SecondaryReasons = d.SecondaryReasons
+}
+
+func modelDisplayFor(m uiModel, role string, current bool) modelDisplay {
+	status := modelStatusLabel(m, current)
+	policy := modelPolicyLabel(m, status)
+	primary := modelPrimaryReason(m, role, status)
+	secondary := modelSecondaryReasons(m.Reasons, primary)
+	return modelDisplay{
+		StatusLabel:      status,
+		PolicyLabel:      policy,
+		PrimaryReason:    primary,
+		SecondaryReasons: secondary,
+	}
+}
+
+func modelStatusLabel(m uiModel, current bool) string {
+	if current {
+		return "Current"
+	}
+	switch {
+	case m.Policy == "manual_deny" || m.Policy == "free_blocked":
+		return "Blocked"
+	case m.Section == "recommended" || m.Recommended:
+		return "Recommended"
+	case m.Section == "untested" || m.Source == "untested" || m.Policy == "free_unverified":
+		return "Untested"
+	case m.Section == "interesting" || m.Source == "near_frontier" || m.Source == "manual" || m.Policy == "manual_allow":
+		return "Interesting"
+	}
+	return "Candidate"
+}
+
+func modelPolicyLabel(m uiModel, status string) string {
+	switch m.Policy {
+	case "", "candidate", "recommended":
+		return ""
+	case "manual_allow":
+		return "Manual allow"
+	case "manual_deny":
+		return "Manual deny"
+	case "free_unverified":
+		return "Free unverified"
+	case "free_verified":
+		return "Free verified"
+	case "free_degraded":
+		return "Free degraded"
+	case "free_blocked":
+		return "Free blocked"
+	default:
+		label := strings.ReplaceAll(m.Policy, "_", " ")
+		return strings.ToUpper(label[:1]) + label[1:]
+	}
+}
+
+func modelPrimaryReason(m uiModel, role, status string) string {
+	switch {
+	case m.Recommended:
+		if role != "" {
+			return "Pareto frontier for " + role
+		}
+		return "Pareto frontier"
+	case m.Policy == "manual_allow":
+		return firstNonEmpty("Manual allow", m.OverrideNote)
+	case m.Policy == "manual_deny":
+		return firstNonEmpty("Manual deny", m.OverrideNote)
+	case m.Policy == "free_blocked":
+		return "Free endpoint is blocked"
+	case m.Policy == "free_degraded":
+		return "Free endpoint degraded"
+	case m.Policy == "free_verified":
+		return "Free endpoint verified"
+	case m.Policy == "free_unverified":
+		return "Free endpoint needs a check"
+	case m.Source == "near_frontier":
+		if role != "" {
+			return "Near frontier for " + role
+		}
+		return "Near recommendation frontier"
+	case status == "Untested":
+		return "Compatible but missing role evidence"
+	}
+	if len(m.Reasons) > 0 {
+		return m.Reasons[0]
+	}
+	return ""
+}
+
+func modelSecondaryReasons(reasons []string, primary string) []string {
+	out := make([]string, 0, len(reasons))
+	for _, reason := range uniqueStrings(reasons) {
+		if reason == "" || reason == primary {
+			continue
+		}
+		out = append(out, reason)
+	}
+	return out
+}
+
+func firstNonEmpty(fallback, value string) string {
+	if strings.TrimSpace(value) != "" {
+		return fallback + ": " + strings.TrimSpace(value)
+	}
+	return fallback
+}
diff --git a/internal/adminapi/recommend_test.go b/internal/adminapi/recommend_test.go
@@ -226,6 +226,50 @@ func TestModelSectionsSplitRecommendationBuckets(t *testing.T) {
 	}
 }
 
+func TestModelDisplayDedupesRecommendedInternals(t *testing.T) {
+	m := uiModel{
+		ID:          "deepseek/deepseek-v3.2",
+		Recommended: true,
+		Section:     "recommended",
+		Policy:      "recommended",
+		Reasons:     []string{"tool calling", "AA coding 35", "ctx 128k"},
+	}
+
+	got := modelDisplayFor(m, "default", false)
+	if got.StatusLabel != "Recommended" {
+		t.Fatalf("status = %q, want Recommended", got.StatusLabel)
+	}
+	if got.PolicyLabel != "" {
+		t.Fatalf("policy label = %q, want empty duplicate-suppressed label", got.PolicyLabel)
+	}
+	if got.PrimaryReason != "Pareto frontier for default" {
+		t.Fatalf("primary reason = %q", got.PrimaryReason)
+	}
+	if containsString(got.SecondaryReasons, "recommended") || containsString(got.SecondaryReasons, "Pareto frontier for default") {
+		t.Fatalf("secondary reasons contain duplicated recommendation internals: %+v", got.SecondaryReasons)
+	}
+}
+
+func TestModelDisplayKeepsCatalogCandidateStatus(t *testing.T) {
+	m := uiModel{
+		ID:      "qwen/qwen3.5-plus",
+		Source:  "catalog",
+		Policy:  "candidate",
+		Reasons: []string{"tool calling", "AA intelligence 90"},
+	}
+
+	got := modelDisplayFor(m, "default", false)
+	if got.StatusLabel != "Candidate" {
+		t.Fatalf("status = %q, want Candidate for plain catalog result", got.StatusLabel)
+	}
+	if got.PolicyLabel != "" {
+		t.Fatalf("policy label = %q, want empty candidate label", got.PolicyLabel)
+	}
+	if got.PrimaryReason != "tool calling" {
+		t.Fatalf("primary reason = %q, want first catalog reason", got.PrimaryReason)
+	}
+}
+
 func TestSectionDiversityCapsModelFamilies(t *testing.T) {
 	models := []uiModel{
 		{ID: "google/gemini-2.5-pro"},
@@ -247,6 +291,15 @@ func TestSectionDiversityCapsModelFamilies(t *testing.T) {
 	}
 }
 
+func containsString(values []string, want string) bool {
+	for _, value := range values {
+		if value == want {
+			return true
+		}
+	}
+	return false
+}
+
 func TestDefaultPresetUsesCodingWhenAgenticIsMissing(t *testing.T) {
 	caps := map[string]llm.Capabilities{
 		"x-ai/grok-4.3": {

diff --git a/internal/adminapi/templates/partials_models_row.html b/internal/adminapi/templates/partials_models_row.html
@@ -33,9 +33,10 @@
     <td>
       <div class="model-id">{{.ID}}</div>
       <div style="margin-top:4px;color:var(--text-secondary);font-size:0.78rem;">
-        {{if .Section}}<span class="badge badge--neutral">{{.Section}}</span>{{end}}
-        {{if .Policy}}<span class="badge badge--neutral">{{.Policy}}</span>{{end}}
-        {{range .Reasons}}<span>{{.}}</span> {{end}}
+        {{if .StatusLabel}}<span class="badge badge--neutral">{{.StatusLabel}}</span>{{end}}
+        {{if .PolicyLabel}}<span class="badge badge--neutral">{{.PolicyLabel}}</span>{{end}}
+        {{if .PrimaryReason}}<span>Why: {{.PrimaryReason}}</span>{{end}}
+        {{range .SecondaryReasons}}<span>{{.}}</span> {{end}}
       </div>
       {{if .Warnings}}
       <div style="margin-top:4px;color:var(--warning, #9a5b00);font-size:0.78rem;">

diff --git a/internal/adminapi/templates/tg_admin.html b/internal/adminapi/templates/tg_admin.html
@@ -586,17 +586,17 @@ <h2 id="sheet-title">Choose model</h2>
           var chips = document.createElement("div");
           chips.className = "chips";
           [
-            model.current ? "current" : "",
-            model.recommended ? "recommended" : "",
-            model.section || "",
-            model.policy || "",
+            model.status_label || (model.current ? "Current" : ""),
+            model.policy_label || "",
             model.tools ? "tools" : "",
             model.vision ? "vision" : "",
             model.reasoning ? "reasoning" : "",
             model.market && model.market.rank ? "market #" + model.market.rank : "",
             model.market && !model.market.rank && (model.market.share || model.market.score) ? "market" : "",
             model.agentic_index ? "agentic " + Math.round(model.agentic_index) : ""
-          ].filter(Boolean).forEach(function (v) {
+          ].filter(Boolean).filter(function (v, i, arr) {
+            return arr.indexOf(v) === i;
+          }).forEach(function (v) {
             var chip = document.createElement("span");
             chip.className = "chip";
             chip.textContent = v;
@@ -605,10 +605,15 @@ <h2 id="sheet-title">Choose model</h2>
           body.appendChild(id);
           body.appendChild(meta);
           body.appendChild(chips);
-          if (model.reasons && model.reasons.length) {
+          if (model.primary_reason || (model.secondary_reasons && model.secondary_reasons.length)) {
             var reasons = document.createElement("div");
             reasons.className = "model-card__note";
-            reasons.textContent = model.reasons.slice(0, 4).join(" · ");
+            var why = [];
+            if (model.primary_reason) why.push("Why: " + model.primary_reason);
+            (model.secondary_reasons || []).slice(0, 3).forEach(function (reason) {
+              why.push(reason);
+            });
+            reasons.textContent = why.join(" · ");
             body.appendChild(reasons);
           }
           if (model.warnings && model.warnings.length) {