diff --git a/README.md b/README.md
index ec3c5e0..666e831 100644
--- a/README.md
+++ b/README.md
@@ -40,11 +40,13 @@
 | WIP 三頁 Vue 3 遷移（Overview/Detail/Hold Detail） | ✅ 已完成 |
 | 設備雙頁 Vue 3 遷移（Status/History） | ✅ 已完成 |
 | 設備快取 DataFrame TTL 一致性修復 | ✅ 已完成 |
+| 中段製程不良追溯分析（TMTT → 上游） | ✅ 已完成 |
 
 ---
 
 ## 開發歷史（Vite 重構後）
 
+- 2026-02-10：完成中段製程不良追溯分析（`/mid-section-defect`）— TMTT 測試站不良回溯至上游機台/站點/製程。三段式資料管線（TMTT 偵測 → SPLITFROMID BFS + COMBINEDASSYLOTS 合批展開 → 上游製程歷史），支援 205 種不良原因篩選、6 張 Pareto 圖表、日趨勢、LOT 明細分頁（200 筆/頁）。Loss reasons 24h Redis 快取、分析結果 5 分鐘快取、Detail API 分離（summary ~16KB + detail ~110KB/page，原 72MB 單次回應）。
 - 2026-02-09：完成設備雙頁 Vue 3 遷移（`/resource`、`/resource-history`）— 兩頁共 1,697 行 vanilla JS + 3,200 行 Jinja2 模板重寫為 Vue 3 SFC。抽取 `resource-shared/` 共用模組（CSS 基底、E10 狀態常數、HierarchyTable 三層展開樹表元件），History 頁 4 個 ECharts 圖表改用 vue-echarts，Status 頁複用 `useAutoRefresh` composable（5 分鐘自動刷新）。
 - 2026-02-09：完成 WIP 三頁 Vue 3 遷移（`/wip-overview`、`/wip-detail`、`/hold-detail`）— 三頁共 1,941 行 vanilla JS + Jinja2 重寫為 Vue 3 SFC。抽取共用 CSS/常數/元件至 `wip-shared/`，Pareto 圖改用 vue-echarts（與 QC-GATE 一致），Hold Detail 新增前端 URL params 判斷取代 Jinja2 注入。
 - 2026-02-09：完成數據表查詢頁面（`/tables`）Vue 3 遷移 — 第二個純 Vite 頁面，建立 `apiPost` POST 請求模式，237 行 vanilla JS 重寫為 Vue 3 SFC 元件。
@@ -460,7 +462,7 @@ A: 請確認瀏覽器允許下載檔案，並檢查查詢結果是否有資料
 
 透過側邊欄抽屜分組導覽切換各功能模組：
 - **報表類**：WIP 即時概況、設備即時概況、設備歷史績效、QC-GATE 即時狀態
-- **查詢類**：設備維修查詢、批次追蹤工具、TMTT 不良分析
+- **查詢類**：設備維修查詢、批次追蹤工具、TMTT 不良分析、中段製程不良追溯
 - **開發工具**（admin only）：數據表查詢、Excel 批次查詢、頁面管理、效能監控
 - 抽屜/頁面配置可由管理員動態管理（新增、重排、刪除）
 
@@ -523,6 +525,21 @@ A: 請確認瀏覽器允許下載檔案，並檢查查詢結果是否有資料
 - 站點排序依 DW_MES_SPEC_WORKCENTER_V 製程順序
 - **技術架構**：第一個純 Vue 3 + Vite 頁面，完全脫離 Jinja2
 
+### 中段製程不良追溯分析
+
+- TMTT 測試站不良回溯至上游機台 / 站點 / 製程的反向追蹤分析
+- 三段式資料管線：
+  1. TMTT 偵測（LOTWIPHISTORY + LOTREJECTHISTORY）
+  2. LOT 族譜解析（CONTAINER.SPLITFROMID BFS 分批鏈 + PJ_COMBINEDASSYLOTS 合批展開）
+  3. 上游製程歷史（LOTWIPHISTORY by ancestor CIDs）
+- 6 張 KPI 卡片 + 6 張 Pareto 圖表（依站點/不良原因/上游機台/TMTT 機台/製程/封裝歸因）
+- 日趨勢折線圖 + LOT 明細分頁表（200 筆/頁，伺服器端 DEFECT_RATE 降序排序）
+- 不良原因多選篩選（205 種，24h Redis 快取）
+- 分析結果 5 分鐘 Redis 快取；summary API (~16 KB) 與 detail API (~110 KB/page) 分離
+- CSV 串流匯出（UTF-8 BOM，完整明細）
+- 5 分鐘自動刷新 + visibilitychange 即時刷新
+- **技術架構**：Vue 3 + Vite，Pareto/趨勢圖使用 vue-echarts，複用 `wip-shared/` 的 Pagination/useAutoRefresh
+
 ### 數據表查詢工具
 
 - 顯示所有 DWH 表格的分類卡片目錄（即時數據表/現況快照表/歷史累積表/輔助表）
@@ -588,8 +605,8 @@ CIRCUIT_BREAKER_RECOVERY_TIMEOUT=30
 | 技術 | 用途 |
 |------|------|
 | Jinja2 | 模板引擎（既有頁面） |
-| Vue 3 | UI 框架（QC-GATE、Tables、WIP 三頁、設備雙頁已遷移，漸進式擴展中） |
-| vue-echarts | ECharts Vue 封裝（QC-GATE、WIP Overview Pareto、Resource History 4 圖表） |
+| Vue 3 | UI 框架（QC-GATE、Tables、WIP 三頁、設備雙頁、中段不良追溯已遷移，漸進式擴展中） |
+| vue-echarts | ECharts Vue 封裝（QC-GATE、WIP Overview Pareto、Resource History 4 圖表、Mid-Section Defect 7 圖表） |
 | Vite 6 | 前端多頁模組打包（含 Vue SFC + HTML entry） |
 | ECharts | 圖表庫（npm tree-shaking + 舊版靜態檔案並存） |
 | Vanilla JS Modules | 互動功能與頁面邏輯（既有頁面） |
@@ -640,7 +657,8 @@ DashBoard_vite/
 │   │   ├── dashboard/          # 儀表板查詢
 │   │   ├── resource/           # 設備查詢
 │   │   ├── wip/                # WIP 查詢
-│   │   └── resource_history/   # 設備歷史查詢
+│   │   ├── resource_history/   # 設備歷史查詢
+│   │   └── mid_section_defect/ # 中段不良追溯查詢
 │   └── templates/              # HTML 模板
 ├── frontend/                   # Vite 前端專案
 │   ├── src/core/               # 共用 API/欄位契約/計算 helper
@@ -655,7 +673,8 @@ DashBoard_vite/
 │   ├── src/wip-shared/         # WIP 三頁共用 CSS/常數/元件
 │   ├── src/wip-overview/       # WIP 即時概況 (Vue 3 SFC)
 │   ├── src/wip-detail/         # WIP 明細查詢 (Vue 3 SFC)
-│   └── src/hold-detail/        # Hold 狀態分析 (Vue 3 SFC)
+│   ├── src/hold-detail/        # Hold 狀態分析 (Vue 3 SFC)
+│   └── src/mid-section-defect/ # 中段不良追溯分析 (Vue 3 SFC)
 ├── shared/
 │   └── field_contracts.json    # 前後端共用欄位契約
 ├── scripts/                    # 腳本
@@ -739,6 +758,22 @@ conda run -n mes-dashboard python scripts/run_cache_benchmarks.py --enforce
 
 ## 變更日誌
 
+### 2026-02-10
+
+- 新增中段製程不良追溯分析頁面（`/mid-section-defect`）：
+  - TMTT 測試站偵測到的不良，反向追蹤至上游機台/站點/製程的歸因分析
+  - 三段式資料管線：TMTT 偵測 → LOT 族譜解析（SPLITFROMID BFS 分批鏈 + COMBINEDASSYLOTS 合批展開）→ 上游製程歷史
+  - 6 張 KPI 卡片（投入/LOT數/不良數/不良率/首要原因/影響機台數）
+  - 6 張 Pareto 圖表（站點/不良原因/上游機台/TMTT 機台/製程/封裝歸因）+ 日趨勢折線
+  - 不良原因多選篩選（205 種，全站 24h Redis 快取，`/api/mid-section-defect/loss-reasons`）
+  - Detail API 分頁分離（`/api/mid-section-defect/analysis` summary ~16 KB + `/api/mid-section-defect/analysis/detail` ~110 KB/page），原 72 MB 單次回應
+  - 伺服器端 DEFECT_RATE 降序排序 + 前端頁內欄位排序
+  - CSV 串流匯出（UTF-8 BOM，完整明細）
+  - 進入頁面不自動查詢，點擊「查詢」後才執行；首次查詢後啟動 5 分鐘自動刷新
+  - Summary + Detail page 1 平行載入（`Promise.all`）
+  - NaN 安全防護（Oracle NULL → pandas NaN，`isinstance(val, str)` 過濾）
+  - 技術架構：Vue 3 + Vite，vue-echarts 7 圖表，複用 `wip-shared/` Pagination/useAutoRefresh
+
 ### 2026-02-09
 
 - 完成設備雙頁 Vue 3 遷移（`/resource`、`/resource-history`）：
@@ -879,5 +914,5 @@ conda run -n mes-dashboard python scripts/run_cache_benchmarks.py --enforce
 
 ---
 
-**文檔版本**: 5.3
-**最後更新**: 2026-02-09
+**文檔版本**: 5.4
+**最後更新**: 2026-02-10
diff --git a/data/page_status.json b/data/page_status.json
index 646a03b..3cdbcd5 100644
--- a/data/page_status.json
+++ b/data/page_status.json
@@ -78,6 +78,13 @@
       "drawer_id": "queries",
       "order": 5
     },
+    {
+      "route": "/mid-section-defect",
+      "name": "中段製程不良追溯",
+      "status": "dev",
+      "drawer_id": "queries",
+      "order": 6
+    },
     {
       "route": "/admin/pages",
       "name": "頁面管理",
diff --git a/frontend/package.json b/frontend/package.json
index 6484262..41d2356 100644
--- a/frontend/package.json
+++ b/frontend/package.json
@@ -5,7 +5,7 @@
   "type": "module",
   "scripts": {
     "dev": "vite --host",
-    "build": "vite build && cp ../src/mes_dashboard/static/dist/src/tables/index.html ../src/mes_dashboard/static/dist/tables.html && cp ../src/mes_dashboard/static/dist/src/qc-gate/index.html ../src/mes_dashboard/static/dist/qc-gate.html && cp ../src/mes_dashboard/static/dist/src/wip-overview/index.html ../src/mes_dashboard/static/dist/wip-overview.html && cp ../src/mes_dashboard/static/dist/src/wip-detail/index.html ../src/mes_dashboard/static/dist/wip-detail.html && cp ../src/mes_dashboard/static/dist/src/hold-detail/index.html ../src/mes_dashboard/static/dist/hold-detail.html && cp ../src/mes_dashboard/static/dist/src/resource-status/index.html ../src/mes_dashboard/static/dist/resource-status.html && cp ../src/mes_dashboard/static/dist/src/resource-history/index.html ../src/mes_dashboard/static/dist/resource-history.html",
+    "build": "vite build && cp ../src/mes_dashboard/static/dist/src/tables/index.html ../src/mes_dashboard/static/dist/tables.html && cp ../src/mes_dashboard/static/dist/src/qc-gate/index.html ../src/mes_dashboard/static/dist/qc-gate.html && cp ../src/mes_dashboard/static/dist/src/wip-overview/index.html ../src/mes_dashboard/static/dist/wip-overview.html && cp ../src/mes_dashboard/static/dist/src/wip-detail/index.html ../src/mes_dashboard/static/dist/wip-detail.html && cp ../src/mes_dashboard/static/dist/src/hold-detail/index.html ../src/mes_dashboard/static/dist/hold-detail.html && cp ../src/mes_dashboard/static/dist/src/resource-status/index.html ../src/mes_dashboard/static/dist/resource-status.html && cp ../src/mes_dashboard/static/dist/src/resource-history/index.html ../src/mes_dashboard/static/dist/resource-history.html && cp ../src/mes_dashboard/static/dist/src/mid-section-defect/index.html ../src/mes_dashboard/static/dist/mid-section-defect.html",
     "test": "node --test tests/*.test.js"
   },
   "devDependencies": {
diff --git a/frontend/src/mid-section-defect/App.vue b/frontend/src/mid-section-defect/App.vue
new file mode 100644
index 0000000..becd8f9
--- /dev/null
+++ b/frontend/src/mid-section-defect/App.vue
@@ -0,0 +1,263 @@
+<script setup>
+import { reactive, ref } from 'vue';
+
+import { apiGet, ensureMesApiAvailable } from '../core/api.js';
+import { useAutoRefresh } from '../wip-shared/composables/useAutoRefresh.js';
+
+import FilterBar from './components/FilterBar.vue';
+import KpiCards from './components/KpiCards.vue';
+import ParetoChart from './components/ParetoChart.vue';
+import TrendChart from './components/TrendChart.vue';
+import DetailTable from './components/DetailTable.vue';
+
+ensureMesApiAvailable();
+
+const API_TIMEOUT = 120000; // 2min (genealogy can be slow)
+const PAGE_SIZE = 200;
+
+const filters = reactive({
+  startDate: '',
+  endDate: '',
+  lossReasons: [],
+});
+
+const availableLossReasons = ref([]);
+
+const analysisData = ref({
+  kpi: {},
+  charts: {},
+  daily_trend: [],
+  genealogy_status: 'ready',
+  detail_total_count: 0,
+});
+
+const detailData = ref([]);
+const detailPagination = ref({
+  page: 1,
+  page_size: PAGE_SIZE,
+  total_count: 0,
+  total_pages: 1,
+});
+const detailLoading = ref(false);
+
+const loading = reactive({
+  initial: false,
+  querying: false,
+});
+
+const hasQueried = ref(false);
+const queryError = ref('');
+
+function setDefaultDates() {
+  const today = new Date();
+  const endDate = new Date(today);
+  endDate.setDate(endDate.getDate() - 1);
+
+  const startDate = new Date(endDate);
+  startDate.setDate(startDate.getDate() - 6);
+
+  filters.startDate = toDateString(startDate);
+  filters.endDate = toDateString(endDate);
+}
+
+function toDateString(value) {
+  return value.toISOString().slice(0, 10);
+}
+
+function unwrapApiResult(result, fallbackMessage) {
+  if (result?.success === true) {
+    return result;
+  }
+  throw new Error(result?.error || fallbackMessage);
+}
+
+function buildFilterParams() {
+  const params = {
+    start_date: filters.startDate,
+    end_date: filters.endDate,
+  };
+  if (filters.lossReasons.length) {
+    params.loss_reasons = filters.lossReasons.join(',');
+  }
+  return params;
+}
+
+async function loadLossReasons() {
+  try {
+    const result = await apiGet('/api/mid-section-defect/loss-reasons');
+    const unwrapped = unwrapApiResult(result, '載入不良原因失敗');
+    availableLossReasons.value = unwrapped.data?.loss_reasons || [];
+  } catch {
+    // Non-blocking — dropdown will be empty until first query
+  }
+}
+
+async function loadDetail(page = 1) {
+  detailLoading.value = true;
+  try {
+    const params = {
+      ...buildFilterParams(),
+      page,
+      page_size: PAGE_SIZE,
+    };
+    const result = await apiGet('/api/mid-section-defect/analysis/detail', {
+      params,
+      timeout: API_TIMEOUT,
+    });
+    const unwrapped = unwrapApiResult(result, '載入明細失敗');
+    detailData.value = unwrapped.data?.detail || [];
+    detailPagination.value = unwrapped.data?.pagination || {
+      page: 1, page_size: PAGE_SIZE, total_count: 0, total_pages: 1,
+    };
+  } catch (err) {
+    console.error('Detail load failed:', err.message);
+    detailData.value = [];
+  } finally {
+    detailLoading.value = false;
+  }
+}
+
+async function loadAnalysis() {
+  queryError.value = '';
+  loading.querying = true;
+
+  try {
+    const params = buildFilterParams();
+
+    // Fire summary and detail page 1 in parallel
+    const [summaryResult] = await Promise.all([
+      apiGet('/api/mid-section-defect/analysis', {
+        params,
+        timeout: API_TIMEOUT,
+      }),
+      loadDetail(1),
+    ]);
+
+    const unwrapped = unwrapApiResult(summaryResult, '查詢失敗');
+    analysisData.value = unwrapped.data;
+    hasQueried.value = true;
+
+    // Start auto-refresh after first successful query
+    if (!autoRefreshStarted) {
+      autoRefreshStarted = true;
+      startAutoRefresh();
+    }
+  } catch (err) {
+    queryError.value = err.message || '查詢失敗，請稍後再試';
+  } finally {
+    loading.querying = false;
+  }
+}
+
+function handleUpdateFilters(updated) {
+  Object.assign(filters, updated);
+}
+
+function handleQuery() {
+  loadAnalysis();
+}
+
+function prevPage() {
+  if (detailPagination.value.page <= 1) return;
+  loadDetail(detailPagination.value.page - 1);
+}
+
+function nextPage() {
+  if (detailPagination.value.page >= detailPagination.value.total_pages) return;
+  loadDetail(detailPagination.value.page + 1);
+}
+
+function exportCsv() {
+  const params = new URLSearchParams({
+    start_date: filters.startDate,
+    end_date: filters.endDate,
+  });
+  if (filters.lossReasons.length) {
+    params.set('loss_reasons', filters.lossReasons.join(','));
+  }
+
+  const link = document.createElement('a');
+  link.href = `/api/mid-section-defect/export?${params}`;
+  link.download = `mid_section_defect_${filters.startDate}_to_${filters.endDate}.csv`;
+  document.body.appendChild(link);
+  link.click();
+  document.body.removeChild(link);
+}
+
+let autoRefreshStarted = false;
+const { startAutoRefresh } = useAutoRefresh({
+  onRefresh: () => loadAnalysis(),
+  intervalMs: 5 * 60 * 1000,
+  autoStart: false,
+  refreshOnVisible: true,
+});
+
+function initPage() {
+  setDefaultDates();
+  loadLossReasons();
+}
+
+void initPage();
+</script>
+
+<template>
+  <div class="page-container">
+    <header class="page-header">
+      <h1>中段製程不良追溯分析</h1>
+      <p class="header-desc">TMTT 測試站不良回溯至上游機台 / 站點 / 製程</p>
+    </header>
+
+    <FilterBar
+      :filters="filters"
+      :loading="loading.querying"
+      :available-loss-reasons="availableLossReasons"
+      @update-filters="handleUpdateFilters"
+      @query="handleQuery"
+    />
+
+    <div v-if="queryError" class="error-banner">{{ queryError }}</div>
+
+    <template v-if="hasQueried">
+      <div v-if="analysisData.genealogy_status === 'error'" class="warning-banner">
+        追溯分析未完成（genealogy 查詢失敗），圖表僅顯示 TMTT 站點數據。
+      </div>
+
+      <KpiCards :kpi="analysisData.kpi" :loading="loading.querying" />
+
+      <div class="charts-section">
+        <div class="charts-row">
+          <ParetoChart title="依站點歸因" :data="analysisData.charts?.by_station" />
+          <ParetoChart title="依不良原因" :data="analysisData.charts?.by_loss_reason" />
+        </div>
+        <div class="charts-row">
+          <ParetoChart title="依上游機台歸因" :data="analysisData.charts?.by_machine" />
+          <ParetoChart title="依 TMTT 機台" :data="analysisData.charts?.by_tmtt_machine" />
+        </div>
+        <div class="charts-row">
+          <ParetoChart title="依製程 (WORKFLOW)" :data="analysisData.charts?.by_workflow" />
+          <ParetoChart title="依封裝 (PACKAGE)" :data="analysisData.charts?.by_package" />
+        </div>
+        <div class="charts-row charts-row-full">
+          <TrendChart :data="analysisData.daily_trend" />
+        </div>
+      </div>
+
+      <DetailTable
+        :data="detailData"
+        :loading="detailLoading"
+        :pagination="detailPagination"
+        @export-csv="exportCsv"
+        @prev-page="prevPage"
+        @next-page="nextPage"
+      />
+    </template>
+
+    <div v-else-if="!loading.querying" class="empty-state">
+      <p>請選擇日期範圍與不良原因，點擊「查詢」開始分析。</p>
+    </div>
+
+    <div class="loading-overlay" :class="{ hidden: !loading.querying }">
+      <div class="loading-spinner"></div>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/mid-section-defect/components/DetailTable.vue b/frontend/src/mid-section-defect/components/DetailTable.vue
new file mode 100644
index 0000000..268a266
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/DetailTable.vue
@@ -0,0 +1,139 @@
+<script setup>
+import { computed, ref } from 'vue';
+
+import Pagination from '../../wip-shared/components/Pagination.vue';
+
+const props = defineProps({
+  data: {
+    type: Array,
+    default: () => [],
+  },
+  loading: {
+    type: Boolean,
+    default: false,
+  },
+  pagination: {
+    type: Object,
+    default: () => ({ page: 1, page_size: 200, total_count: 0, total_pages: 1 }),
+  },
+});
+
+const emit = defineEmits(['export-csv', 'prev-page', 'next-page']);
+
+const sortField = ref('DEFECT_RATE');
+const sortAsc = ref(false);
+
+const COLUMNS = [
+  { key: 'CONTAINERNAME', label: 'LOT ID', width: '140px' },
+  { key: 'PJ_TYPE', label: 'TYPE', width: '80px' },
+  { key: 'PRODUCTLINENAME', label: 'PACKAGE', width: '90px' },
+  { key: 'WORKFLOW', label: 'WORKFLOW', width: '100px' },
+  { key: 'TMTT_EQUIPMENTNAME', label: 'TMTT 設備', width: '110px' },
+  { key: 'INPUT_QTY', label: '投入數', width: '70px', numeric: true },
+  { key: 'LOSS_REASON', label: '不良原因', width: '130px' },
+  { key: 'DEFECT_QTY', label: '不良數', width: '70px', numeric: true },
+  { key: 'DEFECT_RATE', label: '不良率(%)', width: '90px', numeric: true },
+  { key: 'ANCESTOR_COUNT', label: '上游LOT數', width: '80px', numeric: true },
+  { key: 'UPSTREAM_MACHINES', label: '上游機台', width: '200px' },
+];
+
+const sortedData = computed(() => {
+  if (!props.data || !props.data.length) return [];
+  const field = sortField.value;
+  const asc = sortAsc.value;
+  return [...props.data].sort((a, b) => {
+    const va = a[field] ?? '';
+    const vb = b[field] ?? '';
+    if (typeof va === 'number' && typeof vb === 'number') {
+      return asc ? va - vb : vb - va;
+    }
+    const sa = String(va);
+    const sb = String(vb);
+    return asc ? sa.localeCompare(sb) : sb.localeCompare(sa);
+  });
+});
+
+const tableInfo = computed(() => {
+  const p = props.pagination;
+  const total = Number(p.total_count || 0);
+  if (total <= 0) return '暫無資料';
+  const page = Number(p.page || 1);
+  const pageSize = Number(p.page_size || 200);
+  const start = (page - 1) * pageSize + 1;
+  const end = Math.min(page * pageSize, total);
+  return `顯示 ${start} - ${end} 筆，共 ${total.toLocaleString()} 筆`;
+});
+
+function toggleSort(field) {
+  if (sortField.value === field) {
+    sortAsc.value = !sortAsc.value;
+  } else {
+    sortField.value = field;
+    sortAsc.value = false;
+  }
+}
+
+function sortIcon(field) {
+  if (sortField.value !== field) return '';
+  return sortAsc.value ? ' ▲' : ' ▼';
+}
+
+function formatCell(value, col) {
+  if (value == null || value === '') return '-';
+  if (col.key === 'DEFECT_RATE') return Number(value).toFixed(2);
+  if (col.numeric) return Number(value).toLocaleString();
+  return value;
+}
+</script>
+
+<template>
+  <section class="section-card">
+    <div class="section-inner">
+      <div class="detail-header">
+        <h3 class="section-title">LOT 明細</h3>
+        <div class="detail-actions">
+          <span class="detail-count">{{ tableInfo }}</span>
+          <button type="button" class="btn-sm" :disabled="loading" @click="$emit('export-csv')">
+            匯出 CSV
+          </button>
+        </div>
+      </div>
+
+      <div class="table-wrapper">
+        <table class="detail-table">
+          <thead>
+            <tr>
+              <th
+                v-for="col in COLUMNS"
+                :key="col.key"
+                :style="{ width: col.width }"
+                class="sortable"
+                @click="toggleSort(col.key)"
+              >
+                {{ col.label }}{{ sortIcon(col.key) }}
+              </th>
+            </tr>
+          </thead>
+          <tbody>
+            <tr v-for="(row, idx) in sortedData" :key="idx">
+              <td v-for="col in COLUMNS" :key="col.key" :class="{ numeric: col.numeric }">
+                {{ formatCell(row[col.key], col) }}
+              </td>
+            </tr>
+            <tr v-if="!sortedData.length">
+              <td :colspan="COLUMNS.length" class="empty-row">暫無資料</td>
+            </tr>
+          </tbody>
+        </table>
+      </div>
+
+      <Pagination
+        :visible="Number(pagination.total_pages || 1) > 1"
+        :page="Number(pagination.page || 1)"
+        :total-pages="Number(pagination.total_pages || 1)"
+        @prev="emit('prev-page')"
+        @next="emit('next-page')"
+      />
+    </div>
+  </section>
+</template>
diff --git a/frontend/src/mid-section-defect/components/FilterBar.vue b/frontend/src/mid-section-defect/components/FilterBar.vue
new file mode 100644
index 0000000..719bf28
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/FilterBar.vue
@@ -0,0 +1,77 @@
+<script setup>
+import MultiSelect from './MultiSelect.vue';
+
+const props = defineProps({
+  filters: {
+    type: Object,
+    required: true,
+  },
+  loading: {
+    type: Boolean,
+    default: false,
+  },
+  availableLossReasons: {
+    type: Array,
+    default: () => [],
+  },
+});
+
+const emit = defineEmits(['update-filters', 'query']);
+
+function updateFilters(patch) {
+  emit('update-filters', {
+    ...props.filters,
+    ...patch,
+  });
+}
+</script>
+
+<template>
+  <section class="section-card">
+    <div class="section-inner">
+      <div class="filter-row">
+        <div class="filter-field">
+          <label for="msd-start-date">開始</label>
+          <input
+            id="msd-start-date"
+            type="date"
+            :value="filters.startDate"
+            :disabled="loading"
+            @input="updateFilters({ startDate: $event.target.value })"
+          />
+        </div>
+
+        <div class="filter-field">
+          <label for="msd-end-date">結束</label>
+          <input
+            id="msd-end-date"
+            type="date"
+            :value="filters.endDate"
+            :disabled="loading"
+            @input="updateFilters({ endDate: $event.target.value })"
+          />
+        </div>
+
+        <div class="filter-field">
+          <label>不良原因</label>
+          <MultiSelect
+            :model-value="filters.lossReasons"
+            :options="availableLossReasons"
+            :disabled="loading"
+            placeholder="全部原因"
+            @update:model-value="updateFilters({ lossReasons: $event })"
+          />
+        </div>
+
+        <button
+          type="button"
+          class="btn btn-primary"
+          :disabled="loading"
+          @click="$emit('query')"
+        >
+          查詢
+        </button>
+      </div>
+    </div>
+  </section>
+</template>
diff --git a/frontend/src/mid-section-defect/components/KpiCards.vue b/frontend/src/mid-section-defect/components/KpiCards.vue
new file mode 100644
index 0000000..a813a49
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/KpiCards.vue
@@ -0,0 +1,83 @@
+<script setup>
+import { computed } from 'vue';
+
+const props = defineProps({
+  kpi: {
+    type: Object,
+    default: () => ({}),
+  },
+  loading: {
+    type: Boolean,
+    default: false,
+  },
+});
+
+const cards = computed(() => [
+  {
+    label: 'TMTT 投入數',
+    value: formatNumber(props.kpi.total_input),
+    unit: 'pcs',
+    color: '#3b82f6',
+  },
+  {
+    label: 'LOT 數量',
+    value: formatNumber(props.kpi.lot_count),
+    unit: 'lots',
+    color: '#6366f1',
+  },
+  {
+    label: '不良總數',
+    value: formatNumber(props.kpi.total_defect_qty),
+    unit: 'pcs',
+    color: '#ef4444',
+  },
+  {
+    label: '不良率',
+    value: formatRate(props.kpi.total_defect_rate),
+    unit: '%',
+    color: '#f59e0b',
+  },
+  {
+    label: '首要不良原因',
+    value: props.kpi.top_loss_reason || '-',
+    unit: '',
+    color: '#8b5cf6',
+    isText: true,
+  },
+  {
+    label: '上游關聯機台',
+    value: formatNumber(props.kpi.affected_machine_count),
+    unit: '台',
+    color: '#10b981',
+  },
+]);
+
+function formatNumber(v) {
+  if (v == null || v === 0) return '0';
+  return Number(v).toLocaleString();
+}
+
+function formatRate(v) {
+  if (v == null) return '0.00';
+  return Number(v).toFixed(2);
+}
+</script>
+
+<template>
+  <section class="kpi-section">
+    <div class="kpi-grid">
+      <div
+        v-for="(card, idx) in cards"
+        :key="idx"
+        class="kpi-card"
+        :style="{ borderTopColor: card.color }"
+      >
+        <div class="kpi-label">{{ card.label }}</div>
+        <div class="kpi-value" :class="{ 'kpi-text': card.isText }">
+          <span>{{ card.value }}</span>
+          <span v-if="card.unit && !card.isText" class="kpi-unit">{{ card.unit }}</span>
+        </div>
+      </div>
+    </div>
+  </section>
+</template>
diff --git a/frontend/src/mid-section-defect/components/MultiSelect.vue b/frontend/src/mid-section-defect/components/MultiSelect.vue
new file mode 100644
index 0000000..e89b23f
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/MultiSelect.vue
@@ -0,0 +1,154 @@
+<script setup>
+import { computed, onBeforeUnmount, onMounted, ref } from 'vue';
+
+const props = defineProps({
+  modelValue: {
+    type: Array,
+    default: () => [],
+  },
+  options: {
+    type: Array,
+    default: () => [],
+  },
+  placeholder: {
+    type: String,
+    default: '請選擇',
+  },
+  disabled: {
+    type: Boolean,
+    default: false,
+  },
+});
+
+const emit = defineEmits(['update:modelValue']);
+
+const rootRef = ref(null);
+const isOpen = ref(false);
+
+const normalizedOptions = computed(() => {
+  return props.options.map((option) => {
+    if (option && typeof option === 'object') {
+      const value = option.value ?? option.name ?? option.label ?? '';
+      const label = option.label ?? option.name ?? option.value ?? '';
+      return {
+        label: String(label),
+        value: String(value),
+      };
+    }
+
+    return {
+      label: String(option),
+      value: String(option),
+    };
+  });
+});
+
+const selectedSet = computed(() => new Set((props.modelValue || []).map((value) => String(value))));
+
+const selectedText = computed(() => {
+  if (!props.modelValue.length) {
+    return props.placeholder;
+  }
+
+  if (props.modelValue.length === 1) {
+    const found = normalizedOptions.value.find(
+      (option) => option.value === String(props.modelValue[0])
+    );
+    return found?.label || String(props.modelValue[0]);
+  }
+
+  return `已選 ${props.modelValue.length} 項`;
+});
+
+function closeDropdown() {
+  isOpen.value = false;
+}
+
+function toggleDropdown() {
+  if (props.disabled) {
+    return;
+  }
+  isOpen.value = !isOpen.value;
+}
+
+function isSelected(value) {
+  return selectedSet.value.has(String(value));
+}
+
+function toggleOption(value) {
+  const next = new Set(selectedSet.value);
+  const key = String(value);
+
+  if (next.has(key)) {
+    next.delete(key);
+  } else {
+    next.add(key);
+  }
+
+  emit('update:modelValue', [...next]);
+}
+
+function selectAll() {
+  emit(
+    'update:modelValue',
+    normalizedOptions.value.map((option) => option.value)
+  );
+}
+
+function clearAll() {
+  emit('update:modelValue', []);
+}
+
+function handleOutsideClick(event) {
+  if (!isOpen.value || !rootRef.value) {
+    return;
+  }
+
+  if (!rootRef.value.contains(event.target)) {
+    isOpen.value = false;
+  }
+}
+
+onMounted(() => {
+  document.addEventListener('click', handleOutsideClick, true);
+});
+
+onBeforeUnmount(() => {
+  document.removeEventListener('click', handleOutsideClick, true);
+});
+</script>
+
+<template>
+  <div ref="rootRef" class="multi-select">
+    <button
+      type="button"
+      class="multi-select-trigger"
+      :disabled="disabled"
+      @click="toggleDropdown"
+    >
+      <span class="multi-select-text">{{ selectedText }}</span>
+      <span class="multi-select-arrow">{{ isOpen ? '▲' : '▼' }}</span>
+    </button>
+
+    <div v-if="isOpen" class="multi-select-dropdown">
+      <div class="multi-select-options">
+        <button
+          v-for="option in normalizedOptions"
+          :key="option.value"
+          type="button"
+          class="multi-select-option"
+          @click="toggleOption(option.value)"
+        >
+          <input type="checkbox" :checked="isSelected(option.value)" tabindex="-1" />
+          <span>{{ option.label }}</span>
+        </button>
+      </div>
+
+      <div class="multi-select-actions">
+        <button type="button" class="btn-sm" @click="selectAll">全選</button>
+        <button type="button" class="btn-sm" @click="clearAll">清除</button>
+        <button type="button" class="btn-sm" @click="closeDropdown">關閉</button>
+      </div>
+    </div>
+  </div>
+</template>
diff --git a/frontend/src/mid-section-defect/components/ParetoChart.vue b/frontend/src/mid-section-defect/components/ParetoChart.vue
new file mode 100644
index 0000000..ec823ad
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/ParetoChart.vue
@@ -0,0 +1,132 @@
+<script setup>
+import { computed } from 'vue';
+import VChart from 'vue-echarts';
+import { use } from 'echarts/core';
+import { CanvasRenderer } from 'echarts/renderers';
+import { BarChart, LineChart } from 'echarts/charts';
+import {
+  GridComponent,
+  LegendComponent,
+  TooltipComponent,
+} from 'echarts/components';
+
+use([CanvasRenderer, BarChart, LineChart, GridComponent, LegendComponent, TooltipComponent]);
+
+const props = defineProps({
+  title: {
+    type: String,
+    default: '',
+  },
+  data: {
+    type: Array,
+    default: () => [],
+  },
+});
+
+const chartOption = computed(() => {
+  if (!props.data || !props.data.length) return null;
+
+  const names = props.data.map((d) => d.name);
+  const defectQty = props.data.map((d) => d.defect_qty);
+  const cumulativePct = props.data.map((d) => d.cumulative_pct);
+  const defectRate = props.data.map((d) => d.defect_rate);
+
+  return {
+    animationDuration: 350,
+    tooltip: {
+      trigger: 'axis',
+      axisPointer: { type: 'cross' },
+      formatter(params) {
+        const idx = params[0]?.dataIndex;
+        if (idx == null) return '';
+        const item = props.data[idx];
+        let html = `<b>${item.name}</b><br/>`;
+        html += `不良數: ${(item.defect_qty || 0).toLocaleString()}<br/>`;
+        html += `投入數: ${(item.input_qty || 0).toLocaleString()}<br/>`;
+        html += `不良率: ${(item.defect_rate || 0).toFixed(2)}%<br/>`;
+        html += `累計占比: ${(item.cumulative_pct || 0).toFixed(1)}%`;
+        return html;
+      },
+    },
+    legend: {
+      data: ['不良數', '不良率', '累計占比'],
+      bottom: 0,
+      textStyle: { fontSize: 11 },
+    },
+    grid: {
+      top: 40,
+      right: 60,
+      bottom: 50,
+      left: 50,
+      containLabel: true,
+    },
+    xAxis: {
+      type: 'category',
+      data: names,
+      axisLabel: {
+        rotate: names.length > 6 ? 30 : 0,
+        fontSize: 11,
+        overflow: 'truncate',
+        width: 80,
+      },
+    },
+    yAxis: [
+      {
+        type: 'value',
+        name: '不良數',
+        nameTextStyle: { fontSize: 11 },
+        axisLabel: { fontSize: 11 },
+      },
+      {
+        type: 'value',
+        name: '%',
+        max: 100,
+        nameTextStyle: { fontSize: 11 },
+        axisLabel: { fontSize: 11, formatter: '{value}%' },
+      },
+    ],
+    series: [
+      {
+        name: '不良數',
+        type: 'bar',
+        data: defectQty,
+        itemStyle: { color: '#6366f1', borderRadius: [3, 3, 0, 0] },
+        barMaxWidth: 40,
+      },
+      {
+        name: '不良率',
+        type: 'line',
+        yAxisIndex: 1,
+        data: defectRate,
+        symbol: 'circle',
+        symbolSize: 6,
+        lineStyle: { color: '#f59e0b', width: 2 },
+        itemStyle: { color: '#f59e0b' },
+      },
+      {
+        name: '累計占比',
+        type: 'line',
+        yAxisIndex: 1,
+        data: cumulativePct,
+        symbol: 'diamond',
+        symbolSize: 6,
+        lineStyle: { color: '#ef4444', width: 2, type: 'dashed' },
+        itemStyle: { color: '#ef4444' },
+      },
+    ],
+  };
+});
+</script>
+
+<template>
+  <div class="chart-card">
+    <h3 class="chart-title">{{ title }}</h3>
+    <VChart
+      v-if="chartOption"
+      class="chart-canvas"
+      :option="chartOption"
+      autoresize
+    />
+    <div v-else class="chart-empty">暫無資料</div>
+  </div>
+</template>
diff --git a/frontend/src/mid-section-defect/components/TrendChart.vue b/frontend/src/mid-section-defect/components/TrendChart.vue
new file mode 100644
index 0000000..fca31f9
--- /dev/null
+++ b/frontend/src/mid-section-defect/components/TrendChart.vue
@@ -0,0 +1,111 @@
+<script setup>
+import { computed } from 'vue';
+import VChart from 'vue-echarts';
+import { use } from 'echarts/core';
+import { CanvasRenderer } from 'echarts/renderers';
+import { BarChart, LineChart } from 'echarts/charts';
+import {
+  GridComponent,
+  LegendComponent,
+  TooltipComponent,
+} from 'echarts/components';
+
+use([CanvasRenderer, BarChart, LineChart, GridComponent, LegendComponent, TooltipComponent]);
+
+const props = defineProps({
+  data: {
+    type: Array,
+    default: () => [],
+  },
+});
+
+const chartOption = computed(() => {
+  if (!props.data || !props.data.length) return null;
+
+  const dates = props.data.map((d) => d.date);
+  const inputQty = props.data.map((d) => d.input_qty);
+  const defectQty = props.data.map((d) => d.defect_qty);
+  const defectRate = props.data.map((d) => d.defect_rate);
+
+  return {
+    animationDuration: 350,
+    tooltip: {
+      trigger: 'axis',
+      axisPointer: { type: 'cross' },
+    },
+    legend: {
+      data: ['投入數', '不良數', '不良率'],
+      bottom: 0,
+      textStyle: { fontSize: 11 },
+    },
+    grid: {
+      top: 40,
+      right: 60,
+      bottom: 50,
+      left: 50,
+      containLabel: true,
+    },
+    xAxis: {
+      type: 'category',
+      data: dates,
+      axisLabel: {
+        fontSize: 11,
+        rotate: dates.length > 14 ? 30 : 0,
+      },
+    },
+    yAxis: [
+      {
+        type: 'value',
+        name: '數量',
+        nameTextStyle: { fontSize: 11 },
+        axisLabel: { fontSize: 11 },
+      },
+      {
+        type: 'value',
+        name: '不良率 %',
+        nameTextStyle: { fontSize: 11 },
+        axisLabel: { fontSize: 11, formatter: '{value}%' },
+      },
+    ],
+    series: [
+      {
+        name: '投入數',
+        type: 'bar',
+        data: inputQty,
+        itemStyle: { color: '#93c5fd', borderRadius: [3, 3, 0, 0] },
+        barMaxWidth: 30,
+      },
+      {
+        name: '不良數',
+        type: 'bar',
+        data: defectQty,
+        itemStyle: { color: '#fca5a5', borderRadius: [3, 3, 0, 0] },
+        barMaxWidth: 30,
+      },
+      {
+        name: '不良率',
+        type: 'line',
+        yAxisIndex: 1,
+        data: defectRate,
+        symbol: 'circle',
+        symbolSize: 5,
+        lineStyle: { color: '#ef4444', width: 2 },
+        itemStyle: { color: '#ef4444' },
+      },
+    ],
+  };
+});
+</script>
+
+<template>
+  <div class="chart-card chart-card-full">
+    <h3 class="chart-title">每日不良趨勢</h3>
+    <VChart
+      v-if="chartOption"
+      class="chart-canvas chart-canvas-wide"
+      :option="chartOption"
+      autoresize
+    />
+    <div v-else class="chart-empty">暫無資料</div>
+  </div>
+</template>
diff --git a/frontend/src/mid-section-defect/index.html b/frontend/src/mid-section-defect/index.html
new file mode 100644
index 0000000..b69f4c1
--- /dev/null
+++ b/frontend/src/mid-section-defect/index.html
@@ -0,0 +1,12 @@
+<!doctype html>
+<html lang="zh-Hant">
+  <head>
+    <meta charset="UTF-8" />
+    <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+    <title>中段製程不良追溯分析</title>
+  </head>
+  <body>
+    <div id="app"></div>
+    <script type="module" src="./main.js"></script>
+  </body>
+</html>
diff --git a/frontend/src/mid-section-defect/main.js b/frontend/src/mid-section-defect/main.js
new file mode 100644
index 0000000..c56440b
--- /dev/null
+++ b/frontend/src/mid-section-defect/main.js
@@ -0,0 +1,6 @@
+import { createApp } from 'vue';
+
+import App from './App.vue';
+import './style.css';
+
+createApp(App).mount('#app');
diff --git a/frontend/src/mid-section-defect/style.css b/frontend/src/mid-section-defect/style.css
new file mode 100644
index 0000000..b1bbff0
--- /dev/null
+++ b/frontend/src/mid-section-defect/style.css
@@ -0,0 +1,498 @@
+:root {
+  --msd-bg: #f5f7fb;
+  --msd-card-bg: #ffffff;
+  --msd-text: #1f2937;
+  --msd-muted: #64748b;
+  --msd-border: #dbe4ef;
+  --msd-shadow: 0 1px 3px rgba(15, 23, 42, 0.08);
+  --msd-shadow-md: 0 8px 22px rgba(15, 23, 42, 0.1);
+  --msd-primary: #6366f1;
+  --msd-primary-dark: #4f46e5;
+}
+
+* {
+  box-sizing: border-box;
+}
+
+body {
+  margin: 0;
+  font-family: 'Microsoft JhengHei', 'PingFang TC', 'Noto Sans TC', sans-serif;
+  background: var(--msd-bg);
+  color: var(--msd-text);
+}
+
+.page-container {
+  min-height: 100vh;
+  padding: 16px;
+  max-width: 1800px;
+  margin: 0 auto;
+}
+
+/* ====== Header ====== */
+.page-header {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+  padding: 16px 20px;
+  border-radius: 12px;
+  margin-bottom: 16px;
+  background: linear-gradient(135deg, #6366f1 0%, #8b5cf6 50%, #a855f7 100%);
+  box-shadow: var(--msd-shadow-md);
+}
+
+.page-header h1 {
+  margin: 0;
+  color: #ffffff;
+  font-size: 24px;
+  letter-spacing: 0.2px;
+}
+
+.header-desc {
+  margin: 0;
+  color: rgba(255, 255, 255, 0.8);
+  font-size: 13px;
+}
+
+/* ====== Section Card ====== */
+.section-card {
+  background: var(--msd-card-bg);
+  border-radius: 12px;
+  box-shadow: var(--msd-shadow);
+  margin-bottom: 16px;
+}
+
+.section-inner {
+  padding: 16px 20px;
+}
+
+.section-title {
+  margin: 0 0 12px;
+  font-size: 16px;
+  font-weight: 700;
+}
+
+/* ====== Filter Bar ====== */
+.filter-row {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 12px;
+  align-items: center;
+}
+
+.filter-field {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.filter-field label {
+  font-size: 13px;
+  color: #475569;
+  white-space: nowrap;
+}
+
+.filter-field input[type='date'] {
+  border: 1px solid var(--msd-border);
+  border-radius: 6px;
+  padding: 8px 10px;
+  font-size: 13px;
+  background: #ffffff;
+  color: #1f2937;
+}
+
+/* ====== Buttons ====== */
+.btn {
+  border: none;
+  border-radius: 8px;
+  padding: 8px 20px;
+  font-size: 14px;
+  font-weight: 600;
+  cursor: pointer;
+  transition: background 0.2s;
+}
+
+.btn-primary {
+  background: var(--msd-primary);
+  color: #ffffff;
+}
+
+.btn-primary:hover:not(:disabled) {
+  background: var(--msd-primary-dark);
+}
+
+.btn:disabled {
+  cursor: not-allowed;
+  opacity: 0.6;
+}
+
+.btn-sm {
+  border: 1px solid var(--msd-border);
+  border-radius: 6px;
+  background: #ffffff;
+  color: #475569;
+  font-size: 12px;
+  padding: 4px 10px;
+  cursor: pointer;
+}
+
+.btn-sm:hover:not(:disabled) {
+  background: #f1f5f9;
+}
+
+/* ====== MultiSelect ====== */
+.multi-select {
+  position: relative;
+  min-width: 200px;
+}
+
+.multi-select-trigger {
+  width: 100%;
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 8px;
+  border: 1px solid var(--msd-border);
+  border-radius: 6px;
+  padding: 8px 10px;
+  font-size: 13px;
+  color: #1f2937;
+  background: #ffffff;
+  cursor: pointer;
+}
+
+.multi-select-trigger:disabled {
+  cursor: not-allowed;
+  opacity: 0.7;
+}
+
+.multi-select-text {
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+}
+
+.multi-select-arrow {
+  flex-shrink: 0;
+  font-size: 10px;
+  color: #94a3b8;
+}
+
+.multi-select-dropdown {
+  position: absolute;
+  top: 100%;
+  left: 0;
+  right: 0;
+  margin-top: 4px;
+  background: #ffffff;
+  border: 1px solid var(--msd-border);
+  border-radius: 8px;
+  box-shadow: var(--msd-shadow-md);
+  z-index: 100;
+  max-height: 300px;
+  display: flex;
+  flex-direction: column;
+}
+
+.multi-select-options {
+  overflow-y: auto;
+  flex: 1;
+  padding: 4px 0;
+}
+
+.multi-select-option {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+  width: 100%;
+  padding: 6px 12px;
+  border: none;
+  background: transparent;
+  cursor: pointer;
+  font-size: 13px;
+  text-align: left;
+}
+
+.multi-select-option:hover {
+  background: #f1f5f9;
+}
+
+.multi-select-option input[type='checkbox'] {
+  pointer-events: none;
+}
+
+.multi-select-actions {
+  display: flex;
+  gap: 6px;
+  padding: 8px 12px;
+  border-top: 1px solid var(--msd-border);
+  background: #f8fafc;
+  border-radius: 0 0 8px 8px;
+}
+
+/* ====== Error / Warning Banners ====== */
+.error-banner {
+  padding: 12px 16px;
+  background: #fef2f2;
+  border: 1px solid #fecaca;
+  border-radius: 8px;
+  color: #dc2626;
+  font-size: 14px;
+  margin-bottom: 16px;
+}
+
+.warning-banner {
+  padding: 12px 16px;
+  background: #fffbeb;
+  border: 1px solid #fde68a;
+  border-radius: 8px;
+  color: #b45309;
+  font-size: 14px;
+  margin-bottom: 16px;
+}
+
+/* ====== KPI Cards ====== */
+.kpi-section {
+  margin-bottom: 16px;
+}
+
+.kpi-grid {
+  display: grid;
+  grid-template-columns: repeat(auto-fit, minmax(180px, 1fr));
+  gap: 12px;
+}
+
+.kpi-card {
+  background: var(--msd-card-bg);
+  border-radius: 10px;
+  box-shadow: var(--msd-shadow);
+  padding: 16px;
+  border-top: 3px solid #6366f1;
+}
+
+.kpi-label {
+  font-size: 12px;
+  color: var(--msd-muted);
+  margin-bottom: 6px;
+}
+
+.kpi-value {
+  font-size: 24px;
+  font-weight: 700;
+  display: flex;
+  align-items: baseline;
+  gap: 4px;
+}
+
+.kpi-value.kpi-text {
+  font-size: 14px;
+  font-weight: 600;
+  word-break: break-all;
+}
+
+.kpi-unit {
+  font-size: 12px;
+  font-weight: 400;
+  color: var(--msd-muted);
+}
+
+/* ====== Charts ====== */
+.charts-section {
+  margin-bottom: 16px;
+}
+
+.charts-row {
+  display: grid;
+  grid-template-columns: 1fr 1fr;
+  gap: 12px;
+  margin-bottom: 12px;
+}
+
+.charts-row-full {
+  grid-template-columns: 1fr;
+}
+
+.chart-card {
+  background: var(--msd-card-bg);
+  border-radius: 10px;
+  box-shadow: var(--msd-shadow);
+  padding: 16px;
+}
+
+.chart-card-full {
+  grid-column: 1 / -1;
+}
+
+.chart-title {
+  margin: 0 0 8px;
+  font-size: 14px;
+  font-weight: 600;
+  color: var(--msd-text);
+}
+
+.chart-canvas {
+  width: 100%;
+  height: 320px;
+}
+
+.chart-canvas-wide {
+  height: 280px;
+}
+
+.chart-empty {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  height: 200px;
+  color: var(--msd-muted);
+  font-size: 14px;
+}
+
+/* ====== Detail Table ====== */
+.detail-header {
+  display: flex;
+  justify-content: space-between;
+  align-items: center;
+  margin-bottom: 12px;
+}
+
+.detail-actions {
+  display: flex;
+  align-items: center;
+  gap: 12px;
+}
+
+.detail-count {
+  font-size: 13px;
+  color: var(--msd-muted);
+}
+
+.table-wrapper {
+  overflow-x: auto;
+  border: 1px solid var(--msd-border);
+  border-radius: 8px;
+}
+
+.detail-table {
+  width: 100%;
+  border-collapse: collapse;
+  font-size: 13px;
+}
+
+.detail-table th {
+  background: #f1f5f9;
+  padding: 10px 12px;
+  text-align: left;
+  font-weight: 600;
+  font-size: 12px;
+  color: #475569;
+  white-space: nowrap;
+  border-bottom: 2px solid var(--msd-border);
+  position: sticky;
+  top: 0;
+}
+
+.detail-table th.sortable {
+  cursor: pointer;
+  user-select: none;
+}
+
+.detail-table th.sortable:hover {
+  background: #e2e8f0;
+}
+
+.detail-table td {
+  padding: 8px 12px;
+  border-bottom: 1px solid #f1f5f9;
+  white-space: nowrap;
+}
+
+.detail-table td.numeric {
+  text-align: right;
+  font-variant-numeric: tabular-nums;
+}
+
+.detail-table tbody tr:hover {
+  background: #f8fafc;
+}
+
+.empty-row {
+  text-align: center;
+  padding: 32px;
+  color: var(--msd-muted);
+}
+
+/* ====== Pagination ====== */
+.pagination {
+  display: flex;
+  justify-content: center;
+  align-items: center;
+  gap: 12px;
+  padding: 16px;
+  border-top: 1px solid var(--msd-border);
+}
+
+.pagination button {
+  padding: 8px 16px;
+  border: 1px solid var(--msd-border);
+  background: #fff;
+  border-radius: 6px;
+  cursor: pointer;
+  font-size: 13px;
+}
+
+.pagination button:hover:not(:disabled) {
+  border-color: var(--msd-primary);
+  color: var(--msd-primary);
+}
+
+.pagination button:disabled {
+  opacity: 0.5;
+  cursor: not-allowed;
+}
+
+.pagination .page-info {
+  font-size: 13px;
+  color: var(--msd-muted);
+}
+
+/* ====== Empty State ====== */
+.empty-state {
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  padding: 80px 20px;
+  color: var(--msd-muted);
+  font-size: 15px;
+}
+
+/* ====== Loading Overlay ====== */
+.loading-overlay {
+  position: fixed;
+  inset: 0;
+  display: flex;
+  align-items: center;
+  justify-content: center;
+  background: rgba(255, 255, 255, 0.7);
+  z-index: 999;
+  transition: opacity 0.2s;
+}
+
+.loading-overlay.hidden {
+  opacity: 0;
+  pointer-events: none;
+}
+
+.loading-spinner {
+  width: 40px;
+  height: 40px;
+  border: 3px solid #e2e8f0;
+  border-top-color: var(--msd-primary);
+  border-radius: 50%;
+  animation: spin 0.8s linear infinite;
+}
+
+@keyframes spin {
+  to {
+    transform: rotate(360deg);
+  }
+}
diff --git a/frontend/vite.config.js b/frontend/vite.config.js
index cd82046..e97f7e1 100644
--- a/frontend/vite.config.js
+++ b/frontend/vite.config.js
@@ -23,7 +23,8 @@ export default defineConfig(({ mode }) => ({
         tables: resolve(__dirname, 'src/tables/index.html'),
         'query-tool': resolve(__dirname, 'src/query-tool/main.js'),
         'tmtt-defect': resolve(__dirname, 'src/tmtt-defect/main.js'),
-        'qc-gate': resolve(__dirname, 'src/qc-gate/index.html')
+        'qc-gate': resolve(__dirname, 'src/qc-gate/index.html'),
+        'mid-section-defect': resolve(__dirname, 'src/mid-section-defect/index.html')
       },
       output: {
         entryFileNames: '[name].js',
diff --git a/src/mes_dashboard/app.py b/src/mes_dashboard/app.py
index 1b2a5f4..146b164 100644
--- a/src/mes_dashboard/app.py
+++ b/src/mes_dashboard/app.py
@@ -488,6 +488,12 @@ def create_app(config_name: str | None = None) -> Flask:
         dist_dir = os.path.join(app.static_folder or "", "dist")
         return send_from_directory(dist_dir, 'qc-gate.html')
 
+    @app.route('/mid-section-defect')
+    def mid_section_defect_page():
+        """Mid-section defect traceability analysis page (pure Vite)."""
+        dist_dir = os.path.join(app.static_folder or "", "dist")
+        return send_from_directory(dist_dir, 'mid-section-defect.html')
+
     # ========================================================
     # Table Query APIs (for table_data_viewer)
     # ========================================================
diff --git a/src/mes_dashboard/core/database.py b/src/mes_dashboard/core/database.py
index b119c02..bdbf279 100644
--- a/src/mes_dashboard/core/database.py
+++ b/src/mes_dashboard/core/database.py
@@ -559,6 +559,75 @@ def read_sql_df(sql: str, params: Optional[Dict[str, Any]] = None) -> pd.DataFra
         raise
 
 
+def read_sql_df_slow(
+    sql: str,
+    params: Optional[Dict[str, Any]] = None,
+    timeout_seconds: int = 120,
+) -> pd.DataFrame:
+    """Execute a slow SQL query with a custom timeout via direct oracledb connection.
+
+    Unlike read_sql_df which uses the pooled engine (55s timeout),
+    this creates a dedicated connection with a longer call_timeout
+    for known-slow queries (e.g. full table scans on large tables).
+
+    Args:
+        sql: SQL query string with Oracle bind variables.
+        params: Optional dict of parameter values.
+        timeout_seconds: Call timeout in seconds (default: 120).
+
+    Returns:
+        DataFrame with query results, or None on connection failure.
+    """
+    start_time = time.time()
+    timeout_ms = timeout_seconds * 1000
+
+    conn = None
+    try:
+        runtime = get_db_runtime_config()
+        conn = oracledb.connect(
+            **DB_CONFIG,
+            tcp_connect_timeout=runtime["tcp_connect_timeout"],
+            retry_count=runtime["retry_count"],
+            retry_delay=runtime["retry_delay"],
+        )
+        conn.call_timeout = timeout_ms
+        logger.debug(
+            "Slow-query connection established (call_timeout_ms=%s)", timeout_ms
+        )
+
+        cursor = conn.cursor()
+        cursor.execute(sql, params or {})
+        columns = [desc[0].upper() for desc in cursor.description]
+        rows = cursor.fetchall()
+        cursor.close()
+
+        df = pd.DataFrame(rows, columns=columns)
+
+        elapsed = time.time() - start_time
+        if elapsed > 1.0:
+            sql_preview = sql.strip().replace('\n', ' ')[:100]
+            logger.warning(f"Slow query ({elapsed:.2f}s): {sql_preview}...")
+        else:
+            logger.debug(f"Query completed in {elapsed:.3f}s, rows={len(df)}")
+
+        return df
+
+    except Exception as exc:
+        elapsed = time.time() - start_time
+        ora_code = _extract_ora_code(exc)
+        sql_preview = sql.strip().replace('\n', ' ')[:100]
+        logger.error(
+            f"Query failed after {elapsed:.2f}s - ORA-{ora_code}: {exc} | SQL: {sql_preview}..."
+        )
+        raise
+    finally:
+        if conn:
+            try:
+                conn.close()
+            except Exception:
+                pass
+
+
 # ============================================================
 # Table Utilities
 # ============================================================
diff --git a/src/mes_dashboard/routes/__init__.py b/src/mes_dashboard/routes/__init__.py
index 0ce5a98..fad3ec1 100644
--- a/src/mes_dashboard/routes/__init__.py
+++ b/src/mes_dashboard/routes/__init__.py
@@ -16,6 +16,7 @@ from .job_query_routes import job_query_bp
 from .query_tool_routes import query_tool_bp
 from .tmtt_defect_routes import tmtt_defect_bp
 from .qc_gate_routes import qc_gate_bp
+from .mid_section_defect_routes import mid_section_defect_bp
 
 
 def register_routes(app) -> None:
@@ -30,6 +31,7 @@ def register_routes(app) -> None:
     app.register_blueprint(query_tool_bp)
     app.register_blueprint(tmtt_defect_bp)
     app.register_blueprint(qc_gate_bp)
+    app.register_blueprint(mid_section_defect_bp)
 
 __all__ = [
     'wip_bp',
@@ -44,5 +46,6 @@ __all__ = [
     'query_tool_bp',
     'tmtt_defect_bp',
     'qc_gate_bp',
+    'mid_section_defect_bp',
     'register_routes',
 ]
diff --git a/src/mes_dashboard/routes/mid_section_defect_routes.py b/src/mes_dashboard/routes/mid_section_defect_routes.py
new file mode 100644
index 0000000..88e1366
--- /dev/null
+++ b/src/mes_dashboard/routes/mid_section_defect_routes.py
@@ -0,0 +1,160 @@
+# -*- coding: utf-8 -*-
+"""Mid-Section Defect Traceability Analysis API routes.
+
+Reverse traceability from TMTT (test) station back to upstream production stations.
+"""
+
+from flask import Blueprint, jsonify, request, Response
+
+from mes_dashboard.services.mid_section_defect_service import (
+    query_analysis,
+    query_analysis_detail,
+    query_all_loss_reasons,
+    export_csv,
+)
+
+mid_section_defect_bp = Blueprint(
+    'mid_section_defect',
+    __name__,
+    url_prefix='/api/mid-section-defect'
+)
+
+
+@mid_section_defect_bp.route('/analysis', methods=['GET'])
+def api_analysis():
+    """API: Get mid-section defect traceability analysis (summary).
+
+    Returns kpi, charts, daily_trend, available_loss_reasons, genealogy_status,
+    and detail_total_count.  Does NOT include the detail array — use
+    /analysis/detail for paginated detail data.
+
+    Query Parameters:
+        start_date: Start date (YYYY-MM-DD), required
+        end_date: End date (YYYY-MM-DD), required
+        loss_reasons: Comma-separated loss reason names, optional
+    """
+    start_date = request.args.get('start_date')
+    end_date = request.args.get('end_date')
+
+    if not start_date or not end_date:
+        return jsonify({
+            'success': False,
+            'error': '必須提供 start_date 和 end_date 參數'
+        }), 400
+
+    loss_reasons_str = request.args.get('loss_reasons', '')
+    loss_reasons = [r.strip() for r in loss_reasons_str.split(',') if r.strip()] or None
+
+    result = query_analysis(start_date, end_date, loss_reasons)
+
+    if result is None:
+        return jsonify({'success': False, 'error': '查詢失敗，請稍後再試'}), 500
+
+    if 'error' in result:
+        return jsonify({'success': False, 'error': result['error']}), 400
+
+    # Return summary only (no detail array) to keep response lightweight
+    summary = {
+        'kpi': result.get('kpi'),
+        'charts': result.get('charts'),
+        'daily_trend': result.get('daily_trend'),
+        'available_loss_reasons': result.get('available_loss_reasons'),
+        'genealogy_status': result.get('genealogy_status'),
+        'detail_total_count': len(result.get('detail', [])),
+    }
+
+    return jsonify({'success': True, 'data': summary})
+
+
+@mid_section_defect_bp.route('/analysis/detail', methods=['GET'])
+def api_analysis_detail():
+    """API: Get paginated detail table for mid-section defect analysis.
+
+    Query Parameters:
+        start_date: Start date (YYYY-MM-DD), required
+        end_date: End date (YYYY-MM-DD), required
+        loss_reasons: Comma-separated loss reason names, optional
+        page: Page number (default 1)
+        page_size: Records per page (default 200, max 500)
+    """
+    start_date = request.args.get('start_date')
+    end_date = request.args.get('end_date')
+
+    if not start_date or not end_date:
+        return jsonify({
+            'success': False,
+            'error': '必須提供 start_date 和 end_date 參數'
+        }), 400
+
+    loss_reasons_str = request.args.get('loss_reasons', '')
+    loss_reasons = [r.strip() for r in loss_reasons_str.split(',') if r.strip()] or None
+
+    page = max(request.args.get('page', 1, type=int), 1)
+    page_size = max(1, min(request.args.get('page_size', 200, type=int), 500))
+
+    result = query_analysis_detail(
+        start_date, end_date, loss_reasons,
+        page=page, page_size=page_size,
+    )
+
+    if result is None:
+        return jsonify({'success': False, 'error': '查詢失敗，請稍後再試'}), 500
+
+    if 'error' in result:
+        return jsonify({'success': False, 'error': result['error']}), 400
+
+    return jsonify({'success': True, 'data': result})
+
+
+@mid_section_defect_bp.route('/loss-reasons', methods=['GET'])
+def api_loss_reasons():
+    """API: Get all TMTT loss reasons (cached daily).
+
+    No parameters required — returns all loss reasons from last 180 days,
+    cached in Redis with 24h TTL for instant dropdown population.
+
+    Returns:
+        JSON with loss_reasons list.
+    """
+    result = query_all_loss_reasons()
+
+    if result is None:
+        return jsonify({'success': False, 'error': '查詢失敗，請稍後再試'}), 500
+
+    return jsonify({'success': True, 'data': result})
+
+
+@mid_section_defect_bp.route('/export', methods=['GET'])
+def api_export():
+    """API: Export mid-section defect detail data as CSV.
+
+    Query Parameters:
+        start_date: Start date (YYYY-MM-DD), required
+        end_date: End date (YYYY-MM-DD), required
+        loss_reasons: Comma-separated loss reason names, optional
+
+    Returns:
+        CSV file download.
+    """
+    start_date = request.args.get('start_date')
+    end_date = request.args.get('end_date')
+
+    if not start_date or not end_date:
+        return jsonify({
+            'success': False,
+            'error': '必須提供 start_date 和 end_date 參數'
+        }), 400
+
+    loss_reasons_str = request.args.get('loss_reasons', '')
+    loss_reasons = [r.strip() for r in loss_reasons_str.split(',') if r.strip()] or None
+
+    filename = f"mid_section_defect_{start_date}_to_{end_date}.csv"
+
+    return Response(
+        export_csv(start_date, end_date, loss_reasons),
+        mimetype='text/csv',
+        headers={
+            'Content-Disposition': f'attachment; filename={filename}',
+            'Content-Type': 'text/csv; charset=utf-8-sig'
+        }
+    )
diff --git a/src/mes_dashboard/services/mid_section_defect_service.py b/src/mes_dashboard/services/mid_section_defect_service.py
new file mode 100644
index 0000000..2843105
--- /dev/null
+++ b/src/mes_dashboard/services/mid_section_defect_service.py
@@ -0,0 +1,1163 @@
+# -*- coding: utf-8 -*-
+"""Mid-Section Defect Traceability Analysis Service.
+
+Reverse traceability from TMTT (test) station back to upstream production stations.
+Traces LOT genealogy (splits + merges) to attribute TMTT defects to upstream machines.
+
+Data Pipeline:
+    Query 1:  TMTT Detection    → TMTT lots with ALL loss reasons
+    Query 2a: Split Chain BFS   → CONTAINER.SPLITFROMID upward traversal
+    Query 2b: Merge Expansion   → COMBINEDASSYLOTS reverse merge lookup
+    Query 3:  Upstream History  → LOTWIPHISTORY for ALL ancestor CONTAINERIDs
+    Python:   Resolve ancestors → Attribute defects → Aggregate
+
+Attribution Method (Sum):
+    For upstream machine M at station S:
+      attributed_rejectqty = SUM(TMTT REJECTQTY for all linked TMTT lots)
+      attributed_trackinqty = SUM(TMTT TRACKINQTY for all linked TMTT lots)
+      rate = attributed_rejectqty / attributed_trackinqty × 100
+"""
+
+import csv
+import io
+import logging
+import math
+from collections import defaultdict
+from datetime import datetime
+from typing import Optional, Dict, List, Any, Set, Tuple, Generator
+
+import pandas as pd
+
+from mes_dashboard.core.database import read_sql_df
+from mes_dashboard.core.cache import cache_get, cache_set, make_cache_key
+from mes_dashboard.sql import SQLLoader, QueryBuilder
+from mes_dashboard.config.workcenter_groups import get_workcenter_group
+
+logger = logging.getLogger('mes_dashboard.mid_section_defect')
+
+# Constants
+MAX_QUERY_DAYS = 180
+CACHE_TTL_TMTT = 300            # 5 min for TMTT detection data
+CACHE_TTL_LOSS_REASONS = 86400  # 24h for loss reason list (daily sync)
+ORACLE_IN_BATCH_SIZE = 1000     # Oracle IN clause limit
+
+# Mid-section workcenter group order range (成型 through 測試)
+MID_SECTION_ORDER_MIN = 4   # 成型
+MID_SECTION_ORDER_MAX = 11  # 測試
+
+# Top N for chart display (rest grouped as "其他")
+TOP_N = 10
+
+# Dimension column mapping for attribution charts
+DIMENSION_MAP = {
+    'by_station': 'WORKCENTER_GROUP',
+    'by_machine': 'EQUIPMENT_NAME',
+    'by_workflow': 'WORKFLOW',
+    'by_package': 'PRODUCTLINENAME',
+    'by_pj_type': 'PJ_TYPE',
+    'by_tmtt_machine': 'TMTT_EQUIPMENTNAME',
+}
+
+# CSV export column config
+CSV_COLUMNS = [
+    ('CONTAINERNAME', 'LOT ID'),
+    ('PJ_TYPE', 'TYPE'),
+    ('PRODUCTLINENAME', 'PACKAGE'),
+    ('WORKFLOW', 'WORKFLOW'),
+    ('FINISHEDRUNCARD', '完工流水碼'),
+    ('TMTT_EQUIPMENTNAME', 'TMTT設備'),
+    ('INPUT_QTY', '投入數'),
+    ('LOSS_REASON', '不良原因'),
+    ('DEFECT_QTY', '不良數'),
+    ('DEFECT_RATE', '不良率(%)'),
+    ('ANCESTOR_COUNT', '上游LOT數'),
+    ('UPSTREAM_MACHINES', '上游機台'),
+]
+
+
+# ============================================================
+# Public API
+# ============================================================
+
+def query_analysis(
+    start_date: str,
+    end_date: str,
+    loss_reasons: Optional[List[str]] = None,
+) -> Optional[Dict[str, Any]]:
+    """Main entry point for mid-section defect traceability analysis.
+
+    Args:
+        start_date: Start date (YYYY-MM-DD)
+        end_date: End date (YYYY-MM-DD)
+        loss_reasons: Optional list of loss reasons to filter (None = all)
+
+    Returns:
+        Dict with kpi, charts, detail, available_loss_reasons, genealogy_status.
+    """
+    error = _validate_date_range(start_date, end_date)
+    if error:
+        return {'error': error}
+
+    # Check full analysis cache
+    cache_key = make_cache_key(
+        "mid_section_defect",
+        filters={
+            'start_date': start_date,
+            'end_date': end_date,
+            'loss_reasons': sorted(loss_reasons) if loss_reasons else None,
+        },
+    )
+    cached = cache_get(cache_key)
+    if cached is not None:
+        return cached
+
+    # Stage 1: TMTT detection data
+    tmtt_df = _fetch_tmtt_data(start_date, end_date)
+    if tmtt_df is None:
+        return None
+    if tmtt_df.empty:
+        return _empty_result()
+
+    # Extract available loss reasons before filtering
+    available_loss_reasons = sorted(
+        tmtt_df.loc[tmtt_df['REJECTQTY'] > 0, 'LOSSREASONNAME']
+        .dropna().unique().tolist()
+    )
+
+    # Apply loss reason filter if specified
+    if loss_reasons:
+        filtered_df = tmtt_df[
+            (tmtt_df['LOSSREASONNAME'].isin(loss_reasons))
+            | (tmtt_df['REJECTQTY'] == 0)
+            | (tmtt_df['LOSSREASONNAME'].isna())
+        ].copy()
+    else:
+        filtered_df = tmtt_df
+
+    # Stage 2: Genealogy resolution (split chain + merge expansion)
+    tmtt_cids = tmtt_df['CONTAINERID'].unique().tolist()
+    tmtt_names = {}
+    for _, r in tmtt_df.drop_duplicates('CONTAINERID').iterrows():
+        tmtt_names[r['CONTAINERID']] = _safe_str(r.get('CONTAINERNAME'))
+
+    ancestors = {}
+    genealogy_status = 'ready'
+
+    if tmtt_cids:
+        try:
+            ancestors = _resolve_full_genealogy(tmtt_cids, tmtt_names)
+        except Exception as exc:
+            logger.error(f"Genealogy resolution failed: {exc}", exc_info=True)
+            genealogy_status = 'error'
+
+    # Stage 3: Upstream history for ALL CIDs (TMTT lots + ancestors)
+    all_query_cids = set(tmtt_cids)
+    for anc_set in ancestors.values():
+        all_query_cids.update(anc_set)
+    # Filter out any non-string values (NaN/None from pandas)
+    all_query_cids = {c for c in all_query_cids if isinstance(c, str) and c}
+
+    upstream_by_cid = {}
+    if all_query_cids:
+        try:
+            upstream_by_cid = _fetch_upstream_history(list(all_query_cids))
+        except Exception as exc:
+            logger.error(f"Upstream history query failed: {exc}", exc_info=True)
+            genealogy_status = 'error'
+    tmtt_data = _build_tmtt_lookup(filtered_df)
+    attribution = _attribute_defects(
+        tmtt_data, ancestors, upstream_by_cid, loss_reasons,
+    )
+
+    result = {
+        'kpi': _build_kpi(filtered_df, attribution, loss_reasons),
+        'available_loss_reasons': available_loss_reasons,
+        'charts': _build_all_charts(attribution, tmtt_data),
+        'daily_trend': _build_daily_trend(filtered_df, loss_reasons),
+        'detail': _build_detail_table(filtered_df, ancestors, upstream_by_cid),
+        'genealogy_status': genealogy_status,
+    }
+
+    # Only cache successful results (don't cache upstream errors)
+    if genealogy_status == 'ready':
+        cache_set(cache_key, result, ttl=CACHE_TTL_TMTT)
+    return result
+
+
+def query_analysis_detail(
+    start_date: str,
+    end_date: str,
+    loss_reasons: Optional[List[str]] = None,
+    page: int = 1,
+    page_size: int = 200,
+) -> Optional[Dict[str, Any]]:
+    """Return a paginated slice of the detail table from cached analysis.
+
+    Calls query_analysis() which handles caching internally.
+    Sorts detail by DEFECT_RATE descending (worst first) before paginating.
+    """
+    result = query_analysis(start_date, end_date, loss_reasons)
+    if result is None:
+        return None
+    if 'error' in result:
+        return result
+
+    detail = result.get('detail', [])
+    detail_sorted = sorted(
+        detail, key=lambda r: r.get('DEFECT_RATE', 0), reverse=True,
+    )
+
+    total_count = len(detail_sorted)
+    total_pages = max(1, (total_count + page_size - 1) // page_size)
+    page = max(1, min(page, total_pages))
+    offset = (page - 1) * page_size
+
+    return {
+        'detail': detail_sorted[offset:offset + page_size],
+        'pagination': {
+            'page': page,
+            'page_size': page_size,
+            'total_count': total_count,
+            'total_pages': total_pages,
+        },
+    }
+
+
+def query_all_loss_reasons() -> Optional[Dict[str, Any]]:
+    """Get all TMTT loss reasons (cached daily in Redis).
+
+    Lightweight query: DISTINCT LOSSREASONNAME from last 180 days.
+    Cached with 24h TTL — suitable for dropdown population on page load.
+
+    Returns:
+        Dict with 'loss_reasons' list, or None on failure.
+    """
+    cache_key = make_cache_key("mid_section_loss_reasons")
+    cached = cache_get(cache_key)
+    if cached is not None:
+        return cached
+
+    try:
+        sql = SQLLoader.load("mid_section_defect/all_loss_reasons")
+        df = read_sql_df(sql, {})
+        if df is None:
+            logger.error("Loss reasons query returned None")
+            return None
+
+        reasons = sorted(df['LOSSREASONNAME'].dropna().unique().tolist())
+        result = {'loss_reasons': reasons}
+        logger.info(f"Loss reasons: {len(reasons)} distinct values cached (24h TTL)")
+        cache_set(cache_key, result, ttl=CACHE_TTL_LOSS_REASONS)
+        return result
+    except Exception as exc:
+        logger.error(f"Loss reasons query failed: {exc}", exc_info=True)
+        return None
+
+
+def export_csv(
+    start_date: str,
+    end_date: str,
+    loss_reasons: Optional[List[str]] = None,
+) -> Generator[str, None, None]:
+    """Stream CSV export of detail data.
+
+    Yields:
+        CSV lines as strings.
+    """
+    result = query_analysis(start_date, end_date, loss_reasons)
+
+    # BOM for Excel UTF-8 compatibility
+    yield '\ufeff'
+
+    output = io.StringIO()
+    writer = csv.writer(output)
+
+    writer.writerow([label for _, label in CSV_COLUMNS])
+    yield output.getvalue()
+    output.seek(0)
+    output.truncate(0)
+
+    if result is None or 'error' in result:
+        return
+
+    for row in result.get('detail', []):
+        writer.writerow([row.get(col, '') for col, _ in CSV_COLUMNS])
+        yield output.getvalue()
+        output.seek(0)
+        output.truncate(0)
+
+
+# ============================================================
+# Helpers
+# ============================================================
+
+def _safe_str(v, default=''):
+    if v is None or (isinstance(v, float) and math.isnan(v)):
+        return default
+    try:
+        if pd.isna(v):
+            return default
+    except (TypeError, ValueError):
+        pass
+    return str(v)
+
+
+def _safe_float(v, default=0.0):
+    if v is None:
+        return default
+    try:
+        f = float(v)
+        if math.isnan(f) or math.isinf(f):
+            return default
+        return f
+    except (TypeError, ValueError):
+        return default
+
+
+def _safe_int(v, default=0):
+    return int(_safe_float(v, float(default)))
+
+
+def _empty_result() -> Dict[str, Any]:
+    return {
+        'kpi': {
+            'total_input': 0, 'lot_count': 0,
+            'total_defect_qty': 0, 'total_defect_rate': 0.0,
+            'top_loss_reason': '', 'affected_machine_count': 0,
+        },
+        'available_loss_reasons': [],
+        'charts': {k: [] for k in DIMENSION_MAP},
+        'daily_trend': [],
+        'detail': [],
+        'genealogy_status': 'ready',
+    }
+
+
+# ============================================================
+# Validation
+# ============================================================
+
+def _validate_date_range(start_date: str, end_date: str) -> Optional[str]:
+    try:
+        start = datetime.strptime(start_date, '%Y-%m-%d')
+        end = datetime.strptime(end_date, '%Y-%m-%d')
+    except (ValueError, TypeError):
+        return '日期格式無效，請使用 YYYY-MM-DD'
+
+    if start > end:
+        return '起始日期不能晚於結束日期'
+
+    if (end - start).days > MAX_QUERY_DAYS:
+        return f'查詢範圍不能超過 {MAX_QUERY_DAYS} 天'
+
+    return None
+
+
+# ============================================================
+# Query 1: TMTT Detection Data
+# ============================================================
+
+def _fetch_tmtt_data(start_date: str, end_date: str) -> Optional[pd.DataFrame]:
+    """Execute tmtt_detection.sql and return raw DataFrame."""
+    cache_key = make_cache_key(
+        "mid_section_tmtt",
+        filters={'start_date': start_date, 'end_date': end_date},
+    )
+    cached = cache_get(cache_key)
+    if cached is not None:
+        # Cache stores list-of-dicts (JSON-serializable), reconstruct DataFrame
+        if isinstance(cached, list):
+            return pd.DataFrame(cached) if cached else pd.DataFrame()
+        return None
+
+    try:
+        sql = SQLLoader.load("mid_section_defect/tmtt_detection")
+        params = {'start_date': start_date, 'end_date': end_date}
+        df = read_sql_df(sql, params)
+        if df is None:
+            logger.error("TMTT detection query returned None")
+            return None
+        logger.info(
+            f"TMTT detection: {len(df)} rows, "
+            f"{df['CONTAINERID'].nunique() if not df.empty else 0} unique lots"
+        )
+        # Cache as list-of-dicts for JSON serialization via Redis
+        cache_set(cache_key, df.to_dict('records'), ttl=CACHE_TTL_TMTT)
+        return df
+    except Exception as exc:
+        logger.error(f"TMTT detection query failed: {exc}", exc_info=True)
+        return None
+
+
+# ============================================================
+# Query 2: LOT Genealogy
+# ============================================================
+
+def _resolve_full_genealogy(
+    tmtt_cids: List[str],
+    tmtt_names: Dict[str, str],
+) -> Dict[str, Set[str]]:
+    """Resolve full genealogy for TMTT lots via SPLITFROMID + COMBINEDASSYLOTS.
+
+    Step 1: BFS upward through DW_MES_CONTAINER.SPLITFROMID
+    Step 2: Merge expansion via DW_MES_PJ_COMBINEDASSYLOTS
+    Step 3: BFS on merge source CIDs (one more round)
+
+    Args:
+        tmtt_cids: TMTT lot CONTAINERIDs
+        tmtt_names: {cid: containername} from TMTT detection data
+
+    Returns:
+        {tmtt_cid: set(all ancestor CIDs)}
+    """
+    # ---- Step 1: Split chain BFS upward ----
+    child_to_parent, cid_to_name = _bfs_split_chain(tmtt_cids, tmtt_names)
+
+    # Build initial ancestor sets per TMTT lot (walk up split chain)
+    ancestors: Dict[str, Set[str]] = {}
+    for tmtt_cid in tmtt_cids:
+        visited: Set[str] = set()
+        current = tmtt_cid
+        while current in child_to_parent:
+            parent = child_to_parent[current]
+            if parent in visited:
+                break  # cycle protection
+            visited.add(parent)
+            current = parent
+        ancestors[tmtt_cid] = visited
+
+    # ---- Step 2: Merge expansion via COMBINEDASSYLOTS ----
+    all_names = set(cid_to_name.values())
+    if not all_names:
+        _log_genealogy_summary(ancestors, tmtt_cids, 0)
+        return ancestors
+
+    merge_source_map = _fetch_merge_sources(list(all_names))
+    if not merge_source_map:
+        _log_genealogy_summary(ancestors, tmtt_cids, 0)
+        return ancestors
+
+    # Reverse map: name → set of CIDs with that name
+    name_to_cids: Dict[str, Set[str]] = defaultdict(set)
+    for cid, name in cid_to_name.items():
+        name_to_cids[name].add(cid)
+
+    # Expand ancestors with merge sources
+    merge_source_cids_all: Set[str] = set()
+    for tmtt_cid in tmtt_cids:
+        self_and_ancestors = ancestors[tmtt_cid] | {tmtt_cid}
+        for cid in list(self_and_ancestors):
+            name = cid_to_name.get(cid)
+            if name and name in merge_source_map:
+                for src_cid in merge_source_map[name]:
+                    if src_cid != cid and src_cid not in self_and_ancestors:
+                        ancestors[tmtt_cid].add(src_cid)
+                        merge_source_cids_all.add(src_cid)
+
+    # ---- Step 3: BFS on merge source CIDs ----
+    seen = set(tmtt_cids) | set(child_to_parent.values()) | set(child_to_parent.keys())
+    new_merge_cids = list(merge_source_cids_all - seen)
+    if new_merge_cids:
+        merge_c2p, _ = _bfs_split_chain(new_merge_cids, {})
+        child_to_parent.update(merge_c2p)
+
+        # Walk up merge sources' split chains for each TMTT lot
+        for tmtt_cid in tmtt_cids:
+            for merge_cid in list(ancestors[tmtt_cid] & merge_source_cids_all):
+                current = merge_cid
+                while current in merge_c2p:
+                    parent = merge_c2p[current]
+                    if parent in ancestors[tmtt_cid]:
+                        break
+                    ancestors[tmtt_cid].add(parent)
+                    current = parent
+
+    _log_genealogy_summary(ancestors, tmtt_cids, len(merge_source_cids_all))
+    return ancestors
+
+
+def _bfs_split_chain(
+    start_cids: List[str],
+    initial_names: Dict[str, str],
+) -> Tuple[Dict[str, str], Dict[str, str]]:
+    """BFS upward through DW_MES_CONTAINER.SPLITFROMID.
+
+    Args:
+        start_cids: Starting CONTAINERIDs
+        initial_names: Pre-known {cid: containername} mappings
+
+    Returns:
+        child_to_parent: {child_cid: parent_cid} for all split edges
+        cid_to_name: {cid: containername} for all encountered CIDs
+    """
+    child_to_parent: Dict[str, str] = {}
+    cid_to_name: Dict[str, str] = dict(initial_names)
+    seen: Set[str] = set(start_cids)
+    frontier = list(start_cids)
+    bfs_round = 0
+
+    while frontier:
+        bfs_round += 1
+        batch_results: List[Dict[str, Any]] = []
+
+        for i in range(0, len(frontier), ORACLE_IN_BATCH_SIZE):
+            batch = frontier[i:i + ORACLE_IN_BATCH_SIZE]
+            builder = QueryBuilder()
+            builder.add_in_condition("c.CONTAINERID", batch)
+            sql = SQLLoader.load_with_params(
+                "mid_section_defect/split_chain",
+                CID_FILTER=builder.get_conditions_sql(),
+            )
+            try:
+                df = read_sql_df(sql, builder.params)
+                if df is not None and not df.empty:
+                    batch_results.extend(df.to_dict('records'))
+            except Exception as exc:
+                logger.warning(f"Split chain BFS round {bfs_round} batch failed: {exc}")
+
+        new_parents: Set[str] = set()
+        for row in batch_results:
+            cid = row['CONTAINERID']
+            split_from = row.get('SPLITFROMID')
+            name = row.get('CONTAINERNAME')
+
+            if isinstance(name, str) and name:
+                cid_to_name[cid] = name
+            if isinstance(split_from, str) and split_from and cid != split_from:
+                child_to_parent[cid] = split_from
+                if split_from not in seen:
+                    new_parents.add(split_from)
+                    seen.add(split_from)
+
+        frontier = list(new_parents)
+        if bfs_round > 20:
+            logger.warning("Split chain BFS exceeded 20 rounds, stopping")
+            break
+
+    logger.info(
+        f"Split chain BFS: {bfs_round} rounds, "
+        f"{len(child_to_parent)} split edges, "
+        f"{len(cid_to_name)} names collected"
+    )
+    return child_to_parent, cid_to_name
+
+
+def _fetch_merge_sources(
+    finished_names: List[str],
+) -> Dict[str, List[str]]:
+    """Find source lots merged into finished lots via COMBINEDASSYLOTS.
+
+    Args:
+        finished_names: CONTAINERNAMEs to look up as FINISHEDNAME
+
+    Returns:
+        {finished_name: [source_cid, ...]}
+    """
+    result: Dict[str, List[str]] = {}
+
+    for i in range(0, len(finished_names), ORACLE_IN_BATCH_SIZE):
+        batch = finished_names[i:i + ORACLE_IN_BATCH_SIZE]
+        builder = QueryBuilder()
+        builder.add_in_condition("ca.FINISHEDNAME", batch)
+        sql = SQLLoader.load_with_params(
+            "mid_section_defect/merge_lookup",
+            FINISHED_NAME_FILTER=builder.get_conditions_sql(),
+        )
+        try:
+            df = read_sql_df(sql, builder.params)
+            if df is not None and not df.empty:
+                for _, row in df.iterrows():
+                    fn = row['FINISHEDNAME']
+                    src = row['SOURCE_CID']
+                    if isinstance(fn, str) and fn and isinstance(src, str) and src:
+                        result.setdefault(fn, []).append(src)
+        except Exception as exc:
+            logger.warning(f"Merge lookup batch failed: {exc}")
+
+    if result:
+        total_sources = sum(len(v) for v in result.values())
+        logger.info(
+            f"Merge lookup: {len(result)} finished names → {total_sources} source CIDs"
+        )
+    return result
+
+
+def _log_genealogy_summary(
+    ancestors: Dict[str, Set[str]],
+    tmtt_cids: List[str],
+    merge_count: int,
+) -> None:
+    total_ancestors = sum(len(v) for v in ancestors.values())
+    lots_with_ancestors = sum(1 for v in ancestors.values() if v)
+    logger.info(
+        f"Genealogy resolved: {lots_with_ancestors}/{len(tmtt_cids)} lots have ancestors, "
+        f"{total_ancestors} total ancestor links, "
+        f"{merge_count} merge sources"
+    )
+
+
+# ============================================================
+# Query 3: Upstream Production History
+# ============================================================
+
+def _fetch_upstream_history(
+    all_cids: List[str],
+) -> Dict[str, List[Dict[str, Any]]]:
+    """Fetch upstream production history for ancestor CONTAINERIDs.
+
+    Batches queries to respect Oracle IN clause limit.
+    Filters by mid-section workcenter groups (order 4-11) in Python.
+
+    Returns:
+        {containerid: [{'workcenter_group': ..., 'equipment_name': ..., ...}, ...]}
+    """
+    if not all_cids:
+        return {}
+
+    unique_cids = list(set(all_cids))
+    all_rows = []
+
+    # Batch query in chunks of ORACLE_IN_BATCH_SIZE
+    for i in range(0, len(unique_cids), ORACLE_IN_BATCH_SIZE):
+        batch = unique_cids[i:i + ORACLE_IN_BATCH_SIZE]
+
+        builder = QueryBuilder()
+        builder.add_in_condition("h.CONTAINERID", batch)
+        conditions_sql = builder.get_conditions_sql()
+        params = builder.params
+
+        sql = SQLLoader.load_with_params(
+            "mid_section_defect/upstream_history",
+            ANCESTOR_FILTER=conditions_sql,
+        )
+
+        try:
+            df = read_sql_df(sql, params)
+            if df is not None and not df.empty:
+                all_rows.append(df)
+        except Exception as exc:
+            logger.error(
+                f"Upstream history batch {i//ORACLE_IN_BATCH_SIZE + 1} failed: {exc}",
+                exc_info=True,
+            )
+
+    if not all_rows:
+        return {}
+
+    combined = pd.concat(all_rows, ignore_index=True)
+
+    # Filter by mid-section workcenter groups in Python
+    result: Dict[str, List[Dict[str, Any]]] = defaultdict(list)
+    for _, row in combined.iterrows():
+        wc_name = row.get('WORKCENTERNAME', '')
+        group_name, order = get_workcenter_group(wc_name)
+        if group_name is None or order < MID_SECTION_ORDER_MIN or order > MID_SECTION_ORDER_MAX:
+            continue
+
+        cid = row['CONTAINERID']
+        result[cid].append({
+            'workcenter_group': group_name,
+            'workcenter_group_order': order,
+            'equipment_id': _safe_str(row.get('EQUIPMENTID')),
+            'equipment_name': _safe_str(row.get('EQUIPMENTNAME')),
+            'spec_name': _safe_str(row.get('SPECNAME')),
+            'track_in_time': _safe_str(row.get('TRACKINTIMESTAMP')),
+        })
+
+    logger.info(
+        f"Upstream history: {len(result)} lots with mid-section records, "
+        f"from {len(unique_cids)} queried CIDs"
+    )
+    return dict(result)
+
+
+# ============================================================
+# TMTT Data Lookup
+# ============================================================
+
+def _build_tmtt_lookup(
+    df: pd.DataFrame,
+) -> Dict[str, Dict[str, Any]]:
+    """Build lookup dict from TMTT DataFrame.
+
+    Returns:
+        {containerid: {
+            'trackinqty': int,
+            'rejectqty_by_reason': {reason: qty},
+            'containername': str,
+            'workflow': str,
+            'productlinename': str,
+            'pj_type': str,
+            'tmtt_equipmentname': str,
+            'trackintimestamp': str,
+        }}
+    """
+    if df.empty:
+        return {}
+
+    lookup: Dict[str, Dict[str, Any]] = {}
+    for _, row in df.iterrows():
+        cid = row['CONTAINERID']
+        if cid not in lookup:
+            lookup[cid] = {
+                'trackinqty': _safe_int(row.get('TRACKINQTY')),
+                'rejectqty_by_reason': {},
+                'containername': _safe_str(row.get('CONTAINERNAME')),
+                'workflow': _safe_str(row.get('WORKFLOW')),
+                'productlinename': _safe_str(row.get('PRODUCTLINENAME')),
+                'pj_type': _safe_str(row.get('PJ_TYPE')),
+                'tmtt_equipmentname': _safe_str(row.get('TMTT_EQUIPMENTNAME')),
+                'trackintimestamp': _safe_str(row.get('TRACKINTIMESTAMP')),
+            }
+
+        reason = row.get('LOSSREASONNAME')
+        qty = _safe_int(row.get('REJECTQTY'))
+        if reason and qty > 0:
+            lookup[cid]['rejectqty_by_reason'][reason] = (
+                lookup[cid]['rejectqty_by_reason'].get(reason, 0) + qty
+            )
+
+    return lookup
+
+
+# ============================================================
+# Defect Attribution Engine
+# ============================================================
+
+def _attribute_defects(
+    tmtt_data: Dict[str, Dict[str, Any]],
+    ancestors: Dict[str, Set[str]],
+    upstream_by_cid: Dict[str, List[Dict[str, Any]]],
+    loss_reasons: Optional[List[str]] = None,
+) -> List[Dict[str, Any]]:
+    """Attribute TMTT defects to upstream machines.
+
+    For each upstream machine M at station S:
+      - Find all TMTT lots whose ancestors (or self) used M
+      - attributed_rejectqty = SUM(selected REJECTQTY)
+      - attributed_trackinqty = SUM(TRACKINQTY)
+      - rate = attributed_rejectqty / attributed_trackinqty × 100
+
+    Returns:
+        List of attribution records, one per (workcenter_group, equipment_name).
+    """
+    # machine_key → set of TMTT lot CIDs
+    machine_to_tmtt: Dict[Tuple[str, str, str], Set[str]] = defaultdict(set)
+
+    for tmtt_cid, data in tmtt_data.items():
+        ancestor_set = ancestors.get(tmtt_cid, set())
+        # Include the TMTT lot itself (it may have upstream history if no split)
+        all_cids = ancestor_set | {tmtt_cid}
+
+        for anc_cid in all_cids:
+            for record in upstream_by_cid.get(anc_cid, []):
+                machine_key = (
+                    record['workcenter_group'],
+                    record['equipment_name'],
+                    record['equipment_id'],
+                )
+                machine_to_tmtt[machine_key].add(tmtt_cid)
+
+    # Calculate attribution per machine
+    attribution = []
+    for machine_key, tmtt_lot_set in machine_to_tmtt.items():
+        wc_group, eq_name, eq_id = machine_key
+
+        total_trackinqty = sum(
+            tmtt_data[cid]['trackinqty'] for cid in tmtt_lot_set
+            if cid in tmtt_data
+        )
+
+        # Sum defects for selected loss reasons
+        total_rejectqty = 0
+        for cid in tmtt_lot_set:
+            if cid not in tmtt_data:
+                continue
+            by_reason = tmtt_data[cid]['rejectqty_by_reason']
+            if loss_reasons:
+                for reason in loss_reasons:
+                    total_rejectqty += by_reason.get(reason, 0)
+            else:
+                total_rejectqty += sum(by_reason.values())
+
+        rate = round(total_rejectqty / total_trackinqty * 100, 4) if total_trackinqty else 0.0
+
+        # Collect dimension metadata from linked TMTT lots
+        workflows = set()
+        packages = set()
+        pj_types = set()
+        tmtt_machines = set()
+        for cid in tmtt_lot_set:
+            if cid not in tmtt_data:
+                continue
+            d = tmtt_data[cid]
+            if d['workflow']:
+                workflows.add(d['workflow'])
+            if d['productlinename']:
+                packages.add(d['productlinename'])
+            if d['pj_type']:
+                pj_types.add(d['pj_type'])
+            if d['tmtt_equipmentname']:
+                tmtt_machines.add(d['tmtt_equipmentname'])
+
+        attribution.append({
+            'WORKCENTER_GROUP': wc_group,
+            'EQUIPMENT_NAME': eq_name,
+            'EQUIPMENT_ID': eq_id,
+            'TMTT_LOT_COUNT': len(tmtt_lot_set),
+            'INPUT_QTY': total_trackinqty,
+            'DEFECT_QTY': total_rejectqty,
+            'DEFECT_RATE': rate,
+            # Flatten multi-valued dimensions for charting
+            'WORKFLOW': ', '.join(sorted(workflows)) if workflows else '(未知)',
+            'PRODUCTLINENAME': ', '.join(sorted(packages)) if packages else '(未知)',
+            'PJ_TYPE': ', '.join(sorted(pj_types)) if pj_types else '(未知)',
+            'TMTT_EQUIPMENTNAME': ', '.join(sorted(tmtt_machines)) if tmtt_machines else '(未知)',
+        })
+
+    # Sort by defect rate DESC
+    attribution.sort(key=lambda x: x['DEFECT_RATE'], reverse=True)
+
+    return attribution
+
+
+# ============================================================
+# KPI Builder
+# ============================================================
+
+def _build_kpi(
+    df: pd.DataFrame,
+    attribution: List[Dict[str, Any]],
+    loss_reasons: Optional[List[str]] = None,
+) -> Dict[str, Any]:
+    """Build KPI summary."""
+    if df.empty:
+        return {
+            'total_input': 0, 'lot_count': 0,
+            'total_defect_qty': 0, 'total_defect_rate': 0.0,
+            'top_loss_reason': '', 'affected_machine_count': 0,
+        }
+
+    # Deduplicate for INPUT
+    unique_lots = df.drop_duplicates(subset=['CONTAINERID'])
+    total_input = int(unique_lots['TRACKINQTY'].sum())
+    lot_count = len(unique_lots)
+
+    # Defect totals
+    defect_rows = df[df['REJECTQTY'] > 0]
+    if loss_reasons:
+        defect_rows = defect_rows[defect_rows['LOSSREASONNAME'].isin(loss_reasons)]
+
+    total_defect_qty = int(defect_rows['REJECTQTY'].sum()) if not defect_rows.empty else 0
+    total_defect_rate = round(
+        total_defect_qty / total_input * 100, 4
+    ) if total_input else 0.0
+
+    # Top loss reason
+    top_reason = ''
+    if not defect_rows.empty:
+        reason_sums = defect_rows.groupby('LOSSREASONNAME')['REJECTQTY'].sum()
+        if not reason_sums.empty:
+            top_reason = _safe_str(reason_sums.idxmax())
+
+    # Count unique upstream machines with defects attributed
+    affected_machines = sum(1 for a in attribution if a['DEFECT_QTY'] > 0)
+
+    return {
+        'total_input': total_input,
+        'lot_count': lot_count,
+        'total_defect_qty': total_defect_qty,
+        'total_defect_rate': total_defect_rate,
+        'top_loss_reason': top_reason,
+        'affected_machine_count': affected_machines,
+    }
+
+
+# ============================================================
+# Chart Builders
+# ============================================================
+
+def _build_chart_data(
+    records: List[Dict[str, Any]],
+    dimension: str,
+) -> List[Dict[str, Any]]:
+    """Build Top N + Other Pareto chart data for a given dimension.
+
+    Groups attribution records by dimension, sums defect qty, takes top N,
+    groups rest as "其他".
+    """
+    if not records:
+        return []
+
+    # Aggregate by dimension
+    dim_agg: Dict[str, Dict[str, Any]] = defaultdict(
+        lambda: {'input_qty': 0, 'defect_qty': 0, 'lot_count': 0}
+    )
+    for rec in records:
+        key = rec.get(dimension, '(未知)') or '(未知)'
+        # For multi-valued dimensions (comma-separated), split and attribute to each
+        if ',' in key:
+            keys = [k.strip() for k in key.split(',')]
+        else:
+            keys = [key]
+        for k in keys:
+            dim_agg[k]['input_qty'] += rec['INPUT_QTY']
+            dim_agg[k]['defect_qty'] += rec['DEFECT_QTY']
+            dim_agg[k]['lot_count'] += rec['TMTT_LOT_COUNT']
+
+    # Sort by defect qty DESC
+    sorted_items = sorted(dim_agg.items(), key=lambda x: x[1]['defect_qty'], reverse=True)
+
+    # Top N + Other
+    items = []
+    other = {'input_qty': 0, 'defect_qty': 0, 'lot_count': 0}
+    for i, (name, data) in enumerate(sorted_items):
+        if i < TOP_N:
+            rate = round(data['defect_qty'] / data['input_qty'] * 100, 4) if data['input_qty'] else 0.0
+            items.append({
+                'name': name,
+                'input_qty': data['input_qty'],
+                'defect_qty': data['defect_qty'],
+                'defect_rate': rate,
+                'lot_count': data['lot_count'],
+            })
+        else:
+            other['input_qty'] += data['input_qty']
+            other['defect_qty'] += data['defect_qty']
+            other['lot_count'] += data['lot_count']
+
+    if other['defect_qty'] > 0 or other['input_qty'] > 0:
+        rate = round(other['defect_qty'] / other['input_qty'] * 100, 4) if other['input_qty'] else 0.0
+        items.append({
+            'name': '其他',
+            'input_qty': other['input_qty'],
+            'defect_qty': other['defect_qty'],
+            'defect_rate': rate,
+            'lot_count': other['lot_count'],
+        })
+
+    # Add cumulative percentage
+    total_defects = sum(item['defect_qty'] for item in items)
+    cumsum = 0
+    for item in items:
+        cumsum += item['defect_qty']
+        item['cumulative_pct'] = round(cumsum / total_defects * 100, 2) if total_defects else 0.0
+
+    return items
+
+
+def _build_loss_reason_chart(
+    df: pd.DataFrame,
+    loss_reasons: Optional[List[str]] = None,
+) -> List[Dict[str, Any]]:
+    """Build loss reason distribution chart from TMTT data (not attribution)."""
+    if df.empty:
+        return []
+
+    defect_rows = df[df['REJECTQTY'] > 0].copy()
+    if loss_reasons:
+        defect_rows = defect_rows[defect_rows['LOSSREASONNAME'].isin(loss_reasons)]
+
+    if defect_rows.empty:
+        return []
+
+    # Aggregate by loss reason
+    reason_agg = defect_rows.groupby('LOSSREASONNAME')['REJECTQTY'].sum()
+    reason_agg = reason_agg.sort_values(ascending=False)
+
+    # Deduplicated input per loss reason
+    unique_lots = df.drop_duplicates(subset=['CONTAINERID'])
+    total_input = int(unique_lots['TRACKINQTY'].sum())
+
+    items = []
+    total_defects = int(reason_agg.sum())
+    cumsum = 0
+    for reason, qty in reason_agg.items():
+        qty_int = int(qty)
+        cumsum += qty_int
+        rate = round(qty_int / total_input * 100, 4) if total_input else 0.0
+        items.append({
+            'name': _safe_str(reason),
+            'defect_qty': qty_int,
+            'defect_rate': rate,
+            'cumulative_pct': round(cumsum / total_defects * 100, 2) if total_defects else 0.0,
+        })
+
+    return items
+
+
+def _build_all_charts(
+    attribution: List[Dict[str, Any]],
+    tmtt_data: Dict[str, Dict[str, Any]],
+) -> Dict[str, List[Dict]]:
+    """Build chart data for all dimensions."""
+    charts = {}
+    for key, dim_col in DIMENSION_MAP.items():
+        charts[key] = _build_chart_data(attribution, dim_col)
+
+    # Loss reason chart is built from TMTT data directly (not attribution)
+    # Reconstruct a minimal df from tmtt_data for the loss reason chart
+    loss_rows = []
+    for cid, data in tmtt_data.items():
+        trackinqty = data['trackinqty']
+        if data['rejectqty_by_reason']:
+            for reason, qty in data['rejectqty_by_reason'].items():
+                loss_rows.append({
+                    'CONTAINERID': cid,
+                    'TRACKINQTY': trackinqty,
+                    'LOSSREASONNAME': reason,
+                    'REJECTQTY': qty,
+                })
+        else:
+            loss_rows.append({
+                'CONTAINERID': cid,
+                'TRACKINQTY': trackinqty,
+                'LOSSREASONNAME': None,
+                'REJECTQTY': 0,
+            })
+    if loss_rows:
+        loss_df = pd.DataFrame(loss_rows)
+        charts['by_loss_reason'] = _build_loss_reason_chart(loss_df)
+    else:
+        charts['by_loss_reason'] = []
+
+    return charts
+
+
+# ============================================================
+# Daily Trend
+# ============================================================
+
+def _build_daily_trend(
+    df: pd.DataFrame,
+    loss_reasons: Optional[List[str]] = None,
+) -> List[Dict[str, Any]]:
+    """Build daily defect rate trend data."""
+    if df.empty:
+        return []
+
+    work_df = df.copy()
+    work_df['DATE'] = pd.to_datetime(work_df['TRACKINTIMESTAMP']).dt.strftime('%Y-%m-%d')
+
+    # Daily INPUT (deduplicated by CONTAINERID per date)
+    daily_input = (
+        work_df.drop_duplicates(subset=['CONTAINERID', 'DATE'])
+        .groupby('DATE')['TRACKINQTY']
+        .sum()
+    )
+
+    # Daily defects
+    defect_rows = work_df[work_df['REJECTQTY'] > 0]
+    if loss_reasons:
+        defect_rows = defect_rows[defect_rows['LOSSREASONNAME'].isin(loss_reasons)]
+
+    daily_defects = (
+        defect_rows.groupby('DATE')['REJECTQTY'].sum()
+        if not defect_rows.empty
+        else pd.Series(dtype=float)
+    )
+
+    combined = pd.DataFrame({
+        'input_qty': daily_input,
+        'defect_qty': daily_defects,
+    }).fillna(0).astype({'defect_qty': int, 'input_qty': int})
+
+    combined['defect_rate'] = (
+        combined['defect_qty'] / combined['input_qty'] * 100
+    ).round(4).where(combined['input_qty'] > 0, 0.0)
+
+    combined = combined.sort_index()
+
+    result = []
+    for date, row in combined.iterrows():
+        result.append({
+            'date': str(date),
+            'input_qty': _safe_int(row['input_qty']),
+            'defect_qty': _safe_int(row['defect_qty']),
+            'defect_rate': _safe_float(row['defect_rate']),
+        })
+
+    return result
+
+
+# ============================================================
+# Detail Table
+# ============================================================
+
+def _build_detail_table(
+    df: pd.DataFrame,
+    ancestors: Dict[str, Set[str]],
+    upstream_by_cid: Dict[str, List[Dict[str, Any]]],
+) -> List[Dict[str, Any]]:
+    """Build LOT-level detail table with upstream machine info."""
+    if df.empty:
+        return []
+
+    # Unique LOT info
+    lot_cols = [
+        'CONTAINERID', 'CONTAINERNAME', 'PJ_TYPE', 'PRODUCTLINENAME',
+        'WORKFLOW', 'FINISHEDRUNCARD', 'TMTT_EQUIPMENTNAME', 'TRACKINQTY',
+    ]
+    lots = df.drop_duplicates(subset=['CONTAINERID'])[
+        [c for c in lot_cols if c in df.columns]
+    ].copy()
+
+    # Aggregate defects per LOT per loss reason
+    defect_rows = df[df['REJECTQTY'] > 0]
+    lot_defects: Dict[str, Dict[str, int]] = defaultdict(lambda: defaultdict(int))
+    for _, row in defect_rows.iterrows():
+        cid = row['CONTAINERID']
+        reason = _safe_str(row.get('LOSSREASONNAME'))
+        qty = _safe_int(row.get('REJECTQTY'))
+        if reason and qty > 0:
+            lot_defects[cid][reason] += qty
+
+    result = []
+    for _, row in lots.iterrows():
+        cid = row['CONTAINERID']
+        input_qty = _safe_int(row.get('TRACKINQTY'))
+        ancestor_set = ancestors.get(cid, set())
+        all_cids = ancestor_set | {cid}
+
+        # Collect upstream machines
+        upstream_machines = set()
+        for anc_cid in all_cids:
+            for rec in upstream_by_cid.get(anc_cid, []):
+                upstream_machines.add(f"{rec['workcenter_group']}/{rec['equipment_name']}")
+
+        # Build one row per loss reason for this LOT
+        reasons = lot_defects.get(cid, {})
+        if reasons:
+            for reason, qty in sorted(reasons.items()):
+                rate = round(qty / input_qty * 100, 4) if input_qty else 0.0
+                result.append({
+                    'CONTAINERNAME': _safe_str(row.get('CONTAINERNAME')),
+                    'PJ_TYPE': _safe_str(row.get('PJ_TYPE')),
+                    'PRODUCTLINENAME': _safe_str(row.get('PRODUCTLINENAME')),
+                    'WORKFLOW': _safe_str(row.get('WORKFLOW')),
+                    'FINISHEDRUNCARD': _safe_str(row.get('FINISHEDRUNCARD')),
+                    'TMTT_EQUIPMENTNAME': _safe_str(row.get('TMTT_EQUIPMENTNAME')),
+                    'INPUT_QTY': input_qty,
+                    'LOSS_REASON': reason,
+                    'DEFECT_QTY': qty,
+                    'DEFECT_RATE': rate,
+                    'ANCESTOR_COUNT': len(ancestor_set),
+                    'UPSTREAM_MACHINES': ', '.join(sorted(upstream_machines)),
+                })
+        else:
+            result.append({
+                'CONTAINERNAME': _safe_str(row.get('CONTAINERNAME')),
+                'PJ_TYPE': _safe_str(row.get('PJ_TYPE')),
+                'PRODUCTLINENAME': _safe_str(row.get('PRODUCTLINENAME')),
+                'WORKFLOW': _safe_str(row.get('WORKFLOW')),
+                'FINISHEDRUNCARD': _safe_str(row.get('FINISHEDRUNCARD')),
+                'TMTT_EQUIPMENTNAME': _safe_str(row.get('TMTT_EQUIPMENTNAME')),
+                'INPUT_QTY': input_qty,
+                'LOSS_REASON': '',
+                'DEFECT_QTY': 0,
+                'DEFECT_RATE': 0.0,
+                'ANCESTOR_COUNT': len(ancestor_set),
+                'UPSTREAM_MACHINES': ', '.join(sorted(upstream_machines)),
+            })
+
+    return result
diff --git a/src/mes_dashboard/sql/mid_section_defect/all_loss_reasons.sql b/src/mes_dashboard/sql/mid_section_defect/all_loss_reasons.sql
new file mode 100644
index 0000000..6a5da8e
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/all_loss_reasons.sql
@@ -0,0 +1,16 @@
+-- Mid-Section Defect - All Loss Reasons (cached daily)
+-- Lightweight query for filter dropdown population.
+-- Returns ALL loss reasons across all stations (not just TMTT).
+--
+-- Tables used:
+--   DWH.DW_MES_LOTREJECTHISTORY (TXNDATE indexed)
+--
+-- Performance:
+--   DISTINCT on one column with date filter only.
+--   Cached 24h in Redis.
+--
+SELECT DISTINCT r.LOSSREASONNAME
+FROM DWH.DW_MES_LOTREJECTHISTORY r
+WHERE r.TXNDATE >= SYSDATE - 180
+  AND r.LOSSREASONNAME IS NOT NULL
+ORDER BY r.LOSSREASONNAME
diff --git a/src/mes_dashboard/sql/mid_section_defect/genealogy_records.sql b/src/mes_dashboard/sql/mid_section_defect/genealogy_records.sql
new file mode 100644
index 0000000..bd878f2
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/genealogy_records.sql
@@ -0,0 +1,36 @@
+-- Mid-Section Defect Traceability - LOT Genealogy Records (Query 2)
+-- Batch query for split/merge records related to work orders
+--
+-- Parameters:
+--   MFG_ORDER_FILTER - Dynamic IN clause for MFGORDERNAME (built by QueryBuilder)
+--
+-- Tables used:
+--   DWH.DW_MES_CONTAINER (MFGORDERNAME indexed → get CONTAINERIDs)
+--   DWH.DW_MES_HM_LOTMOVEOUT (48M rows, no CONTAINERID index)
+--
+-- Performance:
+--   Full scan on HM_LOTMOVEOUT filtered by CONTAINERIDs from work orders.
+--   CDONAME filter reduces result set to only split/merge operations.
+--   Estimated 30-120s. Use aggressive caching (30-min TTL).
+--
+WITH work_order_lots AS (
+    SELECT CONTAINERID
+    FROM DWH.DW_MES_CONTAINER
+    WHERE {{ MFG_ORDER_FILTER }}
+)
+SELECT
+    h.CDONAME AS OPERATION_TYPE,
+    h.CONTAINERID AS TARGET_CID,
+    h.CONTAINERNAME AS TARGET_LOT,
+    h.FROMCONTAINERID AS SOURCE_CID,
+    h.FROMCONTAINERNAME AS SOURCE_LOT,
+    h.QTY,
+    h.TXNDATE
+FROM DWH.DW_MES_HM_LOTMOVEOUT h
+WHERE (
+    h.CONTAINERID IN (SELECT CONTAINERID FROM work_order_lots)
+    OR h.FROMCONTAINERID IN (SELECT CONTAINERID FROM work_order_lots)
+)
+  AND h.FROMCONTAINERID IS NOT NULL
+  AND (UPPER(h.CDONAME) LIKE '%SPLIT%' OR UPPER(h.CDONAME) LIKE '%COMBINE%')
+ORDER BY h.TXNDATE
diff --git a/src/mes_dashboard/sql/mid_section_defect/merge_lookup.sql b/src/mes_dashboard/sql/mid_section_defect/merge_lookup.sql
new file mode 100644
index 0000000..a643628
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/merge_lookup.sql
@@ -0,0 +1,21 @@
+-- Mid-Section Defect Traceability - Merge Lookup (Query 2b)
+-- Find source lots that were merged into finished lots
+-- via DW_MES_PJ_COMBINEDASSYLOTS
+--
+-- Parameters:
+--   Dynamically built IN clause for FINISHEDNAME values
+--
+-- Tables used:
+--   DWH.DW_MES_PJ_COMBINEDASSYLOTS (1.97M rows, FINISHEDNAME indexed)
+--
+-- Performance:
+--   FINISHEDNAME has index. Batch IN clause (up to 1000 per query).
+--   Each batch <1s.
+--
+SELECT
+    ca.CONTAINERID   AS SOURCE_CID,
+    ca.CONTAINERNAME AS SOURCE_NAME,
+    ca.FINISHEDNAME,
+    ca.LOTID         AS FINISHED_CID
+FROM DWH.DW_MES_PJ_COMBINEDASSYLOTS ca
+WHERE {{ FINISHED_NAME_FILTER }}
diff --git a/src/mes_dashboard/sql/mid_section_defect/split_chain.sql b/src/mes_dashboard/sql/mid_section_defect/split_chain.sql
new file mode 100644
index 0000000..6aecb20
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/split_chain.sql
@@ -0,0 +1,23 @@
+-- Mid-Section Defect Traceability - Split Chain (Query 2a)
+-- Resolve split ancestors via DW_MES_CONTAINER.SPLITFROMID
+--
+-- Parameters:
+--   Dynamically built IN clause for CONTAINERIDs
+--
+-- Tables used:
+--   DWH.DW_MES_CONTAINER (5.2M rows, CONTAINERID UNIQUE index)
+--
+-- Performance:
+--   CONTAINERID has UNIQUE index. Batch IN clause (up to 1000 per query).
+--   Each batch <1s.
+--
+-- Note: SPLITFROMID may be NULL for lots that were not split from another.
+--   BFS caller uses SPLITFROMID to walk upward; NULL means chain terminus.
+--
+SELECT
+    c.CONTAINERID,
+    c.SPLITFROMID,
+    c.ORIGINALCONTAINERID,
+    c.CONTAINERNAME
+FROM DWH.DW_MES_CONTAINER c
+WHERE {{ CID_FILTER }}
diff --git a/src/mes_dashboard/sql/mid_section_defect/tmtt_detection.sql b/src/mes_dashboard/sql/mid_section_defect/tmtt_detection.sql
new file mode 100644
index 0000000..ad47584
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/tmtt_detection.sql
@@ -0,0 +1,93 @@
+-- Mid-Section Defect Traceability - TMTT Detection Data (Query 1)
+-- Returns LOT-level data with TMTT input, ALL defects, and lot metadata
+--
+-- Parameters:
+--   :start_date - Start date (YYYY-MM-DD)
+--   :end_date   - End date (YYYY-MM-DD)
+--
+-- Tables used:
+--   DWH.DW_MES_LOTWIPHISTORY (TMTT station records)
+--   DWH.DW_MES_LOTREJECTHISTORY (defect records - ALL loss reasons)
+--   DWH.DW_MES_CONTAINER (product info + MFGORDERNAME for genealogy)
+--   DWH.DW_MES_WIP (WORKFLOWNAME)
+--
+-- Changes from tmtt_defect/base_data.sql:
+--   1. Removed hardcoded LOSSREASONNAME filter → fetches ALL loss reasons
+--   2. Added MFGORDERNAME from DW_MES_CONTAINER (needed for genealogy batch)
+--   3. Removed MOLD equipment lookup (upstream tracing done separately)
+--   4. Kept existing dedup logic (ROW_NUMBER by CONTAINERID, latest TRACKINTIMESTAMP)
+
+WITH tmtt_records AS (
+    SELECT /*+ MATERIALIZE */
+        h.CONTAINERID,
+        h.EQUIPMENTID AS TMTT_EQUIPMENTID,
+        h.EQUIPMENTNAME AS TMTT_EQUIPMENTNAME,
+        h.TRACKINQTY,
+        h.TRACKINTIMESTAMP,
+        h.TRACKOUTTIMESTAMP,
+        h.FINISHEDRUNCARD,
+        h.SPECNAME,
+        h.WORKCENTERNAME,
+        ROW_NUMBER() OVER (
+            PARTITION BY h.CONTAINERID
+            ORDER BY h.TRACKINTIMESTAMP DESC, h.TRACKOUTTIMESTAMP DESC NULLS LAST
+        ) AS rn
+    FROM DWH.DW_MES_LOTWIPHISTORY h
+    WHERE h.TRACKINTIMESTAMP >= TO_DATE(:start_date, 'YYYY-MM-DD')
+      AND h.TRACKINTIMESTAMP < TO_DATE(:end_date, 'YYYY-MM-DD') + 1
+      AND (UPPER(h.WORKCENTERNAME) LIKE '%TMTT%' OR h.WORKCENTERNAME LIKE '%測試%')
+      AND h.EQUIPMENTID IS NOT NULL
+      AND h.TRACKINTIMESTAMP IS NOT NULL
+),
+tmtt_deduped AS (
+    SELECT * FROM tmtt_records WHERE rn = 1
+),
+tmtt_rejects AS (
+    SELECT /*+ MATERIALIZE */
+        r.CONTAINERID,
+        r.LOSSREASONNAME,
+        SUM(NVL(r.REJECTQTY, 0) + NVL(r.STANDBYQTY, 0) + NVL(r.QTYTOPROCESS, 0)
+            + NVL(r.INPROCESSQTY, 0) + NVL(r.PROCESSEDQTY, 0)) AS REJECTQTY
+    FROM DWH.DW_MES_LOTREJECTHISTORY r
+    WHERE r.TXNDATE >= TO_DATE(:start_date, 'YYYY-MM-DD')
+      AND r.TXNDATE < TO_DATE(:end_date, 'YYYY-MM-DD') + 1
+      AND (UPPER(r.WORKCENTERNAME) LIKE '%TMTT%' OR r.WORKCENTERNAME LIKE '%測試%')
+    GROUP BY r.CONTAINERID, r.LOSSREASONNAME
+),
+lot_metadata AS (
+    SELECT /*+ MATERIALIZE */
+        c.CONTAINERID,
+        c.CONTAINERNAME,
+        c.MFGORDERNAME,
+        c.PJ_TYPE,
+        c.PRODUCTLINENAME
+    FROM DWH.DW_MES_CONTAINER c
+    WHERE c.CONTAINERID IN (SELECT CONTAINERID FROM tmtt_deduped)
+),
+workflow_info AS (
+    SELECT /*+ MATERIALIZE */
+        DISTINCT w.CONTAINERID,
+        w.WORKFLOWNAME
+    FROM DWH.DW_MES_WIP w
+    WHERE w.CONTAINERID IN (SELECT CONTAINERID FROM tmtt_deduped)
+      AND w.PRODUCTLINENAME <> '點測'
+)
+SELECT
+    t.CONTAINERID,
+    m.CONTAINERNAME,
+    m.MFGORDERNAME,
+    m.PJ_TYPE,
+    m.PRODUCTLINENAME,
+    NVL(wf.WORKFLOWNAME, t.SPECNAME) AS WORKFLOW,
+    t.FINISHEDRUNCARD,
+    t.TMTT_EQUIPMENTID,
+    t.TMTT_EQUIPMENTNAME,
+    t.TRACKINQTY,
+    t.TRACKINTIMESTAMP,
+    r.LOSSREASONNAME,
+    NVL(r.REJECTQTY, 0) AS REJECTQTY
+FROM tmtt_deduped t
+LEFT JOIN lot_metadata m ON t.CONTAINERID = m.CONTAINERID
+LEFT JOIN workflow_info wf ON t.CONTAINERID = wf.CONTAINERID
+LEFT JOIN tmtt_rejects r ON t.CONTAINERID = r.CONTAINERID
+ORDER BY t.TRACKINTIMESTAMP
diff --git a/src/mes_dashboard/sql/mid_section_defect/upstream_history.sql b/src/mes_dashboard/sql/mid_section_defect/upstream_history.sql
new file mode 100644
index 0000000..f7f6b20
--- /dev/null
+++ b/src/mes_dashboard/sql/mid_section_defect/upstream_history.sql
@@ -0,0 +1,40 @@
+-- Mid-Section Defect Traceability - Upstream Production History (Query 3)
+-- Get production history for ancestor LOTs at all stations
+--
+-- Parameters:
+--   Dynamically built IN clause for ancestor CONTAINERIDs
+--
+-- Tables used:
+--   DWH.DW_MES_LOTWIPHISTORY (53M rows, CONTAINERID indexed → fast)
+--
+-- Performance:
+--   CONTAINERID has index. Batch IN clause (up to 1000 per query).
+--   Estimated 1-5s per batch.
+--
+WITH ranked_history AS (
+    SELECT
+        h.CONTAINERID,
+        h.WORKCENTERNAME,
+        h.EQUIPMENTID,
+        h.EQUIPMENTNAME,
+        h.SPECNAME,
+        h.TRACKINTIMESTAMP,
+        ROW_NUMBER() OVER (
+            PARTITION BY h.CONTAINERID, h.WORKCENTERNAME, h.EQUIPMENTNAME
+            ORDER BY h.TRACKINTIMESTAMP DESC
+        ) AS rn
+    FROM DWH.DW_MES_LOTWIPHISTORY h
+    WHERE {{ ANCESTOR_FILTER }}
+      AND h.EQUIPMENTID IS NOT NULL
+      AND h.TRACKINTIMESTAMP IS NOT NULL
+)
+SELECT
+    CONTAINERID,
+    WORKCENTERNAME,
+    EQUIPMENTID,
+    EQUIPMENTNAME,
+    SPECNAME,
+    TRACKINTIMESTAMP
+FROM ranked_history
+WHERE rn = 1
+ORDER BY CONTAINERID, TRACKINTIMESTAMP