fix(ops): show first-token latency in TTFT request details

backend/internal/repository/ops_repo_request_details.go

@@ -93,6 +93,7 @@ WITH combined AS (
     COALESCE(NULLIF(g.platform, ''), NULLIF(a.platform, ''), '') AS platform,
     ul.model AS model,
     ul.duration_ms AS duration_ms,
+    ul.first_token_ms AS first_token_ms,
     NULL::INT AS status_code,
     NULL::BIGINT AS error_id,
     NULL::TEXT AS phase,
@@ -117,6 +118,7 @@ WITH combined AS (
     COALESCE(NULLIF(o.platform, ''), NULLIF(g.platform, ''), NULLIF(a.platform, ''), '') AS platform,
     o.model AS model,
     o.duration_ms AS duration_ms,
+    o.time_to_first_token_ms AS first_token_ms,
     o.status_code AS status_code,
     o.id AS error_id,
     o.error_phase AS phase,
@@ -152,6 +154,8 @@ WITH combined AS (
 			// default
 		case "duration_desc":
 			sort = "ORDER BY duration_ms DESC NULLS LAST, created_at DESC"
+		case "ttft_desc":
+			sort = "ORDER BY first_token_ms DESC NULLS LAST, created_at DESC"
 		default:
 			return nil, 0, fmt.Errorf("invalid sort")
 		}
@@ -166,6 +170,7 @@ SELECT
   platform,
   model,
   duration_ms,
+  first_token_ms,
   status_code,
   error_id,
   phase,
@@ -213,9 +218,10 @@ LIMIT $%d OFFSET $%d
 			platform  sql.NullString
 			model     sql.NullString
 
-			durationMs sql.NullInt64
-			statusCode sql.NullInt64
-			errorID    sql.NullInt64
+			durationMs   sql.NullInt64
+			firstTokenMs sql.NullInt64
+			statusCode   sql.NullInt64
+			errorID      sql.NullInt64
 
 			phase    sql.NullString
 			severity sql.NullString
@@ -236,6 +242,7 @@ LIMIT $%d OFFSET $%d
 			&platform,
 			&model,
 			&durationMs,
+			&firstTokenMs,
 			&statusCode,
 			&errorID,
 			&phase,
@@ -257,12 +264,13 @@ LIMIT $%d OFFSET $%d
 			Platform:  strings.TrimSpace(platform.String),
 			Model:     strings.TrimSpace(model.String),
 
-			DurationMs: toIntPtr(durationMs),
-			StatusCode: toIntPtr(statusCode),
-			ErrorID:    toInt64Ptr(errorID),
-			Phase:      phase.String,
-			Severity:   severity.String,
-			Message:    message.String,
+			DurationMs:   toIntPtr(durationMs),
+			FirstTokenMs: toIntPtr(firstTokenMs),
+			StatusCode:   toIntPtr(statusCode),
+			ErrorID:      toInt64Ptr(errorID),
+			Phase:        phase.String,
+			Severity:     severity.String,
+			Message:      message.String,
 
 			UserID:    toInt64Ptr(userID),
 			APIKeyID:  toInt64Ptr(apiKeyID),

backend/internal/repository/ops_repo_request_details_test.go

@@ -0,0 +1,63 @@
+package repository
+
+import (
+	"context"
+	"testing"
+	"time"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	"github.com/Wei-Shaw/sub2api/internal/service"
+	"github.com/stretchr/testify/require"
+)
+
+func TestOpsRepositoryListRequestDetails_LatencySort(t *testing.T) {
+	for _, tc := range []struct {
+		name  string
+		sort  string
+		order string
+	}{
+		{name: "TTFT", sort: "ttft_desc", order: "first_token_ms DESC NULLS LAST, created_at DESC"},
+		{name: "duration", sort: "duration_desc", order: "duration_ms DESC NULLS LAST, created_at DESC"},
+		{name: "default", order: "created_at DESC"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			db, mock := newSQLMock(t)
+			repo := &opsRepository{db: db}
+			start := time.Date(2026, 9, 10, 0, 0, 0, 0, time.UTC)
+			end := start.Add(time.Hour)
+			filter := &service.OpsRequestDetailFilter{
+				StartTime: &start,
+				EndTime:   &end,
+				Sort:      tc.sort,
+				Page:      2,
+				PageSize:  10,
+			}
+
+			mock.ExpectQuery(`SELECT COUNT\(1\) FROM combined`).
+				WithArgs(start, end).
+				WillReturnRows(sqlmock.NewRows([]string{"count"}).AddRow(13))
+			rows := sqlmock.NewRows([]string{
+				"kind", "created_at", "request_id", "platform", "model", "duration_ms", "first_token_ms",
+				"status_code", "error_id", "phase", "severity", "message", "user_id", "api_key_id", "account_id", "group_id", "stream",
+			}).
+				AddRow("success", start, "req-slow", "openai", "gpt-5.5", 12000, 800, nil, nil, nil, nil, nil, 1, 2, 3, 4, true).
+				AddRow("error", start, "req-zero", "openai", "gpt-5.5", 9000, 0, 502, 5, "upstream", "error", "failed", 1, 2, 3, 4, true).
+				AddRow("success", start, "req-missing", "openai", "gpt-5.5", 5000, nil, nil, nil, nil, nil, nil, 1, 2, 3, 4, false)
+			mock.ExpectQuery(`(?s)ul\.first_token_ms AS first_token_ms.*o\.time_to_first_token_ms AS first_token_ms.*SELECT.*duration_ms,\s+first_token_ms,.*ORDER BY ` + tc.order + `\s+LIMIT \$3 OFFSET \$4`).
+				WithArgs(start, end, 10, 10).
+				WillReturnRows(rows)
+
+			items, total, err := repo.ListRequestDetails(context.Background(), filter)
+			require.NoError(t, err)
+			require.EqualValues(t, 13, total)
+			require.Len(t, items, 3)
+			require.NotNil(t, items[0].FirstTokenMs)
+			require.Equal(t, 800, *items[0].FirstTokenMs)
+			require.Equal(t, 12000, *items[0].DurationMs)
+			require.NotNil(t, items[1].FirstTokenMs)
+			require.Zero(t, *items[1].FirstTokenMs)
+			require.Nil(t, items[2].FirstTokenMs)
+			require.NoError(t, mock.ExpectationsWereMet())
+		})
+	}
+}

backend/internal/service/ops_request_details.go

@@ -22,8 +22,9 @@ type OpsRequestDetail struct {
 	Platform string `json:"platform,omitempty"`
 	Model    string `json:"model,omitempty"`
 
-	DurationMs *int `json:"duration_ms,omitempty"`
-	StatusCode *int `json:"status_code,omitempty"`
+	DurationMs   *int `json:"duration_ms,omitempty"`
+	FirstTokenMs *int `json:"first_token_ms,omitempty"`
+	StatusCode   *int `json:"status_code,omitempty"`
 
 	// When Kind == "error", ErrorID links to /admin/ops/errors/:id.
 	ErrorID *int64 `json:"error_id,omitempty"`
@@ -61,7 +62,7 @@ type OpsRequestDetailFilter struct {
 	MinDurationMs *int
 	MaxDurationMs *int
 
-	// Sort: created_at_desc (default) or duration_desc.
+	// Sort: created_at_desc (default), duration_desc or ttft_desc.
 	Sort string
 
 	Page     int