From d4d5d40569899d468c28b7bba4f2a0211da14dc3 Mon Sep 17 00:00:00 2001
From: caoxiaozhu <caoxiaozhu@users.noreply.github.com>
Date: Wed, 27 May 2026 17:31:27 +0800
Subject: [PATCH] =?UTF-8?q?feat:=20=E6=96=B0=E5=A2=9E=E9=A2=84=E7=AE=97?=
 =?UTF-8?q?=E8=B4=B9=E6=8E=A7=E6=A8=A1=E5=9E=8B=E4=B8=8E=E6=8A=A5=E9=94=80?=
 =?UTF-8?q?=E5=AE=A1=E6=89=B9=E6=B5=81=E5=BC=95=E6=93=8E?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

后端新增预算费控服务和报销单审批流模块，引入申请人费用画像
算法，优化知识库 RAG 运行时和同步逻辑，完善报销单工作流常
量和明细同步，更新差旅报销规则电子表格，前端新增预算分析
组件和数字员工模型，完善审批对话框和洞察面板交互，优化侧
边栏和顶栏样式，补充单元测试。
---
 .../index.html                                | 1472 +++++++++++++++++
 .../finance-rules/公司差旅费报销规则.xlsx     |  Bin 19825 -> 19827 bytes
 server/src/app/algorithem/README.md           |   21 +
 server/src/app/algorithem/__init__.py         |   13 +
 .../algorithem/applicant_expense_profile.py   |  445 +++++
 .../applicant_expense_profile_formula.md      |  224 +++
 .../app/api/v1/endpoints/reimbursements.py    |   36 +-
 server/src/app/schemas/budget.py              |   12 +
 server/src/app/services/agent_foundation.py   |    5 +-
 server/src/app/services/agent_runs.py         |    3 +-
 server/src/app/services/budget.py             |    4 +
 .../app/services/budget_expense_control.py    |  194 +++
 server/src/app/services/budget_support.py     |   17 +
 .../services/expense_claim_access_policy.py   |  188 ++-
 .../expense_claim_application_handoff.py      |    5 +-
 .../services/expense_claim_approval_flow.py   |  173 ++
 .../app/services/expense_claim_constants.py   |    1 +
 .../app/services/expense_claim_item_sync.py   |   13 +-
 .../app/services/expense_claim_read_model.py  |    2 +
 .../expense_claim_workflow_constants.py       |    6 +
 server/src/app/services/expense_claims.py     |  227 +--
 server/src/app/services/knowledge.py          |   10 +-
 server/src/app/services/knowledge_rag.py      |  115 +-
 .../src/app/services/knowledge_rag_runtime.py |    8 +-
 server/src/app/services/settings.py           |   13 +-
 server/src/app/services/user_agent.py         |   17 +-
 .../src/app/services/user_agent_knowledge.py  |  167 +-
 .../src/app/services/user_agent_response.py   |    9 +-
 server/storage/knowledge/.index.json          |  146 +-
 server/tests/test_agent_runs_service.py       |   21 +
 ...est_applicant_expense_profile_algorithm.py |   74 +
 server/tests/test_budget_endpoints.py         |   91 +
 server/tests/test_expense_claim_service.py    |  322 +++-
 server/tests/test_knowledge_rag_runtime.py    |   35 +
 server/tests/test_knowledge_rag_service.py    |    6 +-
 server/tests/test_knowledge_service.py        |   39 +
 server/tests/test_knowledge_sync.py           |   65 +
 server/tests/test_settings_persistence.py     |   18 +-
 server/tests/test_user_agent_service.py       |   44 +-
 .../assets/styles/components/sidebar-rail.css |   23 +
 web/src/assets/styles/components/top-bar.css  |    6 +
 .../travel-reimbursement-insight-panel.css    |   14 +
 web/src/assets/styles/detail-page-corners.css |  210 +++
 .../styles/views/documents-center-view.css    |    7 +-
 web/src/assets/styles/views/logs-view.css     |   37 +-
 web/src/assets/styles/views/policies-view.css |   95 +-
 .../styles/views/settings-view-form.css       |    6 +-
 web/src/components/layout/SidebarRail.vue     |   89 +-
 web/src/components/layout/TopBar.vue          |    2 +-
 .../components/shared/ReturnReasonDialog.vue  |   64 +-
 .../TravelReimbursementInsightPanel.vue       |    3 +-
 .../travel/TravelRequestApprovalDialog.vue    |   19 -
 .../travel/TravelRequestBudgetAnalysis.vue    |  293 ++++
 web/src/composables/useRequests.js            |   57 +-
 web/src/main.js                               |    1 +
 web/src/services/reimbursements.js            |    4 +
 web/src/utils/accessControl.js                |   45 +-
 web/src/utils/approvalInbox.js                |    5 +
 web/src/utils/detailAlerts.js                 |   77 +-
 web/src/views/DocumentsCenterView.vue         |   15 +
 web/src/views/LogsView.vue                    |    2 -
 web/src/views/PoliciesView.vue                |   33 +-
 web/src/views/TravelRequestDetailView.vue     |    9 +-
 .../views/scripts/EmployeeManagementView.js   |   10 +-
 web/src/views/scripts/PoliciesView.js         |   54 +-
 .../views/scripts/TravelRequestDetailView.js  |   70 +-
 .../scripts/auditViewDigitalEmployeeModel.js  |  263 +++
 web/src/views/scripts/auditViewMetadata.js    |   44 +
 web/src/views/scripts/auditViewModel.js       |   72 +-
 web/tests/accessControl.test.mjs              |   50 +
 web/tests/app-shell-detail-alerts.test.mjs    |   30 +-
 .../documents-center-status-filter.test.mjs   |    8 +-
 web/tests/requestProgressSteps.test.mjs       |   62 +-
 ...el-request-detail-leader-approval.test.mjs |   34 +-
 ...travel-request-detail-risk-advice.test.mjs |    5 +-
 75 files changed, 5393 insertions(+), 686 deletions(-)
 create mode 100644 document/development/budget-expense-control-model-plan/index.html
 create mode 100644 server/src/app/algorithem/README.md
 create mode 100644 server/src/app/algorithem/__init__.py
 create mode 100644 server/src/app/algorithem/applicant_expense_profile.py
 create mode 100644 server/src/app/algorithem/applicant_expense_profile_formula.md
 create mode 100644 server/src/app/services/budget_expense_control.py
 create mode 100644 server/src/app/services/expense_claim_approval_flow.py
 create mode 100644 server/src/app/services/expense_claim_workflow_constants.py
 create mode 100644 server/tests/test_applicant_expense_profile_algorithm.py
 create mode 100644 server/tests/test_knowledge_rag_runtime.py
 create mode 100644 server/tests/test_knowledge_sync.py
 create mode 100644 web/src/assets/styles/detail-page-corners.css
 create mode 100644 web/src/components/travel/TravelRequestBudgetAnalysis.vue
 create mode 100644 web/src/views/scripts/auditViewDigitalEmployeeModel.js

diff --git a/document/development/budget-expense-control-model-plan/index.html b/document/development/budget-expense-control-model-plan/index.html
new file mode 100644
index 0000000..8fbbb63
--- /dev/null
+++ b/document/development/budget-expense-control-model-plan/index.html
@@ -0,0 +1,1472 @@
+<!doctype html>
+<html lang="zh-CN">
+<head>
+  <meta charset="utf-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1">
+  <title>X-Financial 预算费用规划推荐模型方案</title>
+  <style>
+    :root {
+      color-scheme: light;
+      --bg: #f5f7f4;
+      --surface: #ffffff;
+      --surface-soft: #f9fbf8;
+      --ink: #132019;
+      --ink-soft: #526257;
+      --muted: #718075;
+      --line: #dbe4dc;
+      --line-strong: #c5d2c7;
+      --green: #0f766e;
+      --green-dark: #0b4f49;
+      --green-soft: #e4f5f2;
+      --blue: #2563eb;
+      --blue-soft: #e8f0ff;
+      --amber: #b45309;
+      --amber-soft: #fff3d7;
+      --red: #c2410c;
+      --red-soft: #fff0e8;
+      --plum: #7c2d12;
+      --plum-soft: #f8eee8;
+      --shadow: 0 16px 34px rgba(37, 48, 40, 0.08);
+      --radius: 8px;
+      --mono: "Cascadia Code", "JetBrains Mono", Consolas, monospace;
+      --font: "Microsoft YaHei UI", "Microsoft YaHei", "PingFang SC", "Segoe UI", sans-serif;
+    }
+
+    * {
+      box-sizing: border-box;
+    }
+
+    html {
+      scroll-behavior: smooth;
+    }
+
+    body {
+      margin: 0;
+      background:
+        linear-gradient(180deg, rgba(228, 245, 242, 0.9), rgba(245, 247, 244, 0.82) 310px),
+        var(--bg);
+      color: var(--ink);
+      font-family: var(--font);
+      font-size: 14px;
+      line-height: 1.68;
+      letter-spacing: 0;
+      overflow-wrap: anywhere;
+    }
+
+    a {
+      color: inherit;
+      text-decoration: none;
+    }
+
+    code {
+      padding: 2px 6px;
+      border-radius: 6px;
+      background: rgba(15, 118, 110, 0.08);
+      color: var(--green-dark);
+      font-family: var(--mono);
+      font-size: 0.92em;
+    }
+
+    pre {
+      margin: 0;
+      min-width: 0;
+      max-width: 100%;
+      padding: 16px;
+      overflow: auto;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: #101a16;
+      color: #eef8f4;
+      font-family: var(--mono);
+      font-size: 12.5px;
+      line-height: 1.72;
+    }
+
+    .layout {
+      display: grid;
+      grid-template-columns: 270px minmax(0, 1fr);
+      min-height: 100dvh;
+    }
+
+    .sidebar {
+      position: sticky;
+      top: 0;
+      height: 100dvh;
+      padding: 24px 18px;
+      border-right: 1px solid var(--line);
+      background: rgba(255, 255, 255, 0.9);
+      backdrop-filter: blur(18px);
+      overflow: auto;
+    }
+
+    .brand {
+      display: flex;
+      gap: 12px;
+      align-items: center;
+      padding-bottom: 22px;
+      border-bottom: 1px solid var(--line);
+    }
+
+    .brand-mark {
+      display: grid;
+      width: 40px;
+      height: 40px;
+      place-items: center;
+      border-radius: 8px;
+      background: #12342f;
+      color: #fff;
+      font-weight: 850;
+      box-shadow: 0 12px 20px rgba(15, 79, 73, 0.24);
+    }
+
+    .brand-title {
+      font-size: 15px;
+      font-weight: 850;
+      line-height: 1.25;
+    }
+
+    .brand-subtitle {
+      margin-top: 3px;
+      color: var(--muted);
+      font-size: 12px;
+    }
+
+    .nav-label {
+      margin: 22px 0 8px;
+      color: var(--muted);
+      font-size: 12px;
+      font-weight: 750;
+    }
+
+    .nav {
+      display: grid;
+      gap: 5px;
+    }
+
+    .nav a {
+      display: flex;
+      align-items: center;
+      min-height: 36px;
+      padding: 7px 10px;
+      border-radius: 8px;
+      color: var(--ink-soft);
+      font-size: 13px;
+      font-weight: 700;
+    }
+
+    .nav a:hover {
+      background: var(--green-soft);
+      color: var(--green-dark);
+    }
+
+    .nav-dot {
+      width: 7px;
+      height: 7px;
+      margin-right: 10px;
+      border-radius: 99px;
+      background: var(--line-strong);
+    }
+
+    .nav a:hover .nav-dot {
+      background: var(--green);
+    }
+
+    .side-note {
+      margin-top: 24px;
+      padding: 12px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface-soft);
+      color: var(--ink-soft);
+      font-size: 12px;
+      line-height: 1.55;
+    }
+
+    main {
+      min-width: 0;
+      padding: 34px 44px 58px;
+    }
+
+    .hero {
+      display: grid;
+      grid-template-columns: minmax(0, 1.1fr) minmax(330px, 0.9fr);
+      gap: 22px;
+      align-items: stretch;
+      margin-bottom: 22px;
+    }
+
+    .hero-copy {
+      min-width: 0;
+      padding: 30px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: rgba(255, 255, 255, 0.88);
+      box-shadow: var(--shadow);
+    }
+
+    .eyebrow {
+      display: inline-flex;
+      gap: 8px;
+      align-items: center;
+      margin-bottom: 14px;
+      color: var(--green-dark);
+      font-size: 12px;
+      font-weight: 850;
+      text-transform: uppercase;
+    }
+
+    .eyebrow::before {
+      content: "";
+      width: 24px;
+      height: 2px;
+      border-radius: 99px;
+      background: var(--green);
+    }
+
+    h1,
+    h2,
+    h3 {
+      margin: 0;
+      line-height: 1.28;
+      letter-spacing: 0;
+    }
+
+    h1 {
+      max-width: 760px;
+      color: #0b1f1b;
+      font-size: 34px;
+      font-weight: 900;
+    }
+
+    .lead {
+      max-width: 780px;
+      margin: 16px 0 0;
+      color: var(--ink-soft);
+      font-size: 15px;
+    }
+
+    .hero-actions {
+      display: flex;
+      flex-wrap: wrap;
+      gap: 10px;
+      margin-top: 22px;
+    }
+
+    .tag {
+      display: inline-flex;
+      align-items: center;
+      min-height: 30px;
+      padding: 5px 10px;
+      border: 1px solid var(--line);
+      border-radius: 999px;
+      background: var(--surface-soft);
+      color: var(--ink-soft);
+      font-size: 12px;
+      font-weight: 750;
+    }
+
+    .hero-panel {
+      display: grid;
+      gap: 10px;
+      min-width: 0;
+      padding: 18px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: #16342f;
+      color: #ecf8f5;
+      box-shadow: var(--shadow);
+    }
+
+    .signal-row {
+      display: flex;
+      justify-content: space-between;
+      gap: 14px;
+      padding: 12px;
+      border: 1px solid rgba(236, 248, 245, 0.14);
+      border-radius: 8px;
+      background: rgba(255, 255, 255, 0.06);
+    }
+
+    .signal-row strong {
+      display: block;
+      font-size: 13px;
+    }
+
+    .signal-row > div:first-child {
+      min-width: 0;
+    }
+
+    .signal-row span {
+      display: block;
+      margin-top: 3px;
+      color: rgba(236, 248, 245, 0.72);
+      font-size: 12px;
+    }
+
+    .signal-value {
+      flex: 0 0 auto;
+      color: #bff4e8;
+      font-weight: 900;
+    }
+
+    section {
+      min-width: 0;
+      margin-top: 18px;
+      padding: 24px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: rgba(255, 255, 255, 0.92);
+      box-shadow: 0 8px 20px rgba(37, 48, 40, 0.04);
+      scroll-margin-top: 20px;
+    }
+
+    .section-head {
+      display: flex;
+      align-items: flex-start;
+      justify-content: space-between;
+      gap: 22px;
+      margin-bottom: 16px;
+    }
+
+    .section-head h2 {
+      font-size: 22px;
+      font-weight: 900;
+    }
+
+    .section-head p {
+      max-width: 680px;
+      margin: 6px 0 0;
+      color: var(--ink-soft);
+    }
+
+    .section-kicker {
+      color: var(--green);
+      font-size: 12px;
+      font-weight: 850;
+      white-space: nowrap;
+    }
+
+    .grid-2,
+    .grid-3,
+    .grid-4 {
+      display: grid;
+      gap: 12px;
+    }
+
+    .grid-2 {
+      grid-template-columns: repeat(2, minmax(0, 1fr));
+    }
+
+    .grid-3 {
+      grid-template-columns: repeat(3, minmax(0, 1fr));
+    }
+
+    .grid-4 {
+      grid-template-columns: repeat(4, minmax(0, 1fr));
+    }
+
+    .card,
+    .metric,
+    .lane,
+    .phase,
+    .check-item {
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface);
+    }
+
+    .card {
+      min-width: 0;
+      padding: 16px;
+    }
+
+    .card h3,
+    .lane h3,
+    .phase h3 {
+      color: var(--ink);
+      font-size: 15px;
+      font-weight: 850;
+    }
+
+    .card p,
+    .lane p,
+    .phase p {
+      margin: 8px 0 0;
+      color: var(--ink-soft);
+      font-size: 13px;
+    }
+
+    .metric {
+      min-width: 0;
+      padding: 14px;
+      background: linear-gradient(180deg, #fff, var(--surface-soft));
+    }
+
+    .metric-label {
+      color: var(--muted);
+      font-size: 12px;
+      font-weight: 750;
+    }
+
+    .metric-value {
+      margin-top: 6px;
+      color: var(--ink);
+      font-size: 22px;
+      font-weight: 900;
+      line-height: 1.1;
+    }
+
+    .metric-desc {
+      margin-top: 7px;
+      color: var(--ink-soft);
+      font-size: 12px;
+      line-height: 1.5;
+    }
+
+    .pill-list {
+      display: flex;
+      flex-wrap: wrap;
+      gap: 8px;
+      margin-top: 12px;
+    }
+
+    .pill {
+      display: inline-flex;
+      align-items: center;
+      min-height: 28px;
+      padding: 4px 9px;
+      border-radius: 999px;
+      background: var(--green-soft);
+      color: var(--green-dark);
+      font-size: 12px;
+      font-weight: 750;
+    }
+
+    .flow {
+      display: grid;
+      grid-template-columns: repeat(5, minmax(0, 1fr));
+      gap: 10px;
+      margin-top: 14px;
+    }
+
+    .flow-step {
+      position: relative;
+      min-height: 120px;
+      padding: 14px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface-soft);
+    }
+
+    .flow-step:not(:last-child)::after {
+      content: "";
+      position: absolute;
+      top: 50%;
+      right: -9px;
+      width: 8px;
+      height: 8px;
+      border-top: 2px solid var(--line-strong);
+      border-right: 2px solid var(--line-strong);
+      transform: translateY(-50%) rotate(45deg);
+      background: transparent;
+    }
+
+    .flow-number {
+      display: inline-grid;
+      width: 24px;
+      height: 24px;
+      place-items: center;
+      margin-bottom: 9px;
+      border-radius: 6px;
+      background: var(--green);
+      color: #fff;
+      font-size: 12px;
+      font-weight: 900;
+    }
+
+    .flow-step strong {
+      display: block;
+      font-size: 13px;
+      line-height: 1.35;
+    }
+
+    .flow-step span {
+      display: block;
+      margin-top: 6px;
+      color: var(--ink-soft);
+      font-size: 12px;
+      line-height: 1.5;
+    }
+
+    .architecture {
+      display: grid;
+      gap: 10px;
+      margin-top: 12px;
+    }
+
+    .lane {
+      display: grid;
+      grid-template-columns: 160px minmax(0, 1fr);
+      gap: 14px;
+      padding: 14px;
+      align-items: start;
+    }
+
+    .lane h3 {
+      color: var(--green-dark);
+    }
+
+    .lane ul,
+    .card ul,
+    .phase ul,
+    .quality-list {
+      margin: 9px 0 0;
+      padding-left: 18px;
+      color: var(--ink-soft);
+    }
+
+    li + li {
+      margin-top: 5px;
+    }
+
+    .formula-grid {
+      display: grid;
+      grid-template-columns: minmax(0, 1fr) minmax(300px, 0.8fr);
+      gap: 14px;
+      align-items: start;
+    }
+
+    .callout {
+      padding: 16px;
+      border: 1px solid var(--line);
+      border-left: 4px solid var(--green);
+      border-radius: var(--radius);
+      background: var(--green-soft);
+      color: var(--green-dark);
+    }
+
+    .callout strong {
+      display: block;
+      margin-bottom: 6px;
+      font-weight: 900;
+    }
+
+    .score-band {
+      display: grid;
+      gap: 9px;
+    }
+
+    .band {
+      display: grid;
+      grid-template-columns: 120px 82px minmax(0, 1fr);
+      gap: 10px;
+      align-items: center;
+      padding: 11px 12px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface-soft);
+    }
+
+    .band strong {
+      font-size: 13px;
+    }
+
+    .band-score {
+      font-family: var(--mono);
+      font-size: 12px;
+      font-weight: 850;
+    }
+
+    .band span {
+      color: var(--ink-soft);
+      font-size: 12px;
+    }
+
+    .band.recommended {
+      border-color: #a6ded3;
+      background: var(--green-soft);
+    }
+
+    .band.caution {
+      border-color: #f0cf88;
+      background: var(--amber-soft);
+    }
+
+    .band.review {
+      border-color: #b8c9f5;
+      background: var(--blue-soft);
+    }
+
+    .band.block {
+      border-color: #efb69a;
+      background: var(--red-soft);
+    }
+
+    .phase {
+      min-width: 0;
+      padding: 16px;
+      border-top: 4px solid var(--green);
+    }
+
+    .phase:nth-child(2) {
+      border-top-color: var(--blue);
+    }
+
+    .phase:nth-child(3) {
+      border-top-color: var(--amber);
+    }
+
+    .phase:nth-child(4) {
+      border-top-color: var(--red);
+    }
+
+    .screen {
+      display: grid;
+      grid-template-columns: 1fr 1fr;
+      gap: 12px;
+      padding: 14px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: #f8faf8;
+    }
+
+    .screen-block {
+      min-width: 0;
+      min-height: 74px;
+      padding: 12px;
+      border: 1px solid var(--line);
+      border-radius: var(--radius);
+      background: var(--surface);
+    }
+
+    .screen-block.wide {
+      grid-column: 1 / -1;
+    }
+
+    .screen-block strong {
+      display: block;
+      font-size: 13px;
+    }
+
+    .screen-block span {
+      display: block;
+      margin-top: 5px;
+      color: var(--ink-soft);
+      font-size: 12px;
+    }
+
+    .checklist {
+      display: grid;
+      gap: 9px;
+      margin-top: 12px;
+    }
+
+    .check-item {
+      display: flex;
+      gap: 10px;
+      align-items: flex-start;
+      padding: 12px;
+      color: var(--ink-soft);
+    }
+
+    .check-item input {
+      width: 16px;
+      height: 16px;
+      margin-top: 4px;
+      accent-color: var(--green);
+    }
+
+    .check-item strong {
+      display: block;
+      color: var(--ink);
+      font-size: 13px;
+    }
+
+    .check-item span {
+      display: block;
+      margin-top: 3px;
+      font-size: 12px;
+    }
+
+    .check-item:has(input:checked) {
+      background: var(--green-soft);
+      color: var(--green-dark);
+    }
+
+    .check-item:has(input:checked) strong,
+    .check-item:has(input:checked) span {
+      text-decoration: line-through;
+      text-decoration-thickness: 1px;
+    }
+
+    .split-note {
+      display: grid;
+      grid-template-columns: 1fr 1fr;
+      gap: 12px;
+      margin-top: 12px;
+    }
+
+    .note-green,
+    .note-amber {
+      padding: 14px;
+      border-radius: var(--radius);
+      font-size: 13px;
+    }
+
+    .note-green {
+      border: 1px solid #a6ded3;
+      background: var(--green-soft);
+      color: var(--green-dark);
+    }
+
+    .note-amber {
+      border: 1px solid #f0cf88;
+      background: var(--amber-soft);
+      color: #74460b;
+    }
+
+    .footer {
+      margin-top: 24px;
+      color: var(--muted);
+      font-size: 12px;
+      text-align: center;
+    }
+
+    @media (max-width: 1100px) {
+      .layout {
+        grid-template-columns: 1fr;
+      }
+
+      .sidebar {
+        position: static;
+        height: auto;
+      }
+
+      .nav {
+        grid-template-columns: repeat(2, minmax(0, 1fr));
+      }
+
+      .hero,
+      .grid-4,
+      .flow,
+      .formula-grid {
+        grid-template-columns: 1fr;
+      }
+
+      .flow-step:not(:last-child)::after {
+        display: none;
+      }
+    }
+
+    @media (max-width: 760px) {
+      main {
+        padding: 22px 14px 38px;
+      }
+
+      .sidebar {
+        padding: 16px 14px;
+      }
+
+      .brand {
+        padding-bottom: 14px;
+      }
+
+      .nav-label {
+        margin: 14px 0 8px;
+      }
+
+      .nav {
+        grid-template-columns: repeat(2, minmax(0, 1fr));
+        gap: 4px 8px;
+      }
+
+      .nav a {
+        min-height: 32px;
+        padding: 6px 8px;
+        font-size: 12px;
+      }
+
+      .side-note {
+        margin-top: 14px;
+        padding: 10px;
+      }
+
+      .signal-row {
+        display: grid;
+        grid-template-columns: 1fr;
+        gap: 4px;
+      }
+
+      .signal-value {
+        justify-self: start;
+      }
+
+      .hero-copy,
+      section {
+        padding: 18px;
+      }
+
+      h1 {
+        font-size: 26px;
+      }
+
+      .grid-2,
+      .grid-3,
+      .split-note,
+      .screen,
+      .lane,
+      .band {
+        grid-template-columns: 1fr;
+      }
+    }
+  </style>
+</head>
+<body>
+  <div class="layout">
+    <aside class="sidebar">
+      <div class="brand">
+        <div class="brand-mark">BF</div>
+        <div>
+          <div class="brand-title">预算费用模型方案</div>
+          <div class="brand-subtitle">Budget & Fee Recommendation</div>
+        </div>
+      </div>
+
+      <div class="nav-label">方案目录</div>
+      <nav class="nav" aria-label="页面目录">
+        <a href="#summary"><span class="nav-dot"></span>方案结论</a>
+        <a href="#loop"><span class="nav-dot"></span>业务闭环</a>
+        <a href="#architecture"><span class="nav-dot"></span>模型架构</a>
+        <a href="#features"><span class="nav-dot"></span>数据与特征</a>
+        <a href="#formula"><span class="nav-dot"></span>计算口径</a>
+        <a href="#score"><span class="nav-dot"></span>评分推荐</a>
+        <a href="#applicant-profile"><span class="nav-dot"></span>画像公式</a>
+        <a href="#output"><span class="nav-dot"></span>输出协议</a>
+        <a href="#ui"><span class="nav-dot"></span>审批展示</a>
+        <a href="#roadmap"><span class="nav-dot"></span>落地路线</a>
+        <a href="#acceptance"><span class="nav-dot"></span>验收清单</a>
+      </nav>
+
+      <div class="side-note">
+        版本：2026-05-27<br>
+        定位：在原“预算费用管控评分模型”基础上，升级为可上线评审的规划推荐模型方案。
+      </div>
+    </aside>
+
+    <main>
+      <header class="hero">
+        <div class="hero-copy">
+          <div class="eyebrow">X-Financial Model Proposal</div>
+          <h1>预算费用规划推荐模型方案</h1>
+          <p class="lead">
+            这个模型不只是给审批页打一个分，而是把“预算是否够、费用是否合理、是否影响预算节奏、下一步该怎么处理”整理成可解释、可审计、可持续迭代的推荐结果。
+          </p>
+          <div class="hero-actions">
+            <span class="tag">规则模型优先</span>
+            <span class="tag">硬约束可审计</span>
+            <span class="tag">LLM 只做解释层</span>
+            <span class="tag">审批反馈可回流</span>
+          </div>
+        </div>
+
+        <div class="hero-panel" aria-label="关键决策信号">
+          <div class="signal-row">
+            <div>
+              <strong>模型主目标</strong>
+              <span>帮预算管理者快速判断是否可承接</span>
+            </div>
+            <div class="signal-value">Recommend</div>
+          </div>
+          <div class="signal-row">
+            <div>
+              <strong>第一版边界</strong>
+              <span>不替代人工审批，不让 LLM 直接做硬判断</span>
+            </div>
+            <div class="signal-value">Guarded</div>
+          </div>
+          <div class="signal-row">
+            <div>
+              <strong>核心口径</strong>
+              <span>扣除当前申请已预占，避免重复计算</span>
+            </div>
+            <div class="signal-value">Traceable</div>
+          </div>
+          <div class="signal-row">
+            <div>
+              <strong>迭代方向</strong>
+              <span>规则评分 + 历史基准 + 解释生成</span>
+            </div>
+            <div class="signal-value">V1 → V3</div>
+          </div>
+        </div>
+      </header>
+
+      <section id="summary">
+        <div class="section-head">
+          <div>
+            <h2>方案结论</h2>
+            <p>建议把模型定位为“预算审批辅助决策引擎”，输出结构化推荐和解释依据，而不是单纯的页面分数。</p>
+          </div>
+          <div class="section-kicker">Executive Summary</div>
+        </div>
+
+        <div class="grid-4">
+          <div class="metric">
+            <div class="metric-label">推荐动作</div>
+            <div class="metric-value">4 类</div>
+            <div class="metric-desc">建议通过、谨慎通过、需要复核、不建议直接通过。</div>
+          </div>
+          <div class="metric">
+            <div class="metric-label">核心评分</div>
+            <div class="metric-value">100 分</div>
+            <div class="metric-desc">预算安全优先，费用必要性与信息完整度共同影响。</div>
+          </div>
+          <div class="metric">
+            <div class="metric-label">硬性约束</div>
+            <div class="metric-value">3 条</div>
+            <div class="metric-desc">超预算、强阻断控制、预算池无法匹配必须显式处理。</div>
+          </div>
+          <div class="metric">
+            <div class="metric-label">上线策略</div>
+            <div class="metric-value">分阶段</div>
+            <div class="metric-desc">先确定性规则，再接入历史基准和解释层。</div>
+          </div>
+        </div>
+
+        <div class="split-note">
+          <div class="note-green">
+            <strong>推荐采用的产品口径：</strong>
+            分数不是结论本身，结论应由“推荐动作 + 风险等级 + 触发依据 + 可执行建议”共同表达。
+          </div>
+          <div class="note-amber">
+            <strong>需要避免的方向：</strong>
+            不要让 LLM 直接判断是否通过，也不要只用一个综合分覆盖预算、业务必要性、历史异常等不同原因。
+          </div>
+        </div>
+      </section>
+
+      <section id="loop">
+        <div class="section-head">
+          <div>
+            <h2>业务闭环</h2>
+            <p>模型要服务完整预算链路：预算池建立、申请预占、审批推荐、报销核销、结果反馈。</p>
+          </div>
+          <div class="section-kicker">Business Loop</div>
+        </div>
+
+        <div class="flow">
+          <div class="flow-step">
+            <span class="flow-number">1</span>
+            <strong>预算计划建立</strong>
+            <span>按部门、项目、科目、成本中心形成预算池和预警线。</span>
+          </div>
+          <div class="flow-step">
+            <span class="flow-number">2</span>
+            <strong>费用申请预占</strong>
+            <span>申请提交后先占用预算，避免后续审批期间额度被重复使用。</span>
+          </div>
+          <div class="flow-step">
+            <span class="flow-number">3</span>
+            <strong>预算审批推荐</strong>
+            <span>预算管理者看到风险等级、计算依据、建议动作和补充要求。</span>
+          </div>
+          <div class="flow-step">
+            <span class="flow-number">4</span>
+            <strong>报销与核销</strong>
+            <span>申请通过后生成报销草稿，实际报销后转为已核销金额。</span>
+          </div>
+          <div class="flow-step">
+            <span class="flow-number">5</span>
+            <strong>结果反馈回流</strong>
+            <span>记录审批采纳、驳回原因、实际报销差异，反哺后续规则。</span>
+          </div>
+        </div>
+      </section>
+
+      <section id="architecture">
+        <div class="section-head">
+          <div>
+            <h2>模型架构</h2>
+            <p>采用“上下文构建 → 特征计算 → 硬约束判定 → 综合评分 → 推荐解释 → 反馈沉淀”的分层结构。</p>
+          </div>
+          <div class="section-kicker">Model Layers</div>
+        </div>
+
+        <div class="architecture">
+          <div class="lane">
+            <h3>上下文层</h3>
+            <div>
+              <p>统一组装申请单、预算池、历史费用、项目计划和审批身份，形成模型输入快照。</p>
+              <ul>
+                <li>复用 <code>build_claim_budget_context()</code> 的预算上下文。</li>
+                <li>补充申请事由、地点、费用类型、项目编号、附件摘要。</li>
+                <li>保留计算时刻的预算快照，方便事后审计。</li>
+              </ul>
+            </div>
+          </div>
+
+          <div class="lane">
+            <h3>特征层</h3>
+            <div>
+              <p>把业务数据转换成可解释特征，而不是直接把原始字段丢给模型。</p>
+              <ul>
+                <li>预算容量：审批前可用、审批后使用率、超预算金额。</li>
+                <li>单笔影响：本次金额占预算比例、是否突破预算节奏。</li>
+                <li>必要性证据：事由、项目、地点、附件、业务场景是否完整。</li>
+                <li>历史基准：同部门、同项目、同费用类型的历史均值和波动。</li>
+              </ul>
+            </div>
+          </div>
+
+          <div class="lane">
+            <h3>决策层</h3>
+            <div>
+              <p>先执行硬规则，再计算综合分，确保强管控场景不会被其他维度“平均掉”。</p>
+              <ul>
+                <li>硬规则：超预算、预算池阻断、预算无法匹配、关键字段缺失。</li>
+                <li>综合评分：预算安全、费用影响、规划匹配、历史异常、信息完整。</li>
+                <li>分档输出：recommended、caution、review、block、reference。</li>
+              </ul>
+            </div>
+          </div>
+
+          <div class="lane">
+            <h3>解释层</h3>
+            <div>
+              <p>第一版直接用模板解释；后续允许 LLM 改写语气，但不能改写分数、等级和硬性结论。</p>
+              <ul>
+                <li>解释必须引用结构化依据，不允许凭空补充原因。</li>
+                <li>建议要能变成审批动作，例如补充材料、拆分费用、调整预算。</li>
+                <li>所有展示文案保留对应的 <code>basis_code</code>，方便追踪。</li>
+              </ul>
+            </div>
+          </div>
+        </div>
+      </section>
+
+      <section id="features">
+        <div class="section-head">
+          <div>
+            <h2>数据与特征</h2>
+            <p>模型输入需要覆盖“预算是否够”和“费用是否该花”两条线，避免只看额度、不看业务必要性。</p>
+          </div>
+          <div class="section-kicker">Feature Design</div>
+        </div>
+
+        <div class="grid-3">
+          <div class="card">
+            <h3>预算上下文</h3>
+            <ul>
+              <li><code>budget_applicable</code>：是否纳入预算管控。</li>
+              <li><code>matched</code>：是否匹配到预算池。</li>
+              <li><code>total_amount</code>、<code>reserved_amount</code>、<code>consumed_amount</code>。</li>
+              <li><code>current_reserved_amount</code>：当前申请已预占金额。</li>
+              <li><code>warning_threshold</code>、<code>control_action</code>。</li>
+            </ul>
+          </div>
+
+          <div class="card">
+            <h3>申请单主数据</h3>
+            <ul>
+              <li>申请金额、费用类型、申请事由。</li>
+              <li>业务地点、项目编号、成本中心。</li>
+              <li>直属领导意见、附件摘要。</li>
+              <li>申请时间、期望发生时间、是否紧急。</li>
+            </ul>
+          </div>
+
+          <div class="card">
+            <h3>增强特征</h3>
+            <ul>
+              <li>同类费用历史均值与分位数。</li>
+              <li>部门剩余预算消耗速度。</li>
+              <li>项目里程碑与预算计划匹配度。</li>
+              <li>申请人近期预算占用频次。</li>
+            </ul>
+          </div>
+        </div>
+
+        <div class="pill-list" aria-label="特征分组">
+          <span class="pill">预算容量</span>
+          <span class="pill">单笔影响</span>
+          <span class="pill">预算节奏</span>
+          <span class="pill">业务必要性</span>
+          <span class="pill">历史异常</span>
+          <span class="pill">信息完整度</span>
+          <span class="pill">审批反馈</span>
+        </div>
+      </section>
+
+      <section id="formula">
+        <div class="section-head">
+          <div>
+            <h2>计算口径</h2>
+            <p>申请提交时已经发生预算预占，所以审批分析必须扣除当前申请已预占金额，避免重复计算。</p>
+          </div>
+          <div class="section-kicker">Calculation</div>
+        </div>
+
+        <div class="formula-grid">
+          <pre><code>已使用基数 = 已预占金额 + 已核销金额 - 当前申请已预占金额
+审批前可用预算 = 预算总额度 - 已使用基数
+审批后占用 = 已使用基数 + 本次申请金额
+此次费用占预算 = 本次申请金额 / 预算总额度
+审批后使用率 = 审批后占用 / 预算总额度
+超预算金额 = max(本次申请金额 - 审批前可用预算, 0)</code></pre>
+
+          <div class="callout">
+            <strong>关键原则</strong>
+            如果当前申请在提交阶段已经进入 <code>reserved_amount</code>，审批页就不能再把它当成新增占用直接叠加。否则预算管理者看到的风险会被放大，尤其是大额申请会明显失真。
+          </div>
+        </div>
+      </section>
+
+      <section id="score">
+        <div class="section-head">
+          <div>
+            <h2>评分推荐</h2>
+            <p>建议采用“硬规则优先 + 加权评分”的组合。硬规则决定是否必须复核或阻断，评分用于排序和解释风险程度。</p>
+          </div>
+          <div class="section-kicker">Decision Strategy</div>
+        </div>
+
+        <div class="grid-2">
+          <div class="card">
+            <h3>建议权重</h3>
+            <ul>
+              <li>预算安全：40 分，覆盖可用额度、审批后使用率、超预算金额。</li>
+              <li>单笔影响：18 分，判断本次费用对预算池的冲击。</li>
+              <li>规划匹配：16 分，比较项目计划、预算周期和费用发生时间。</li>
+              <li>历史基准：14 分，识别同类费用异常偏高或频繁占用。</li>
+              <li>信息完整：12 分，衡量事由、项目、地点、附件是否足够。</li>
+            </ul>
+          </div>
+
+          <div class="score-band">
+            <div class="band recommended">
+              <strong>recommended</strong>
+              <div class="band-score">85-100</div>
+              <span>预算充足、信息完整、费用影响可承接，建议通过。</span>
+            </div>
+            <div class="band caution">
+              <strong>caution</strong>
+              <div class="band-score">70-84</div>
+              <span>预算可承接但影响较明显，建议谨慎通过并关注后续节奏。</span>
+            </div>
+            <div class="band review">
+              <strong>review</strong>
+              <div class="band-score">50-69</div>
+              <span>存在信息缺口、历史异常或接近预警线，需要复核。</span>
+            </div>
+            <div class="band block">
+              <strong>block</strong>
+              <div class="band-score">0-49</div>
+              <span>超预算或触发强管控动作，不建议直接通过。</span>
+            </div>
+          </div>
+        </div>
+
+        <div class="split-note">
+          <div class="note-green">
+            <strong>硬规则示例：</strong>
+            <code>control_action = block</code> 且超预算时，直接进入 <code>block</code>；预算池未匹配时输出 <code>reference</code>，不能伪装成低风险。
+          </div>
+          <div class="note-amber">
+            <strong>推荐动作要具体：</strong>
+            不只写“需关注”，而要给出“补充业务必要性、拆分费用、调整预算池、发起预算调剂”等可执行动作。
+          </div>
+        </div>
+      </section>
+
+      <section id="applicant-profile">
+        <div class="section-head">
+          <div>
+            <h2>申请人费用画像公式</h2>
+            <p>画像只用于调整复核强度和审核建议，不能直接把申请人定义为“异常人员”。核心是用同组基准解释个人费用节奏是否偏离。</p>
+          </div>
+          <div class="section-kicker">Applicant Profile</div>
+        </div>
+
+        <div class="formula-grid">
+          <pre><code>申请人画像风险分 =
+  高频申请分 * 20%
++ 高金额占用分 * 25%
++ 同组偏离分 * 25%
++ 历史调减退回分 * 15%
++ 本次申请偏离分 * 15%
+
+等级：
+0-39   正常
+40-59  需关注
+60-79  重点复核
+80-100 强复核 / 升级审批</code></pre>
+
+          <div class="callout">
+            <strong>同组基准口径</strong>
+            同组不按全公司粗暴比较，而应按 <code>部门 + 岗位 + 费用类型 + 城市等级 + 项目类型 + 近 90/180 天</code> 聚合。销售、项目经理和研发不能混在一个基准池里比较。
+          </div>
+        </div>
+
+        <div class="grid-3" style="margin-top: 12px;">
+          <div class="card">
+            <h3>出差天数建议</h3>
+            <pre><code>同类基准天数 = peer_group.travel_days_p75
+天数偏离率 = 本次出差天数 / 同类基准天数
+建议天数 = min(
+  本次出差天数,
+  同类基准天数 + 业务缓冲天数
+)</code></pre>
+            <p>当偏离率超过 1.5 时，建议补充行程安排或压缩天数；超过 2.0 时进入重点复核。</p>
+          </div>
+
+          <div class="card">
+            <h3>出差费用建议</h3>
+            <pre><code>本次日均费用 = 本次申请金额 / 本次出差天数
+日均费用偏离率 =
+  本次日均费用 / 同城市同职级日均基准
+建议金额上限 =
+  建议天数 * 日均基准 * 容忍系数</code></pre>
+            <p>容忍系数第一版建议使用 1.15 到 1.20，避免模型因为小幅波动给出过硬建议。</p>
+          </div>
+
+          <div class="card">
+            <h3>招待费用建议</h3>
+            <pre><code>人均招待金额 = 本次招待金额 / 参与人数
+人均偏离率 = 人均招待金额 / 招待标准上限
+同客户招待频率 = 近 90 天同客户招待次数
+个人招待分位 = 个人近 90 天招待金额在同组分位</code></pre>
+            <p>人均偏离率超过 1.2 时建议调低标准；同客户 90 天内多次招待时要求补充客户推进阶段。</p>
+          </div>
+        </div>
+
+        <div class="split-note">
+          <div class="note-green">
+            <strong>建议生成公式：</strong>
+            <code>审核建议强度 = max(画像风险等级, 本次偏离等级, 硬规则等级)</code>。展示时必须说明是哪一类指标触发，而不是只给一个结论。
+          </div>
+          <div class="note-amber">
+            <strong>审核建议示例：</strong>
+            “该申请人近 90 天费用占用处于同组 P88，本次出差天数为同类 P75 的 1.67 倍。建议将 5 天调整为 4 天，或补充客户拜访安排和项目阶段说明。”
+          </div>
+        </div>
+      </section>
+
+      <section id="output">
+        <div class="section-head">
+          <div>
+            <h2>输出协议</h2>
+            <p>接口输出需要同时服务审批页展示、看板统计和后续审计，因此自然语言必须和结构化字段分开。</p>
+          </div>
+          <div class="section-kicker">API Contract</div>
+        </div>
+
+        <pre><code>{
+  "score": 82,
+  "rating": "caution",
+  "risk_level": "medium",
+  "recommendation": "cautious_approve",
+  "summary": "预算整体可承接，但本次费用对预算池已有明显影响。",
+  "metrics": {
+    "claim_amount": "12000.00",
+    "total_amount": "50000.00",
+    "available_before_approval": "38000.00",
+    "claim_amount_ratio": "24.00",
+    "after_usage_rate": "72.00",
+    "over_budget_amount": "0.00"
+  },
+  "applicant_profile": {
+    "profile_score": 66,
+    "profile_level": "review",
+    "peer_percentile": 88,
+    "travel_days_ratio": "1.67",
+    "daily_cost_ratio": "1.34",
+    "suggested_days": 4,
+    "suggested_amount_cap": "6800.00"
+  },
+  "evidence": [
+    {
+      "basis_code": "budget.after_usage_rate.warning_near",
+      "text": "审批后预算使用率为 72.00%，接近预警区间。"
+    }
+  ],
+  "suggested_actions": [
+    {
+      "action": "approve_with_note",
+      "text": "可继续审批，但建议备注本次费用对应的项目阶段和后续预算节奏。"
+    }
+  ],
+  "explainable_snapshot": {
+    "budget_no": "BUD-TEST",
+    "control_action": "warn",
+    "warning_threshold": "80.00"
+  }
+}</code></pre>
+      </section>
+
+      <section id="ui">
+        <div class="section-head">
+          <div>
+            <h2>审批展示</h2>
+            <p>预算管理者最需要看到的是“为什么是这个建议”和“如果要通过，还要补什么”。页面不应只展示一个分数。</p>
+          </div>
+          <div class="section-kicker">Approval UX</div>
+        </div>
+
+        <div class="screen">
+          <div class="screen-block wide">
+            <strong>顶部结论条</strong>
+            <span>展示推荐动作、风险等级、综合分和一句话摘要，例如“谨慎通过：预算可承接，但审批后使用率接近预警线”。</span>
+          </div>
+          <div class="screen-block">
+            <strong>关键指标</strong>
+            <span>本次金额、预算总额、审批前可用、审批后使用率、超预算金额。</span>
+          </div>
+          <div class="screen-block">
+            <strong>触发依据</strong>
+            <span>逐条展示规则依据，保留 basis_code，方便定位到模型逻辑。</span>
+          </div>
+          <div class="screen-block">
+            <strong>建议动作</strong>
+            <span>通过、谨慎通过、补充材料、拆分费用、预算调剂、退回修改。</span>
+          </div>
+          <div class="screen-block">
+            <strong>申请人费用画像</strong>
+            <span>展示近 90 天频率、金额分位、天数偏离率和建议压缩区间，只作为审批参考。</span>
+          </div>
+          <div class="screen-block">
+            <strong>审批反馈</strong>
+            <span>记录预算管理者是否采纳建议，以及不采纳时的原因。</span>
+          </div>
+        </div>
+      </section>
+
+      <section id="roadmap">
+        <div class="section-head">
+          <div>
+            <h2>落地路线</h2>
+            <p>先把确定性口径做稳，再逐步接入历史基准和解释层，避免第一版就变成难以审计的黑盒。</p>
+          </div>
+          <div class="section-kicker">Implementation</div>
+        </div>
+
+        <div class="grid-4">
+          <div class="phase">
+            <h3>P0：口径对齐</h3>
+            <p>确认预算预占、审批前可用、审批后占用的计算口径。</p>
+            <ul>
+              <li>梳理预算上下文字段。</li>
+              <li>固化不重复计算规则。</li>
+              <li>定义输出协议。</li>
+            </ul>
+          </div>
+          <div class="phase">
+            <h3>P1：规则评分</h3>
+            <p>上线可审计的确定性模型，覆盖预算审批主流程。</p>
+            <ul>
+              <li>实现特征计算。</li>
+              <li>实现硬规则和分档。</li>
+              <li>补齐单元测试。</li>
+            </ul>
+          </div>
+          <div class="phase">
+            <h3>P2：历史基准</h3>
+            <p>引入同类费用历史均值、部门消耗速度和预算节奏。</p>
+            <ul>
+              <li>沉淀历史统计任务。</li>
+              <li>增加异常检测特征。</li>
+              <li>看板展示采纳率。</li>
+            </ul>
+          </div>
+          <div class="phase">
+            <h3>P3：解释增强</h3>
+            <p>在结构化依据上增加 LLM 文案层，提升可读性。</p>
+            <ul>
+              <li>LLM 不改分数和等级。</li>
+              <li>解释引用 basis_code。</li>
+              <li>增加人工反馈闭环。</li>
+            </ul>
+          </div>
+        </div>
+      </section>
+
+      <section id="acceptance">
+        <div class="section-head">
+          <div>
+            <h2>验收清单</h2>
+            <p>下面的清单会保存在当前浏览器本地，方便后续评审和开发跟踪。</p>
+          </div>
+          <div class="section-kicker">Checklist</div>
+        </div>
+
+        <div class="checklist">
+          <label class="check-item">
+            <input type="checkbox" data-check="contract">
+            <span>
+              <strong>输出协议稳定</strong>
+              <span>评分、等级、推荐动作、指标、依据、建议动作都有结构化字段。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="double-count">
+            <span>
+              <strong>预算预占不重复计算</strong>
+              <span>测试覆盖当前申请已预占、未预占、部分预占三类场景。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="hard-rules">
+            <span>
+              <strong>硬规则不可被平均抵消</strong>
+              <span>超预算和 block 控制动作必须进入高风险或阻断建议。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="ui">
+            <span>
+              <strong>审批页展示可解释</strong>
+              <span>预算管理者能看到风险原因、计算指标和可执行下一步。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="applicant-profile">
+            <span>
+              <strong>申请人画像公式可审计</strong>
+              <span>同组基准、偏离率、建议天数、建议金额上限都能追溯到结构化字段。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="feedback">
+            <span>
+              <strong>审批反馈可回流</strong>
+              <span>记录采纳、覆盖、退回、预算调剂等结果，为后续模型迭代准备数据。</span>
+            </span>
+          </label>
+          <label class="check-item">
+            <input type="checkbox" data-check="llm-boundary">
+            <span>
+              <strong>LLM 边界清晰</strong>
+              <span>LLM 只改写解释文案，不参与硬性通过、阻断和预算额度计算。</span>
+            </span>
+          </label>
+        </div>
+      </section>
+
+      <div class="footer">
+        X-Financial · 预算费用规划推荐模型方案 · 适用于预算审批、费用申请和报销核销闭环。
+      </div>
+    </main>
+  </div>
+
+  <script>
+    const storagePrefix = "x-financial-budget-model-plan:";
+
+    document.querySelectorAll("[data-check]").forEach((checkbox) => {
+      const key = storagePrefix + checkbox.dataset.check;
+      checkbox.checked = localStorage.getItem(key) === "1";
+      checkbox.addEventListener("change", () => {
+        localStorage.setItem(key, checkbox.checked ? "1" : "0");
+      });
+    });
+  </script>
+</body>
+</html>
diff --git a/server/rules/finance-rules/公司差旅费报销规则.xlsx b/server/rules/finance-rules/公司差旅费报销规则.xlsx
index cca003bdf4bb317d4b495eb4a0046ecf1cc41af1..22ca3d855043fd4a858f8bcf738c134524d5b0bb 100644
GIT binary patch
delta 1523
zcmZ8hdpMM76#wStw#_mzu7fcop=YK{qOF8AH5V;X4dohLsHWU+HH=#k&H74WRjAO#
zC6b7;LSrUVq=v~PLs+F0ZLwo^=G)!p*?RwZfA8;{^PcmZ^Lx%rA?#Bj4DCTjAawxc
zNuf2@q61VBio1@b8VQHMn0hfnUBT==lyKX%8q7xH*urpaXPqwj+2}`W226(zHkXLK
zj8cCsMTIBYf<x4piD^c`d2%HouF_5O4|fa_l^s{(uhEhGMMyp%BPKQT?5XBg9{Q)r
zR<R^K;|r3aJ9UjOPnsV(r$%-B#PuO@DFs4Ax1mWwy(f=|cefV6aYrfZ>o;5Y`uiS?
ztlh?zCT&g~V<cF4U-RjWp3Jo9y!D_cEv3B2?G(14w1j^Z@4z)Pv(Am;lW-e4&y>JQ
zEUMav=hn?p&oA68AgxJOTN;FF<8$ppDcQ2A;b7Ys-^cfaQcHXoxipNJPFVSw-mfBs
zzaVUB+M|0rH<_OzCJzUN(v15q<YiS78`-1L+ksiq#N=OoL&X)q2?1p_dWgK-v#<Z2
z2`?jJy7#gk#cCfVKe8=?LhAdnukgUBBhOFOkDwmAl<oD`n{RBx&VW4}UPv&Q*MR$~
zy>05Cr|yS_%s03bVZ=HCp?7dHE?dQVmK)mGk$ty0&E93qELOsC(<jI9RI1|D*tQ0o
zi$Ba4*rk_ca_KrS*09LJ#gm<6=+f5}Ot%wA4BJ=D9)3SsA&-&Ij!#LRR)3P3%1*1c
zq>ig=Zmd*6H2kj7%Fuwn%O!i34r}$}u?5M-Q7N&DlQn;SoYLVnaYFnoi({PR$3M5m
zMu@~tjtiehZY+dmt&K&ordL<9<*#dA9gAF)HOi;^_1jxHLFHroFkSAtP>V**dTO{;
z&Cp5aZc){fEf~*EjTbdY--jTYHyWtwYKXLOq&@8optEi6<kivcCOY;N1U&1C+N--}
zxm$U|{tD(}EQ|79PV>6>k{LKZ<#PQ_>3(d6rpU0X?ZZxj@omfP^Ur;qI0+Nf0j8f%
zvd#M)wm-4NxyASWd>X-6TYeP(IySL&WMc~tfg3^1wb|vk344mJIQ^$_HKoSu0=4$5
zL)zvW5MfJ|+FlZ$C2)lSKs+1(FaUUzuFcNL^JAsrbu#2xuO*zq<CrVFhS@Opd_gFG
zTJ5)n5Dt=gQ+_;b*NeR^qGhd`zeplU0mji-;|KBEi{Q(&IRsK^e)n6A*&S6<R94b3
z<>NJbk4w7ieQS3!t;1A_RSm-&xM{SBw6X>@m>IaD)Bmk=q|>BYU%GhkfJsxCm>tCL
z(0c{js(wBQokBNpm`<-X&Y8GzCfIwBwG$0rLGr%ThCB^YIqU&o>q4_$#%R_v>w4Yk
z(sM^HpZ3oQ>?4DgMo+5P94C8Wb$dnaYS~pJKA-NKjxY)9n*Vwz@kT__Tx~#L$1zb<
zk;Iy-xl-z714{2Z3)eYRES_X~<^A|5u19EoQGhS2vfhA>p-Sqy$|G`^v|b;^)uXqp
zs4jLcyua#*(UJ0<6+-l45Mz9qPj3jdeBkU1A<$JSew8i*zU}*AoB<5#dzP_}3IyJm
zuSG%tZ<UCEzy+Ho9SH2&l>U!N`twH!fFkNfr3Eyg*+aJ%&EbEG38axBwwq=I_tpYg
zTDqW&wgjS<(2d~t4ZuMIoQ($^rWmnu-9b|LZt&0+1Qa?@oEh*49dD`>hbyt8Yg)lP
zP#Clz8}P5s9<Z8j3Z-~TH-fRjcXTu8g<%-M#NbK>0k4d|w8uby3jzS{sR4lgH}P02
z(AgaerZF&(V@a!5My8^+aPWIa&<1ksW?-QIB@83D$0mr<a>qbLz6%ztXh8yiwxYss
PlZe1g?lvlXsxtH6$l;8&

delta 1608
zcmex7i}B+uM&1B#W)=|!5b&?vypcDaof$}P&Sx)X0yCa+&SmBVip2O>&2nBhxtmv}
zJ~~_2QRMjkQ&vw*Z+e^G&U>-#%CUqc83%9KO*{Ec#k!<i?Tp{7UZo#F-F7STyA(7c
zj<CP>UML~3-)Hi*{D(Jv&8E9~b=C@gD|>e4&%ft$EUizkE|k#t?cy@UnEfRyuV%~6
zJB!ymT;Lu(^^Z`HQs>Ne=5rm*Rw_=bwNK~jeaot!^Imhq99O<iOLL}$m=@IvJz8;K
zc1z>(bFLk#GZrm$^L5-H{<GzTK>)As<|_ihZEP)*cJMV{e|<$Y_1!k7Lr)l#JQ`G4
z<bQ6FOx<!}FQ49#r1e+N?6%2aTI9y8WvwZqxiET_nS}TbF^5YV<3f@cr?YNfcku9u
zDM@$dOypXrT~UAJ^(KZFi5E8iyuHdclu=-2|B~!g>--m|t**44#QF5ja!>u9z-x1K
zvX5*!W)S;3eb&0(!eeu`JnBeWP-vXbReI^cq#d5=2NR4o_<CAP{T2@HEZ6v4aodqw
z%;oAyoxL4X%6{#2wtZ=R;=-Fp8E^C!^2S@8h%-qLWb3Wn)SA74r@qcq{!h$><8wS^
zY+7QP5}w>pa|$Xil5h%EPx3ZCbC2t=;+MSlUv~We_TbI)*Y`fI@189Gx4(Vw;pd$E
z%R=@tMkjF^Pi}J9t?W8Q!@}jNH|w&5_SgrpHCC_o{;_g4TBjg6-822}`Tg_vzMm`i
zmCx_|{=F9O>t;UGjX4;-<IL;&dW%=(_hV}8U+%9qJ!r5#VMgZ9O|vSO_(^_z$DDgh
zitEiWMsDuiqEa48ry?Fn^)-lnNO3l@yx~8;O3t${Xv?STm+qfh`?M$MU7X)_vFTSi
z=k9d<nt5@iu9Q|-@zL2AfA60ZvfNy3#=B(!uhuF=w4Z0sX->}np*=Ao&&8-#Ti4M}
z_REqrYuD=@oVs1RZKk>J=`;7&>D)aT-L~-d)304SBR0;z_k@w}W7Dq(GmkAwdw%wm
zJS>TGfRgy;M1c%WaMC^_sn1vsO4<S4v1v|>3=EY_3=D$6jF6I_98i>BP^_PvUzDm>
zk(<*y>7w5u1Ch4))R%v+P2)`3-0Q*{K52%<;*aYx`7i7!^Np6QuHMebtRx;+_W!#6
zf6LfyH(qPqYW%#)%yJQ@qo72fcdD6XmHzaPGOl(7%9WwgCjz<Ubh~^a>eaH?*RR|5
zpovMoRPOQ24VJgs5+sks<v144FKKyssPwG~|J{qZ8=gLzozLLUF*A|3(Ob^qUF$~K
z-WMyTB!%3P59ejlGzeL6kflG<>Ele^2V3@6CVJ?3taZwMs8+@0UiJPehneqWla~!;
ze>MNb?A^n=_}Hm0vxOSEe7MwPJ_OY7UNryyw%=NZ#pQpqhTUlo-Trbb)6W}XF<Gfb
zf6QM*vRhxZ4|IIl(DVF_#ixTjMz2<FI-IcliE?d~-t-kSCY;n!yFBBZw^O8-#LoY(
zykqa$bAR1<u)c?1b?zqXLziy0epJzDJ9d4)_eXA&NQiOPKYIKfFsL_I$_OxmW1~kl
zk`2swp={2=2a1*j$&wNt%nS^BI3|B^k=*R25zY)&ut0m42$&J;(8>a4Jazd5W=y{2
zX$KOREbV0mrbE0ekP5G<UM67CGhWh6!2**D1jHu)_u>Hy@p(%#?whRTtr>?cL4Zm}
zVEO=p4^mi?M}RjYFy(*>QE&=_d+q>Gj{sO3Onc(wjo$L|AS*s*JnhtBV_;Y(&%hvr
zqNaB8dv8Usn!|hB?{cy*Fz5<1FaQx4K%8-Nvc8WJSPdeO7#Ji_)Cf;5@KFbQZkCTU
z(?spbh8m)i&-e&}g&qKf{2V|+hQ^>2j_!rGNoO?o0<9JXx*yfpYc61eB78;JPI`hY
F1OS@`msbD)

diff --git a/server/src/app/algorithem/README.md b/server/src/app/algorithem/README.md
new file mode 100644
index 0000000..ae62a73
--- /dev/null
+++ b/server/src/app/algorithem/README.md
@@ -0,0 +1,21 @@
+# X-Financial 核心算法推演目录
+
+> 目录名 `algorithem` 沿用当前任务指定拼写。该目录用于沉淀核心算法推演、公式口径和可审计实现，避免把算法细节直接堆进 `services`。
+
+## 目录职责
+
+- 保存预算、费用、风控、知识检索等核心算法的推演文档。
+- 记录公式、权重、阈值、输入输出协议和边界案例。
+- 为后续 Python 实现、单元测试和接口协议提供依据。
+
+## 当前算法主题
+
+- `applicant_expense_profile_formula.md`：申请人费用画像与审核建议量化公式。
+- `applicant_expense_profile.py`：申请人费用画像评分的第一版纯算法实现。
+
+## 落地原则
+
+- 算法先有可解释公式，再进入业务服务实现。
+- 硬规则、评分权重和自然语言解释要分层。
+- 所有核心算法模块都要遵守 800 行上限，按职责拆分。
+- 涉及审批建议时，输出“依据 + 建议动作”，不要直接给人贴负面标签。
diff --git a/server/src/app/algorithem/__init__.py b/server/src/app/algorithem/__init__.py
new file mode 100644
index 0000000..4f60fcc
--- /dev/null
+++ b/server/src/app/algorithem/__init__.py
@@ -0,0 +1,13 @@
+"""Core algorithm derivations for X-Financial."""
+
+from .applicant_expense_profile import (
+    ApplicantExpenseProfileInput,
+    ApplicantExpenseProfileResult,
+    evaluate_applicant_expense_profile,
+)
+
+__all__ = [
+    "ApplicantExpenseProfileInput",
+    "ApplicantExpenseProfileResult",
+    "evaluate_applicant_expense_profile",
+]
diff --git a/server/src/app/algorithem/applicant_expense_profile.py b/server/src/app/algorithem/applicant_expense_profile.py
new file mode 100644
index 0000000..8c6eea3
--- /dev/null
+++ b/server/src/app/algorithem/applicant_expense_profile.py
@@ -0,0 +1,445 @@
+"""Applicant expense profile scoring algorithm.
+
+The module is intentionally pure and framework-free. Service layers can build the
+input snapshot from database records, while this module only owns the formula,
+scores, thresholds, and explainable result.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass, field
+from decimal import ROUND_HALF_UP, Decimal, InvalidOperation
+from typing import Any
+
+ZERO = Decimal("0")
+ONE = Decimal("1")
+HUNDRED = Decimal("100")
+
+LEVEL_NORMAL = "normal"
+LEVEL_WATCH = "watch"
+LEVEL_REVIEW = "review"
+LEVEL_ESCALATION = "escalation"
+
+
+@dataclass(slots=True)
+class ApplicantExpenseProfileInput:
+    """Inputs for applicant expense behavior scoring.
+
+    Values should be pre-aggregated by a comparable peer group, such as
+    department + role + expense type + city grade + project type.
+    """
+
+    applicant_claim_count_90d: int = 0
+    peer_claim_count_p75_90d: Any = ZERO
+    applicant_amount_90d: Any = ZERO
+    available_peer_budget_90d: Any = ZERO
+    amount_percentile: Any = ZERO
+    peer_amount_median_90d: Any = ZERO
+    adjusted_or_returned_count_180d: int = 0
+    approved_claim_count_180d: int = 0
+    requested_days: Any = ZERO
+    peer_travel_days_p75: Any = ZERO
+    business_buffer_days: Any = ONE
+    claim_amount: Any = ZERO
+    peer_daily_cost_baseline: Any = ZERO
+    tolerance_factor: Any = Decimal("1.20")
+    entertainment_amount: Any = ZERO
+    attendee_count: int = 0
+    entertainment_standard_cap: Any = ZERO
+    same_customer_frequency_90d: int = 0
+    applicant_entertainment_percentile: Any = ZERO
+    hard_rule_score: int = 0
+
+
+@dataclass(slots=True)
+class ApplicantExpenseProfileResult:
+    profile_score: int
+    profile_level: str
+    recommendation_score: int
+    recommendation_level: str
+    frequency_score: int
+    amount_occupancy_score: int
+    peer_deviation_score: int
+    adjustment_history_score: int
+    current_claim_deviation_score: int
+    travel_days_deviation_score: int
+    daily_cost_deviation_score: int
+    entertainment_deviation_score: int
+    frequency_ratio: Decimal
+    budget_share_ratio: Decimal
+    peer_deviation_ratio: Decimal
+    adjustment_ratio: Decimal
+    travel_days_ratio: Decimal
+    daily_cost_ratio: Decimal
+    per_capita_entertainment_ratio: Decimal
+    suggested_days: Decimal | None
+    suggested_amount_cap: Decimal | None
+    basis_codes: list[str] = field(default_factory=list)
+
+    def as_dict(self) -> dict[str, Any]:
+        return {
+            "profile_score": self.profile_score,
+            "profile_level": self.profile_level,
+            "recommendation_score": self.recommendation_score,
+            "recommendation_level": self.recommendation_level,
+            "scores": {
+                "frequency_score": self.frequency_score,
+                "amount_occupancy_score": self.amount_occupancy_score,
+                "peer_deviation_score": self.peer_deviation_score,
+                "adjustment_history_score": self.adjustment_history_score,
+                "current_claim_deviation_score": self.current_claim_deviation_score,
+                "travel_days_deviation_score": self.travel_days_deviation_score,
+                "daily_cost_deviation_score": self.daily_cost_deviation_score,
+                "entertainment_deviation_score": self.entertainment_deviation_score,
+            },
+            "metrics": {
+                "frequency_ratio": _format_decimal(self.frequency_ratio),
+                "budget_share_ratio": _format_decimal(self.budget_share_ratio),
+                "peer_deviation_ratio": _format_decimal(self.peer_deviation_ratio),
+                "adjustment_ratio": _format_decimal(self.adjustment_ratio),
+                "travel_days_ratio": _format_decimal(self.travel_days_ratio),
+                "daily_cost_ratio": _format_decimal(self.daily_cost_ratio),
+                "per_capita_entertainment_ratio": _format_decimal(
+                    self.per_capita_entertainment_ratio
+                ),
+                "suggested_days": _format_decimal(self.suggested_days),
+                "suggested_amount_cap": _format_decimal(self.suggested_amount_cap),
+            },
+            "basis_codes": list(self.basis_codes),
+        }
+
+
+def evaluate_applicant_expense_profile(
+    payload: ApplicantExpenseProfileInput,
+) -> ApplicantExpenseProfileResult:
+    frequency_ratio = _ratio(payload.applicant_claim_count_90d, payload.peer_claim_count_p75_90d)
+    frequency_score = _score_frequency_ratio(frequency_ratio)
+
+    budget_share_ratio = _ratio(payload.applicant_amount_90d, payload.available_peer_budget_90d)
+    amount_percentile_score = _score_percentile(_to_decimal(payload.amount_percentile))
+    budget_share_score = _score_budget_share_ratio(budget_share_ratio)
+    amount_occupancy_score = max(amount_percentile_score, budget_share_score)
+
+    peer_deviation_ratio = _ratio(payload.applicant_amount_90d, payload.peer_amount_median_90d)
+    peer_deviation_score = _score_peer_deviation_ratio(peer_deviation_ratio)
+
+    adjustment_ratio = _ratio(
+        payload.adjusted_or_returned_count_180d,
+        payload.approved_claim_count_180d,
+    )
+    adjustment_history_score = _score_adjustment_ratio(adjustment_ratio)
+
+    travel_days_ratio = _ratio(payload.requested_days, payload.peer_travel_days_p75)
+    travel_days_deviation_score = _score_travel_days_ratio(travel_days_ratio)
+
+    requested_days = _to_decimal(payload.requested_days)
+    daily_cost = _ratio(payload.claim_amount, requested_days)
+    daily_cost_ratio = _ratio(daily_cost, payload.peer_daily_cost_baseline)
+    daily_cost_deviation_score = _score_daily_cost_ratio(daily_cost_ratio)
+
+    per_capita_amount = _ratio(payload.entertainment_amount, payload.attendee_count)
+    per_capita_entertainment_ratio = _ratio(
+        per_capita_amount,
+        payload.entertainment_standard_cap,
+    )
+    entertainment_deviation_score = max(
+        _score_entertainment_per_capita_ratio(per_capita_entertainment_ratio),
+        _score_same_customer_frequency(payload.same_customer_frequency_90d),
+        _score_percentile(_to_decimal(payload.applicant_entertainment_percentile)),
+    )
+
+    current_claim_deviation_score = max(
+        travel_days_deviation_score,
+        daily_cost_deviation_score,
+        entertainment_deviation_score,
+    )
+
+    profile_score = _weighted_profile_score(
+        frequency_score=frequency_score,
+        amount_occupancy_score=amount_occupancy_score,
+        peer_deviation_score=peer_deviation_score,
+        adjustment_history_score=adjustment_history_score,
+        current_claim_deviation_score=current_claim_deviation_score,
+    )
+
+    hard_rule_score = _clamp_score(payload.hard_rule_score)
+    recommendation_score = max(profile_score, current_claim_deviation_score, hard_rule_score)
+
+    suggested_days = _suggest_days(
+        requested_days=requested_days,
+        baseline_days=_to_decimal(payload.peer_travel_days_p75),
+        business_buffer_days=_to_decimal(payload.business_buffer_days),
+    )
+    suggested_amount_cap = _suggest_amount_cap(
+        suggested_days=suggested_days,
+        daily_cost_baseline=_to_decimal(payload.peer_daily_cost_baseline),
+        tolerance_factor=_to_decimal(payload.tolerance_factor),
+    )
+
+    basis_codes = _build_basis_codes(
+        frequency_ratio=frequency_ratio,
+        amount_percentile=_to_decimal(payload.amount_percentile),
+        budget_share_ratio=budget_share_ratio,
+        peer_deviation_ratio=peer_deviation_ratio,
+        adjustment_ratio=adjustment_ratio,
+        travel_days_ratio=travel_days_ratio,
+        daily_cost_ratio=daily_cost_ratio,
+        per_capita_entertainment_ratio=per_capita_entertainment_ratio,
+        same_customer_frequency_90d=payload.same_customer_frequency_90d,
+        applicant_entertainment_percentile=_to_decimal(payload.applicant_entertainment_percentile),
+        hard_rule_score=hard_rule_score,
+    )
+
+    return ApplicantExpenseProfileResult(
+        profile_score=profile_score,
+        profile_level=level_from_score(profile_score),
+        recommendation_score=recommendation_score,
+        recommendation_level=level_from_score(recommendation_score),
+        frequency_score=frequency_score,
+        amount_occupancy_score=amount_occupancy_score,
+        peer_deviation_score=peer_deviation_score,
+        adjustment_history_score=adjustment_history_score,
+        current_claim_deviation_score=current_claim_deviation_score,
+        travel_days_deviation_score=travel_days_deviation_score,
+        daily_cost_deviation_score=daily_cost_deviation_score,
+        entertainment_deviation_score=entertainment_deviation_score,
+        frequency_ratio=frequency_ratio,
+        budget_share_ratio=budget_share_ratio,
+        peer_deviation_ratio=peer_deviation_ratio,
+        adjustment_ratio=adjustment_ratio,
+        travel_days_ratio=travel_days_ratio,
+        daily_cost_ratio=daily_cost_ratio,
+        per_capita_entertainment_ratio=per_capita_entertainment_ratio,
+        suggested_days=suggested_days,
+        suggested_amount_cap=suggested_amount_cap,
+        basis_codes=basis_codes,
+    )
+
+
+def level_from_score(score: int) -> str:
+    normalized = _clamp_score(score)
+    if normalized >= 80:
+        return LEVEL_ESCALATION
+    if normalized >= 60:
+        return LEVEL_REVIEW
+    if normalized >= 40:
+        return LEVEL_WATCH
+    return LEVEL_NORMAL
+
+
+def _weighted_profile_score(
+    *,
+    frequency_score: int,
+    amount_occupancy_score: int,
+    peer_deviation_score: int,
+    adjustment_history_score: int,
+    current_claim_deviation_score: int,
+) -> int:
+    weighted = (
+        Decimal(frequency_score) * Decimal("0.20")
+        + Decimal(amount_occupancy_score) * Decimal("0.25")
+        + Decimal(peer_deviation_score) * Decimal("0.25")
+        + Decimal(adjustment_history_score) * Decimal("0.15")
+        + Decimal(current_claim_deviation_score) * Decimal("0.15")
+    )
+    return _clamp_score(int(weighted.quantize(Decimal("1"), rounding=ROUND_HALF_UP)))
+
+
+def _score_frequency_ratio(ratio: Decimal) -> int:
+    return _score_ratio(
+        ratio,
+        [(Decimal("1.0"), 0), (Decimal("1.2"), 30), (Decimal("1.5"), 60), (Decimal("2.0"), 80)],
+    )
+
+
+def _score_budget_share_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("0.10"), 0), (Decimal("0.20"), 40), (Decimal("0.35"), 70)])
+
+
+def _score_peer_deviation_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("1.0"), 0), (Decimal("1.3"), 40), (Decimal("1.8"), 70)])
+
+
+def _score_adjustment_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("0.05"), 0), (Decimal("0.15"), 40), (Decimal("0.30"), 70)])
+
+
+def _score_travel_days_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("1.2"), 0), (Decimal("1.5"), 40), (Decimal("2.0"), 70)])
+
+
+def _score_daily_cost_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("1.1"), 0), (Decimal("1.3"), 40), (Decimal("1.6"), 70)])
+
+
+def _score_entertainment_per_capita_ratio(ratio: Decimal) -> int:
+    return _score_ratio(ratio, [(Decimal("1.0"), 0), (Decimal("1.2"), 40), (Decimal("1.5"), 70)])
+
+
+def _score_same_customer_frequency(frequency: int) -> int:
+    if frequency >= 5:
+        return 100
+    if frequency >= 3:
+        return 70
+    if frequency >= 2:
+        return 40
+    return 0
+
+
+def _score_percentile(percentile: Decimal) -> int:
+    normalized = max(ZERO, min(HUNDRED, percentile))
+    if normalized <= Decimal("75"):
+        return 0
+    if normalized <= Decimal("85"):
+        return 40
+    if normalized <= Decimal("95"):
+        return 70
+    return 100
+
+
+def _score_ratio(ratio: Decimal, bands: list[tuple[Decimal, int]]) -> int:
+    if ratio <= ZERO:
+        return 0
+    for upper_bound, score in bands:
+        if ratio <= upper_bound:
+            return score
+    return 100
+
+
+def _suggest_days(
+    *,
+    requested_days: Decimal,
+    baseline_days: Decimal,
+    business_buffer_days: Decimal,
+) -> Decimal | None:
+    if requested_days <= ZERO:
+        return None
+    if baseline_days <= ZERO:
+        return _quantize(requested_days)
+    return _quantize(min(requested_days, baseline_days + max(ZERO, business_buffer_days)))
+
+
+def _suggest_amount_cap(
+    *,
+    suggested_days: Decimal | None,
+    daily_cost_baseline: Decimal,
+    tolerance_factor: Decimal,
+) -> Decimal | None:
+    if suggested_days is None or suggested_days <= ZERO or daily_cost_baseline <= ZERO:
+        return None
+    factor = tolerance_factor if tolerance_factor > ZERO else ONE
+    return _quantize_money(suggested_days * daily_cost_baseline * factor)
+
+
+def _build_basis_codes(
+    *,
+    frequency_ratio: Decimal,
+    amount_percentile: Decimal,
+    budget_share_ratio: Decimal,
+    peer_deviation_ratio: Decimal,
+    adjustment_ratio: Decimal,
+    travel_days_ratio: Decimal,
+    daily_cost_ratio: Decimal,
+    per_capita_entertainment_ratio: Decimal,
+    same_customer_frequency_90d: int,
+    applicant_entertainment_percentile: Decimal,
+    hard_rule_score: int,
+) -> list[str]:
+    basis_codes: list[str] = []
+
+    if frequency_ratio > Decimal("1.5"):
+        basis_codes.append("applicant.frequency.ratio_review")
+    elif frequency_ratio > Decimal("1.2"):
+        basis_codes.append("applicant.frequency.ratio_watch")
+
+    if amount_percentile > Decimal("95"):
+        basis_codes.append("applicant.amount_percentile.p95")
+    elif amount_percentile > Decimal("85"):
+        basis_codes.append("applicant.amount_percentile.p85")
+
+    if budget_share_ratio > Decimal("0.35"):
+        basis_codes.append("applicant.budget_share.high")
+    elif budget_share_ratio > Decimal("0.20"):
+        basis_codes.append("applicant.budget_share.watch")
+
+    if peer_deviation_ratio > Decimal("1.8"):
+        basis_codes.append("applicant.peer_deviation.escalation")
+    elif peer_deviation_ratio > Decimal("1.3"):
+        basis_codes.append("applicant.peer_deviation.review")
+
+    if adjustment_ratio > Decimal("0.30"):
+        basis_codes.append("applicant.adjustment_history.escalation")
+    elif adjustment_ratio > Decimal("0.15"):
+        basis_codes.append("applicant.adjustment_history.review")
+
+    if travel_days_ratio > Decimal("2.0"):
+        basis_codes.append("travel.days_ratio.escalation")
+    elif travel_days_ratio > Decimal("1.5"):
+        basis_codes.append("travel.days_ratio.review")
+    elif travel_days_ratio > Decimal("1.2"):
+        basis_codes.append("travel.days_ratio.watch")
+
+    if daily_cost_ratio > Decimal("1.6"):
+        basis_codes.append("travel.daily_cost_ratio.escalation")
+    elif daily_cost_ratio > Decimal("1.3"):
+        basis_codes.append("travel.daily_cost_ratio.review")
+    elif daily_cost_ratio > Decimal("1.1"):
+        basis_codes.append("travel.daily_cost_ratio.watch")
+
+    if per_capita_entertainment_ratio > Decimal("1.5"):
+        basis_codes.append("entertainment.per_capita.escalation")
+    elif per_capita_entertainment_ratio > Decimal("1.2"):
+        basis_codes.append("entertainment.per_capita.review")
+
+    if same_customer_frequency_90d >= 5:
+        basis_codes.append("entertainment.same_customer.frequency_escalation")
+    elif same_customer_frequency_90d >= 3:
+        basis_codes.append("entertainment.same_customer.frequency_review")
+
+    if applicant_entertainment_percentile > Decimal("95"):
+        basis_codes.append("entertainment.amount_percentile.p95")
+    elif applicant_entertainment_percentile > Decimal("85"):
+        basis_codes.append("entertainment.amount_percentile.p85")
+
+    if hard_rule_score >= 80:
+        basis_codes.append("hard_rule.escalation")
+    elif hard_rule_score >= 60:
+        basis_codes.append("hard_rule.review")
+
+    return basis_codes
+
+
+def _ratio(numerator: Any, denominator: Any) -> Decimal:
+    divisor = _to_decimal(denominator)
+    if divisor <= ZERO:
+        return ZERO
+    return _quantize(_to_decimal(numerator) / divisor)
+
+
+def _to_decimal(value: Any) -> Decimal:
+    if value is None:
+        return ZERO
+    if isinstance(value, Decimal):
+        return value
+    try:
+        return Decimal(str(value))
+    except (InvalidOperation, ValueError):
+        return ZERO
+
+
+def _quantize(value: Decimal) -> Decimal:
+    return value.quantize(Decimal("0.0001"), rounding=ROUND_HALF_UP)
+
+
+def _quantize_money(value: Decimal) -> Decimal:
+    return value.quantize(Decimal("0.01"), rounding=ROUND_HALF_UP)
+
+
+def _format_decimal(value: Decimal | None) -> str | None:
+    if value is None:
+        return None
+    return format(value.normalize(), "f")
+
+
+def _clamp_score(score: int) -> int:
+    return max(0, min(100, int(score)))
diff --git a/server/src/app/algorithem/applicant_expense_profile_formula.md b/server/src/app/algorithem/applicant_expense_profile_formula.md
new file mode 100644
index 0000000..1691e42
--- /dev/null
+++ b/server/src/app/algorithem/applicant_expense_profile_formula.md
@@ -0,0 +1,224 @@
+# 申请人费用画像与审核建议量化公式
+
+## 目标
+
+申请人费用画像用于回答：
+
+- 该申请人的近期费用节奏是否高于同组基准。
+- 本次申请的天数、金额、招待频率是否明显偏离。
+- 审批时应普通通过、谨慎通过、重点复核，还是升级审批。
+
+画像结果只作为审批参考，不应直接定义员工为异常人员。
+
+## 同组基准
+
+同组基准必须按业务可比口径聚合：
+
+```text
+peer_group =
+  部门
++ 岗位/职级
++ 费用类型
++ 城市等级/业务区域
++ 项目类型/客户阶段
++ 近 90 天或近 180 天窗口
+```
+
+不能用全公司平均值直接比较销售、研发、项目经理等不同岗位。
+
+## 画像风险分
+
+```text
+profile_score =
+  frequency_score * 0.20
++ amount_occupancy_score * 0.25
++ peer_deviation_score * 0.25
++ adjustment_history_score * 0.15
++ current_claim_deviation_score * 0.15
+```
+
+等级：
+
+```text
+0-39   normal          正常
+40-59  watch           需关注
+60-79  review          重点复核
+80-100 escalation      强复核 / 升级审批
+```
+
+## 子分计算
+
+### 高频申请分
+
+```text
+frequency_ratio = applicant_claim_count_90d / peer_claim_count_p75_90d
+
+frequency_score =
+  0   if frequency_ratio <= 1.0
+  30  if 1.0 < frequency_ratio <= 1.2
+  60  if 1.2 < frequency_ratio <= 1.5
+  80  if 1.5 < frequency_ratio <= 2.0
+  100 if frequency_ratio > 2.0
+```
+
+### 高金额占用分
+
+```text
+budget_share_ratio = applicant_amount_90d / available_peer_budget_90d
+amount_percentile = applicant_amount_90d 在同组中的分位数
+
+amount_occupancy_score =
+  max(
+    percentile_score(amount_percentile),
+    ratio_score(budget_share_ratio)
+  )
+```
+
+建议第一版分位规则：
+
+```text
+P0-P75   -> 0
+P75-P85  -> 40
+P85-P95  -> 70
+P95-P100 -> 100
+```
+
+### 同组偏离分
+
+```text
+peer_deviation_ratio =
+  applicant_amount_90d / peer_amount_median_90d
+
+peer_deviation_score =
+  0   if peer_deviation_ratio <= 1.0
+  40  if 1.0 < peer_deviation_ratio <= 1.3
+  70  if 1.3 < peer_deviation_ratio <= 1.8
+  100 if peer_deviation_ratio > 1.8
+```
+
+### 历史调减退回分
+
+```text
+adjustment_ratio =
+  adjusted_or_returned_count_180d / approved_claim_count_180d
+
+adjustment_history_score =
+  0   if adjustment_ratio <= 0.05
+  40  if 0.05 < adjustment_ratio <= 0.15
+  70  if 0.15 < adjustment_ratio <= 0.30
+  100 if adjustment_ratio > 0.30
+```
+
+### 本次申请偏离分
+
+```text
+current_claim_deviation_score =
+  max(
+    travel_days_deviation_score,
+    daily_cost_deviation_score,
+    entertainment_deviation_score
+  )
+```
+
+## 出差天数建议
+
+```text
+baseline_days = peer_group.travel_days_p75
+travel_days_ratio = requested_days / baseline_days
+suggested_days = min(
+  requested_days,
+  baseline_days + business_buffer_days
+)
+```
+
+建议阈值：
+
+```text
+travel_days_ratio <= 1.2       正常
+1.2 < ratio <= 1.5             提醒关注
+1.5 < ratio <= 2.0             建议压缩天数或补充说明
+ratio > 2.0                    重点复核 / 升级审批
+```
+
+## 出差费用建议
+
+```text
+daily_cost = claim_amount / requested_days
+daily_cost_ratio = daily_cost / peer_city_grade_daily_cost_baseline
+suggested_amount_cap =
+  suggested_days
+  * peer_city_grade_daily_cost_baseline
+  * tolerance_factor
+```
+
+第一版 `tolerance_factor` 建议取 `1.15` 到 `1.20`。
+
+建议阈值：
+
+```text
+daily_cost_ratio <= 1.1       正常
+1.1 < ratio <= 1.3            提醒关注
+1.3 < ratio <= 1.6            建议调减费用标准
+ratio > 1.6                   重点复核 / 升级审批
+```
+
+## 招待费用建议
+
+```text
+per_capita_entertainment_amount = entertainment_amount / attendee_count
+per_capita_ratio = per_capita_entertainment_amount / entertainment_standard_cap
+same_customer_frequency_90d = count(customer_id, applicant_id, 90d)
+applicant_entertainment_percentile =
+  applicant_entertainment_amount_90d 在同组中的分位数
+```
+
+建议阈值：
+
+```text
+per_capita_ratio <= 1.0            正常
+1.0 < ratio <= 1.2                 提醒关注
+1.2 < ratio <= 1.5                 建议调低招待标准
+ratio > 1.5                        重点复核 / 升级审批
+same_customer_frequency_90d >= 3   要求补充客户推进阶段
+percentile >= P85                  重点复核
+percentile >= P95                  升级审批
+```
+
+## 审核建议强度
+
+```text
+recommendation_strength =
+  max(
+    level(profile_score),
+    level(current_claim_deviation_score),
+    level(hard_rule_score)
+  )
+```
+
+输出文案必须引用触发指标：
+
+```text
+该申请人近 90 天费用占用处于同组 P88，
+本次出差天数为同类 P75 的 1.67 倍。
+建议将出差天数由 5 天调整为 4 天，
+或补充客户拜访安排和项目阶段说明。
+```
+
+## 输出字段建议
+
+```json
+{
+  "profile_score": 66,
+  "profile_level": "review",
+  "peer_percentile": 88,
+  "travel_days_ratio": "1.67",
+  "daily_cost_ratio": "1.34",
+  "same_customer_frequency_90d": 3,
+  "suggested_days": 4,
+  "suggested_amount_cap": "6800.00",
+  "basis_codes": [
+    "applicant.peer_percentile.p85",
+    "travel.days_ratio.review"
+  ]
+}
+```
diff --git a/server/src/app/api/v1/endpoints/reimbursements.py b/server/src/app/api/v1/endpoints/reimbursements.py
index 65b002c..e17561c 100644
--- a/server/src/app/api/v1/endpoints/reimbursements.py
+++ b/server/src/app/api/v1/endpoints/reimbursements.py
@@ -7,6 +7,7 @@ from fastapi.responses import FileResponse
 from sqlalchemy.orm import Session
 
 from app.api.deps import CurrentUserContext, get_current_user, get_db
+from app.schemas.budget import BudgetClaimAnalysisRead
 from app.schemas.common import ErrorResponse
 from app.schemas.reimbursement import (
     ExpenseClaimAttachmentActionResponse,
@@ -25,6 +26,7 @@ from app.schemas.reimbursement import (
     TravelReimbursementCalculatorResponse,
 )
 from app.services.expense_claims import ExpenseClaimService
+from app.services.budget import BudgetService
 from app.services.reimbursement import ReimbursementService
 from app.services.travel_reimbursement_calculator import TravelReimbursementCalculatorService
 
@@ -126,6 +128,38 @@ def get_expense_claim(claim_id: str, db: DbSession, current_user: CurrentUser) -
     return claim
 
 
+@router.get(
+    "/claims/{claim_id}/budget-analysis",
+    response_model=BudgetClaimAnalysisRead,
+    summary="读取申请单预算分析",
+    description="根据当前预算池、申请金额和预算管控模型，返回费用申请的预算影响和评分建议。",
+    responses={
+        status.HTTP_404_NOT_FOUND: {
+            "model": ErrorResponse,
+            "description": "单据不存在。",
+        },
+        status.HTTP_403_FORBIDDEN: {
+            "model": ErrorResponse,
+            "description": "当前用户无权查看预算分析。",
+        },
+    },
+)
+def get_expense_claim_budget_analysis(
+    claim_id: str,
+    db: DbSession,
+    current_user: CurrentUser,
+) -> BudgetClaimAnalysisRead:
+    service = ExpenseClaimService(db)
+    if not service.can_view_budget_analysis(current_user):
+        raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail="只有预算监控员或高级财务人员可以查看预算分析。")
+    claim = service.get_claim(claim_id, current_user)
+    if claim is None:
+        raise HTTPException(status_code=status.HTTP_404_NOT_FOUND, detail="Claim not found")
+    if not service.can_view_budget_analysis(current_user, claim):
+        raise HTTPException(status_code=status.HTTP_403_FORBIDDEN, detail="只有该部门 P8 预算监控员或高级财务人员可以查看预算分析。")
+    return BudgetService(db).analyze_claim_budget(claim)
+
+
 @router.patch(
     "/claims/{claim_id}",
     response_model=ExpenseClaimRead,
@@ -538,7 +572,7 @@ def return_expense_claim(
     "/claims/{claim_id}/approve",
     response_model=ExpenseClaimRead,
     summary="审批通过单据",
-    description="费用申请由直属领导审批通过后完成；报销单直属领导审批后流转到财务审批，财务终审通过后进入归档入账。",
+    description="费用申请由直属领导审批后流转到预算管理者审批，预算审核通过后生成报销草稿；报销单直属领导审批后流转到财务审批。",
     responses={
         status.HTTP_404_NOT_FOUND: {
             "model": ErrorResponse,
diff --git a/server/src/app/schemas/budget.py b/server/src/app/schemas/budget.py
index c8d3702..f85b53e 100644
--- a/server/src/app/schemas/budget.py
+++ b/server/src/app/schemas/budget.py
@@ -2,6 +2,7 @@ from __future__ import annotations
 
 from datetime import datetime
 from decimal import Decimal
+from typing import Any
 
 from pydantic import BaseModel, ConfigDict, Field
 
@@ -144,6 +145,17 @@ class BudgetCheckRead(BaseModel):
     allocation: BudgetAllocationRead | None = None
 
 
+class BudgetClaimAnalysisRead(BaseModel):
+    budget_context: dict[str, Any] = Field(default_factory=dict)
+    score: int
+    rating: str
+    risk_level: str
+    summary: str
+    metrics: dict[str, Any] = Field(default_factory=dict)
+    basis: list[str] = Field(default_factory=list)
+    suggestions: list[str] = Field(default_factory=list)
+
+
 class BudgetOperationRead(BaseModel):
     ok: bool
     message: str
diff --git a/server/src/app/services/agent_foundation.py b/server/src/app/services/agent_foundation.py
index 32cbd12..fadbe2f 100644
--- a/server/src/app/services/agent_foundation.py
+++ b/server/src/app/services/agent_foundation.py
@@ -87,7 +87,10 @@ class AgentFoundationService(
 
     def _foundation_cache_key(self) -> str:
         bind = self.db.get_bind()
-        return str(getattr(bind, "url", "") or id(bind))
+        url = str(getattr(bind, "url", "") or "")
+        if url.endswith("/:memory:"):
+            return f"{url}:{id(bind)}"
+        return url or str(id(bind))
 
     def _ensure_financial_record_schema(self) -> None:
         bind = self.db.get_bind()
diff --git a/server/src/app/services/agent_runs.py b/server/src/app/services/agent_runs.py
index 0e735d6..886ceb8 100644
--- a/server/src/app/services/agent_runs.py
+++ b/server/src/app/services/agent_runs.py
@@ -18,6 +18,7 @@ from app.services.knowledge_ingest_log import enrich_knowledge_ingest_route_json
 logger = get_logger("app.services.agent_runs")
 
 KNOWLEDGE_SYNC_HEARTBEAT_TIMEOUT = timedelta(minutes=30)
+KNOWLEDGE_SYNC_JOB_TYPES = {"knowledge_index_sync", "llm_wiki_sync"}
 
 
 class AgentRunService:
@@ -262,7 +263,7 @@ class AgentRunService:
                 continue
 
             route_json = dict(run.route_json or {})
-            if str(route_json.get("job_type") or "").strip() != "knowledge_index_sync":
+            if str(route_json.get("job_type") or "").strip() not in KNOWLEDGE_SYNC_JOB_TYPES:
                 continue
 
             heartbeat_at = self._parse_heartbeat_time(
diff --git a/server/src/app/services/budget.py b/server/src/app/services/budget.py
index 9150612..934a07a 100644
--- a/server/src/app/services/budget.py
+++ b/server/src/app/services/budget.py
@@ -20,6 +20,7 @@ from app.schemas.budget import (
     BudgetSummaryRead,
     BudgetTransactionRead,
 )
+from app.services.budget_expense_control import BudgetExpenseControlModel
 from app.services.budget_support import BudgetSupportMixin
 from app.services.budget_types import BudgetControlError, SUPPORTED_BUDGET_SUBJECT_CODES
 
@@ -112,6 +113,9 @@ class BudgetService(BudgetSupportMixin):
             warnings=warnings,
         )
 
+    def analyze_claim_budget(self, claim: ExpenseClaim) -> dict[str, Any]:
+        return BudgetExpenseControlModel().assess(self.build_claim_budget_context(claim), claim)
+
     def create_or_update_allocation(
         self,
         payload: BudgetAllocationCreate,
diff --git a/server/src/app/services/budget_expense_control.py b/server/src/app/services/budget_expense_control.py
new file mode 100644
index 0000000..23ef660
--- /dev/null
+++ b/server/src/app/services/budget_expense_control.py
@@ -0,0 +1,194 @@
+from __future__ import annotations
+
+from decimal import Decimal, InvalidOperation
+from typing import Any
+
+from app.models.financial_record import ExpenseClaim
+
+
+class BudgetExpenseControlModel:
+    """预算费用管控模型：用预算容量、单据影响、规则动作和信息完整度给出建议。"""
+
+    def assess(self, budget_context: dict[str, Any], claim: ExpenseClaim | None = None) -> dict[str, Any]:
+        context = dict(budget_context or {})
+        amount = self._money(context.get("claim_amount"))
+        if not context.get("budget_applicable", True):
+            return self._build_result(
+                context=context,
+                score=72,
+                rating="reference",
+                risk_level="low",
+                summary="该费用类型暂未纳入预算强管控，本次仅作为申请合理性参考。",
+                basis=["当前费用科目未启用预算控制。"],
+                suggestions=["预算管理者可结合项目必要性和历史同类费用进行人工判断。"],
+            )
+        if not context.get("matched"):
+            return self._build_result(
+                context=context,
+                score=38,
+                rating="block",
+                risk_level="high",
+                summary="未匹配到可用预算池，建议先完成预算编制或调整预算维度后再审批。",
+                basis=["系统按部门、成本中心、项目和费用科目未找到匹配预算额度。"],
+                suggestions=["请预算编制者补充对应预算池，或核对申请部门、项目和费用类型是否填写正确。"],
+            )
+
+        total = self._money(context.get("total_amount"))
+        reserved = self._money(context.get("reserved_amount"))
+        consumed = self._money(context.get("consumed_amount"))
+        current_reserved = self._money(context.get("current_reserved_amount"))
+        warning_threshold = self._money(context.get("warning_threshold") or "80")
+        used_before = max(reserved + consumed - current_reserved, Decimal("0.00"))
+        available_before = max(total - used_before, Decimal("0.00"))
+        after_used = used_before + amount
+        claim_ratio = self._percent(amount, total)
+        after_usage_rate = self._percent(after_used, total)
+        over_budget_amount = max(amount - available_before, Decimal("0.00"))
+
+        score = 100
+        basis = [
+            f"预算池 {context.get('budget_no') or '未命名'} 总额度 {self._fmt(total)} 元。",
+            f"本次申请金额 {self._fmt(amount)} 元，占预算 {self._fmt(claim_ratio)}%。",
+            f"审批后预算使用率预计 {self._fmt(after_usage_rate)}%，预警线 {self._fmt(warning_threshold)}%。",
+        ]
+        suggestions: list[str] = []
+
+        if over_budget_amount > Decimal("0.00"):
+            score -= 55
+            basis.append(f"按当前预算余额测算，本次申请将超出预算 {self._fmt(over_budget_amount)} 元。")
+            suggestions.append("建议先追加或调剂预算，再允许申请继续流转。")
+        elif after_usage_rate >= Decimal("100.00"):
+            score -= 38
+            basis.append("审批后预算使用率将达到或超过 100%。")
+            suggestions.append("建议预算管理者复核剩余额度，并确认是否需要预算调剂。")
+        elif after_usage_rate >= warning_threshold:
+            score -= 20
+            basis.append("审批后预算使用率将触达预算预警线。")
+            suggestions.append("建议关注后续同类费用，必要时提前调整预算节奏。")
+        elif after_usage_rate >= Decimal("70.00"):
+            score -= 8
+            basis.append("审批后预算使用率较高，但尚未触达预警线。")
+
+        if claim_ratio >= Decimal("50.00"):
+            score -= 20
+            basis.append("单笔申请占预算比例超过 50%，对预算池影响较大。")
+            suggestions.append("建议补充业务必要性、交付范围和费用拆分依据。")
+        elif claim_ratio >= Decimal("30.00"):
+            score -= 12
+            basis.append("单笔申请占预算比例超过 30%，需要关注预算节奏。")
+        elif claim_ratio >= Decimal("15.00"):
+            score -= 5
+            basis.append("单笔申请占预算比例超过 15%，属于中等预算影响。")
+
+        missing_fields = self._collect_context_gaps(claim)
+        if missing_fields:
+            score -= min(12, len(missing_fields) * 4)
+            basis.append(f"申请信息仍缺少：{'、'.join(missing_fields)}。")
+            suggestions.append("建议申请人补齐业务背景，便于预算管理者判断费用必要性。")
+
+        control_action = str(context.get("control_action") or "").strip().lower()
+        if control_action == "block" and over_budget_amount > Decimal("0.00"):
+            suggestions.append("该预算池为硬控制口径，超预算时不建议直接通过。")
+
+        score = max(0, min(100, int(round(score))))
+        rating, risk_level = self._rate(score, over_budget_amount)
+        if not suggestions:
+            suggestions.append("预算额度与本次费用影响基本匹配，可以结合业务必要性继续审批。")
+
+        return self._build_result(
+            context={
+                **context,
+                "claim_amount_ratio": str(claim_ratio),
+                "after_usage_rate": str(after_usage_rate),
+                "available_before_amount": str(available_before),
+                "over_budget_amount": str(over_budget_amount),
+            },
+            score=score,
+            rating=rating,
+            risk_level=risk_level,
+            summary=self._summary(rating),
+            basis=basis,
+            suggestions=suggestions,
+        )
+
+    @staticmethod
+    def _collect_context_gaps(claim: ExpenseClaim | None) -> list[str]:
+        if claim is None:
+            return []
+        gaps = []
+        if not str(claim.reason or "").strip():
+            gaps.append("申请事由")
+        if not str(claim.location or "").strip():
+            gaps.append("地点")
+        if not str(claim.project_code or "").strip():
+            gaps.append("项目编号")
+        return gaps
+
+    @staticmethod
+    def _rate(score: int, over_budget_amount: Decimal) -> tuple[str, str]:
+        if over_budget_amount > Decimal("0.00") or score < 50:
+            return "block", "high"
+        if score < 70:
+            return "review", "medium"
+        if score < 85:
+            return "caution", "medium"
+        return "recommended", "low"
+
+    @staticmethod
+    def _summary(rating: str) -> str:
+        summaries = {
+            "recommended": "预算容量充足，单据费用与当前预算节奏基本匹配。",
+            "caution": "预算整体可承接，但本次费用对预算池已有一定影响。",
+            "review": "预算影响偏高，建议预算管理者结合业务必要性复核后再通过。",
+            "block": "预算风险较高，不建议在未补充依据或调整预算前直接通过。",
+        }
+        return summaries.get(rating, "已完成预算费用合理性测算。")
+
+    @staticmethod
+    def _build_result(
+        *,
+        context: dict[str, Any],
+        score: int,
+        rating: str,
+        risk_level: str,
+        summary: str,
+        basis: list[str],
+        suggestions: list[str],
+    ) -> dict[str, Any]:
+        return {
+            "budget_context": context,
+            "score": score,
+            "rating": rating,
+            "risk_level": risk_level,
+            "summary": summary,
+            "basis": basis,
+            "suggestions": suggestions,
+            "metrics": {
+                "claim_amount": context.get("claim_amount"),
+                "total_amount": context.get("total_amount"),
+                "claim_amount_ratio": context.get("claim_amount_ratio", "0.00"),
+                "usage_rate": context.get("usage_rate", "0.00"),
+                "after_usage_rate": context.get("after_usage_rate", context.get("usage_rate", "0.00")),
+                "available_amount": context.get("available_amount"),
+                "available_before_amount": context.get("available_before_amount"),
+                "over_budget_amount": context.get("over_budget_amount", "0.00"),
+            },
+        }
+
+    @staticmethod
+    def _money(value: Any) -> Decimal:
+        try:
+            return Decimal(str(value if value is not None else "0")).quantize(Decimal("0.01"))
+        except (InvalidOperation, ValueError):
+            return Decimal("0.00")
+
+    @staticmethod
+    def _percent(numerator: Decimal, denominator: Decimal) -> Decimal:
+        if denominator <= Decimal("0.00"):
+            return Decimal("0.00")
+        return ((numerator / denominator) * Decimal("100")).quantize(Decimal("0.01"))
+
+    @staticmethod
+    def _fmt(value: Decimal) -> str:
+        return f"{value.quantize(Decimal('0.01'))}"
+
diff --git a/server/src/app/services/budget_support.py b/server/src/app/services/budget_support.py
index 4acc809..4d01e1e 100644
--- a/server/src/app/services/budget_support.py
+++ b/server/src/app/services/budget_support.py
@@ -310,6 +310,14 @@ class BudgetSupportMixin:
             }
 
         balance = self.get_balance(allocation)
+        reservation_source_type = self._reservation_source_type_from_claim(claim)
+        current_reservation = self._find_active_reservation(
+            source_type=reservation_source_type,
+            source_id=claim.id,
+        )
+        current_reserved_amount = self._money(
+            current_reservation.amount if current_reservation is not None else Decimal("0.00")
+        )
         over_budget_amount = max(amount - balance.available_amount, Decimal("0.00"))
         return {
             "matched": True,
@@ -319,6 +327,7 @@ class BudgetSupportMixin:
             "claim_amount": str(amount),
             "total_amount": str(balance.total_amount),
             "reserved_amount": str(balance.reserved_amount),
+            "current_reserved_amount": str(current_reserved_amount),
             "consumed_amount": str(balance.consumed_amount),
             "available_amount": str(balance.available_amount),
             "usage_rate": str(balance.usage_rate),
@@ -335,6 +344,14 @@ class BudgetSupportMixin:
             "project_code": allocation.project_code,
         }
 
+    @staticmethod
+    def _reservation_source_type_from_claim(claim: ExpenseClaim) -> str:
+        claim_no = str(claim.claim_no or "").strip().upper()
+        expense_type = str(claim.expense_type or "").strip().lower()
+        if claim_no.startswith(("AP-", "APP-")) or expense_type == "application" or expense_type.endswith("_application"):
+            return "application"
+        return "claim"
+
     def _find_allocation_for_claim(self, claim: ExpenseClaim) -> BudgetAllocation | None:
         fiscal_year, period_key = self._period_from_claim(claim)
         return self._find_allocation_for_dimension(
diff --git a/server/src/app/services/expense_claim_access_policy.py b/server/src/app/services/expense_claim_access_policy.py
index 0d6e0e0..4eb04e1 100644
--- a/server/src/app/services/expense_claim_access_policy.py
+++ b/server/src/app/services/expense_claim_access_policy.py
@@ -10,14 +10,25 @@ from app.api.deps import CurrentUserContext
 from app.models.employee import Employee
 from app.models.financial_record import ExpenseClaim
 from app.models.organization import OrganizationUnit
+from app.models.role import Role
+from app.services.expense_claim_workflow_constants import (
+    APPROVAL_DONE_STAGE,
+    ARCHIVE_ACCOUNTING_STAGE,
+    BUDGET_MANAGER_APPROVAL_STAGE,
+    DIRECT_MANAGER_APPROVAL_STAGE,
+    FINANCE_APPROVAL_STAGE,
+)
 
 
 PRIVILEGED_CLAIM_ROLE_CODES = {"finance", "executive"}
 ARCHIVE_CENTER_ROLE_CODES = {"finance", "executive"}
 APPROVAL_VISIBLE_CLAIM_ROLE_CODES = {"manager", "approver"}
+BUDGET_APPROVAL_ROLE_CODES = {"budget_monitor", "executive"}
+BUDGET_MONITOR_ROLE_CODE = "budget_monitor"
+BUDGET_MONITOR_APPROVAL_GRADE = "P8"
 CLAIM_DELETE_ROLE_CODES = {"executive"}
 ARCHIVED_CLAIM_STATUSES = ("approved", "completed", "paid")
-APPLICATION_ARCHIVED_STAGES = ("审批完成", "申请归档", "completed")
+APPLICATION_ARCHIVED_STAGES = (APPROVAL_DONE_STAGE, "申请归档", "completed")
 
 
 class ExpenseClaimAccessPolicy:
@@ -49,7 +60,7 @@ class ExpenseClaimAccessPolicy:
             normalized_type.like("%\\_application", escape="\\"),
         )
         return or_(
-            stage == "归档入账",
+            stage == ARCHIVE_ACCOUNTING_STAGE,
             stage == "completed",
             and_(
                 application_condition,
@@ -61,7 +72,7 @@ class ExpenseClaimAccessPolicy:
                 or_(
                     stage == "",
                     stage.is_(None),
-                    stage == "归档入账",
+                    stage == ARCHIVE_ACCOUNTING_STAGE,
                     stage == "completed",
                 ),
             ),
@@ -77,7 +88,7 @@ class ExpenseClaimAccessPolicy:
     def is_archived_claim(claim: ExpenseClaim) -> bool:
         normalized_status = str(claim.status or "").strip().lower()
         stage = str(claim.approval_stage or "").strip()
-        if stage in {"归档入账", "completed"}:
+        if stage in {ARCHIVE_ACCOUNTING_STAGE, "completed"}:
             return True
         normalized_type = str(claim.expense_type or "").strip().lower()
         claim_no = str(claim.claim_no or "").strip().upper()
@@ -92,7 +103,7 @@ class ExpenseClaimAccessPolicy:
             and stage in APPLICATION_ARCHIVED_STAGES
         ):
             return True
-        return normalized_status in ARCHIVED_CLAIM_STATUSES and stage in {"", "归档入账", "completed"}
+        return normalized_status in ARCHIVED_CLAIM_STATUSES and stage in {"", ARCHIVE_ACCOUNTING_STAGE, "completed"}
 
     def can_return_claim(self, current_user: CurrentUserContext, claim: ExpenseClaim) -> bool:
         normalized_status = str(claim.status or "").strip().lower()
@@ -100,9 +111,11 @@ class ExpenseClaimAccessPolicy:
             return False
 
         stage = str(claim.approval_stage or "").strip()
-        if stage == "直属领导审批":
+        if stage == DIRECT_MANAGER_APPROVAL_STAGE:
             return self.is_current_direct_manager_approver(current_user, claim)
-        if stage == "财务审批":
+        if stage == BUDGET_MANAGER_APPROVAL_STAGE:
+            return self.is_budget_manager_approver(current_user, claim)
+        if stage == FINANCE_APPROVAL_STAGE:
             return self.has_privileged_claim_access(current_user) and not self.is_claim_owned_by_current_user(
                 claim,
                 current_user,
@@ -111,9 +124,11 @@ class ExpenseClaimAccessPolicy:
 
     def can_approve_claim(self, current_user: CurrentUserContext, claim: ExpenseClaim) -> bool:
         stage = str(claim.approval_stage or "").strip()
-        if stage == "直属领导审批":
+        if stage == DIRECT_MANAGER_APPROVAL_STAGE:
             return self.is_current_direct_manager_approver(current_user, claim)
-        if stage == "财务审批":
+        if stage == BUDGET_MANAGER_APPROVAL_STAGE:
+            return self.is_budget_manager_approver(current_user, claim)
+        if stage == FINANCE_APPROVAL_STAGE:
             role_codes = self.normalize_role_codes(current_user)
             return (
                 (current_user.is_admin or "finance" in role_codes)
@@ -127,7 +142,7 @@ class ExpenseClaimAccessPolicy:
             return False
         if str(claim.status or "").strip().lower() != "submitted":
             return False
-        if str(claim.approval_stage or "").strip() != "直属领导审批":
+        if str(claim.approval_stage or "").strip() != DIRECT_MANAGER_APPROVAL_STAGE:
             return False
 
         current_employee = self.resolve_current_employee(current_user)
@@ -149,6 +164,65 @@ class ExpenseClaimAccessPolicy:
 
         return self.resolve_claim_manager_name(claim) == approver_name
 
+    def is_budget_manager_approver(self, current_user: CurrentUserContext, claim: ExpenseClaim) -> bool:
+        if str(claim.status or "").strip().lower() != "submitted":
+            return False
+        if str(claim.approval_stage or "").strip() != BUDGET_MANAGER_APPROVAL_STAGE:
+            return False
+        if self.is_claim_owned_by_current_user(claim, current_user):
+            return False
+        if current_user.is_admin:
+            return True
+        role_codes = self.normalize_role_codes(current_user)
+        if "executive" in role_codes:
+            return True
+        return self.is_department_p8_budget_monitor(current_user, claim)
+
+    def is_budget_manager_user(self, current_user: CurrentUserContext) -> bool:
+        if current_user.is_admin:
+            return True
+        role_codes = self.normalize_role_codes(current_user)
+        return bool(role_codes & BUDGET_APPROVAL_ROLE_CODES)
+
+    def is_department_p8_budget_monitor(self, current_user: CurrentUserContext, claim: ExpenseClaim) -> bool:
+        role_codes = self.normalize_role_codes(current_user)
+        if BUDGET_MONITOR_ROLE_CODE not in role_codes:
+            return False
+
+        current_employee = self.resolve_current_employee(current_user)
+        if current_employee is None:
+            return False
+        if not self._employee_has_budget_approval_grade(current_employee):
+            return False
+
+        return self._employee_matches_claim_department(current_employee, current_user, claim)
+
+    def resolve_department_budget_manager(self, claim: ExpenseClaim) -> Employee | None:
+        department_ids, department_names = self._collect_claim_department_identity(claim)
+        department_conditions = []
+        if department_ids:
+            department_conditions.append(Employee.organization_unit_id.in_(department_ids))
+        if department_names:
+            department_conditions.append(Employee.organization_unit.has(OrganizationUnit.name.in_(department_names)))
+        if not department_conditions:
+            return None
+
+        stmt = (
+            select(Employee)
+            .options(selectinload(Employee.organization_unit), selectinload(Employee.roles))
+            .where(
+                func.upper(func.coalesce(Employee.grade, "")) == BUDGET_MONITOR_APPROVAL_GRADE,
+                Employee.roles.any(Role.role_code == BUDGET_MONITOR_ROLE_CODE),
+                or_(*department_conditions),
+            )
+            .order_by(Employee.name.asc(), Employee.employee_no.asc())
+            .limit(1)
+        )
+        claim_employee_id = str(claim.employee_id or "").strip()
+        if claim_employee_id:
+            stmt = stmt.where(Employee.id != claim_employee_id)
+        return self.db.scalar(stmt)
+
     @staticmethod
     def normalize_role_codes(current_user: CurrentUserContext) -> set[str]:
         return {
@@ -157,6 +231,51 @@ class ExpenseClaimAccessPolicy:
             if str(item).strip()
         }
 
+    @staticmethod
+    def _employee_has_budget_approval_grade(employee: Employee) -> bool:
+        return str(employee.grade or "").strip().upper() == BUDGET_MONITOR_APPROVAL_GRADE
+
+    def _employee_matches_claim_department(
+        self,
+        employee: Employee,
+        current_user: CurrentUserContext,
+        claim: ExpenseClaim,
+    ) -> bool:
+        claim_department_ids, claim_department_names = self._collect_claim_department_identity(claim)
+        employee_department_ids = {
+            str(employee.organization_unit_id or "").strip(),
+        }
+        employee_department_names = {
+            str(current_user.department_name or "").strip(),
+        }
+        if employee.organization_unit is not None:
+            employee_department_names.add(str(employee.organization_unit.name or "").strip())
+
+        employee_department_ids.discard("")
+        employee_department_names.discard("")
+        return bool(
+            (claim_department_ids and employee_department_ids & claim_department_ids)
+            or (claim_department_names and employee_department_names & claim_department_names)
+        )
+
+    def _collect_claim_department_identity(self, claim: ExpenseClaim) -> tuple[set[str], set[str]]:
+        department_ids = {
+            str(claim.department_id or "").strip(),
+        }
+        department_names = {
+            str(claim.department_name or "").strip(),
+        }
+
+        claim_employee = self.resolve_claim_employee_for_backfill(claim)
+        if claim_employee is not None:
+            department_ids.add(str(claim_employee.organization_unit_id or "").strip())
+            if claim_employee.organization_unit is not None:
+                department_names.add(str(claim_employee.organization_unit.name or "").strip())
+
+        department_ids.discard("")
+        department_names.discard("")
+        return department_ids, department_names
+
     def resolve_current_employee(self, current_user: CurrentUserContext) -> Employee | None:
         return self.resolve_employee_by_identity_candidates(
             [
@@ -375,7 +494,7 @@ class ExpenseClaimAccessPolicy:
         ).strip()
         pending_leader_approval_parts = [
             ExpenseClaim.status == "submitted",
-            ExpenseClaim.approval_stage == "直属领导审批",
+            ExpenseClaim.approval_stage == DIRECT_MANAGER_APPROVAL_STAGE,
         ]
         if employee is not None:
             pending_leader_approval_parts.append(
@@ -399,17 +518,55 @@ class ExpenseClaimAccessPolicy:
 
         return conditions
 
+    def build_budget_approval_claim_conditions(self, current_user: CurrentUserContext) -> list[Any]:
+        role_codes = self.normalize_role_codes(current_user)
+        if BUDGET_MONITOR_ROLE_CODE not in role_codes:
+            return []
+        employee = self.resolve_current_employee(current_user)
+        if employee is None or not self._employee_has_budget_approval_grade(employee):
+            return []
+
+        pending_budget_approval_parts = [
+            ExpenseClaim.status == "submitted",
+            ExpenseClaim.approval_stage == BUDGET_MANAGER_APPROVAL_STAGE,
+        ]
+        pending_budget_approval_parts.append(
+            or_(ExpenseClaim.employee_id.is_(None), ExpenseClaim.employee_id != employee.id)
+        )
+        if employee.name:
+            pending_budget_approval_parts.append(ExpenseClaim.employee_name != employee.name)
+
+        department_conditions = []
+        department_name = str(current_user.department_name or "").strip()
+        if employee.organization_unit_id:
+            department_conditions.append(ExpenseClaim.department_id == employee.organization_unit_id)
+            subordinate_department_employee_ids = select(Employee.id).where(
+                Employee.organization_unit_id == employee.organization_unit_id
+            )
+            department_conditions.append(ExpenseClaim.employee_id.in_(subordinate_department_employee_ids))
+        if employee.organization_unit is not None and employee.organization_unit.name:
+            department_conditions.append(ExpenseClaim.department_name == employee.organization_unit.name)
+        if department_name:
+            department_conditions.append(ExpenseClaim.department_name == department_name)
+        if not department_conditions:
+            return []
+
+        pending_budget_approval_parts.append(or_(*department_conditions))
+        return [and_(*pending_budget_approval_parts)]
+
     def apply_approval_claim_scope(self, stmt: Any, current_user: CurrentUserContext) -> Any:
         role_codes = self.normalize_role_codes(current_user)
         if current_user.is_admin or "executive" in role_codes:
             return stmt.where(ExpenseClaim.status == "submitted")
+        conditions = []
         if "finance" in role_codes:
-            return stmt.where(
+            conditions.append(and_(
                 ExpenseClaim.status == "submitted",
-                ExpenseClaim.approval_stage == "财务审批",
-            )
+                ExpenseClaim.approval_stage == FINANCE_APPROVAL_STAGE,
+            ))
 
-        conditions = self.build_approval_claim_conditions(current_user)
+        conditions.extend(self.build_budget_approval_claim_conditions(current_user))
+        conditions.extend(self.build_approval_claim_conditions(current_user))
         if not conditions:
             return stmt.where(ExpenseClaim.id == "__no_visible_claim__")
 
@@ -440,6 +597,7 @@ class ExpenseClaimAccessPolicy:
             return stmt.where(ExpenseClaim.id == "__no_visible_claim__")
 
         if include_approval_scope:
+            conditions.extend(self.build_budget_approval_claim_conditions(current_user))
             conditions.extend(self.build_approval_claim_conditions(current_user))
 
         return stmt.where(or_(*conditions))
diff --git a/server/src/app/services/expense_claim_application_handoff.py b/server/src/app/services/expense_claim_application_handoff.py
index baab82b..f3327cf 100644
--- a/server/src/app/services/expense_claim_application_handoff.py
+++ b/server/src/app/services/expense_claim_application_handoff.py
@@ -68,7 +68,10 @@ class ExpenseClaimApplicationHandoffMixin:
                     "application_claim_no": application_claim.claim_no,
                     "application_budget_amount": str(application_claim.amount or Decimal("0.00")),
                     "application_approval_event_id": str(approval_flag.get("approval_event_id") or ""),
-                    "leader_opinion": str(approval_flag.get("opinion") or "").strip(),
+                    "leader_opinion": str(
+                        approval_flag.get("leader_opinion") or approval_flag.get("opinion") or ""
+                    ).strip(),
+                    "budget_opinion": str(approval_flag.get("budget_opinion") or "").strip(),
                     "created_at": created_at.isoformat(),
                 }
             ],
diff --git a/server/src/app/services/expense_claim_approval_flow.py b/server/src/app/services/expense_claim_approval_flow.py
new file mode 100644
index 0000000..4ff3a7e
--- /dev/null
+++ b/server/src/app/services/expense_claim_approval_flow.py
@@ -0,0 +1,173 @@
+from __future__ import annotations
+
+import uuid
+from datetime import UTC, datetime
+from typing import Any
+
+from app.api.deps import CurrentUserContext
+from app.services.expense_claim_workflow_constants import (
+    APPROVAL_DONE_STAGE,
+    ARCHIVE_ACCOUNTING_STAGE,
+    BUDGET_MANAGER_APPROVAL_STAGE,
+    DIRECT_MANAGER_APPROVAL_STAGE,
+    FINANCE_APPROVAL_STAGE,
+)
+
+
+class ExpenseClaimApprovalFlowMixin:
+    def approve_claim(
+        self,
+        claim_id: str,
+        current_user: CurrentUserContext,
+        *,
+        opinion: str | None = None,
+    ):
+        claim = self.get_claim(claim_id, current_user)
+        if claim is None:
+            return None
+
+        normalized_status = str(claim.status or "").strip().lower()
+        if normalized_status != "submitted":
+            raise ValueError("只有审批中的单据可以审批通过。")
+
+        previous_stage = str(claim.approval_stage or "").strip()
+        is_application_claim = self._is_expense_application_claim(claim)
+        next_budget_manager = None
+        if previous_stage == DIRECT_MANAGER_APPROVAL_STAGE:
+            if not self._access_policy.can_approve_claim(current_user, claim):
+                raise ValueError("只有当前直属领导审批人可以审批通过该单据。")
+            approval_source = "manual_approval"
+            event_type = "expense_application_approval" if is_application_claim else "expense_claim_approval"
+            label = "领导审批通过"
+            if is_application_claim:
+                next_budget_manager = self._access_policy.resolve_department_budget_manager(claim)
+                next_status = "submitted"
+                next_stage = BUDGET_MANAGER_APPROVAL_STAGE
+                default_message = "{operator} 已确认直属领导审核，流转至预算管理者审批。"
+            else:
+                next_status = "submitted"
+                next_stage = FINANCE_APPROVAL_STAGE
+                default_message = "{operator} 已审批通过，流转至{next_stage}。"
+        elif previous_stage == BUDGET_MANAGER_APPROVAL_STAGE:
+            if not is_application_claim:
+                raise ValueError("只有费用申请需要预算管理者审批。")
+            if not self._access_policy.can_approve_claim(current_user, claim):
+                raise ValueError("只有当前预算管理者可以审批通过该费用申请。")
+            approval_source = "budget_approval"
+            event_type = "expense_application_budget_approval"
+            label = "预算管理者审核通过"
+            next_status = "approved"
+            next_stage = APPROVAL_DONE_STAGE
+            default_message = "{operator} 已完成预算审核，申请流程完成并生成报销草稿。"
+        elif previous_stage == FINANCE_APPROVAL_STAGE:
+            if is_application_claim:
+                raise ValueError("费用申请需先完成预算管理者审批。")
+            if not self._access_policy.can_approve_claim(current_user, claim):
+                raise ValueError("只有财务人员可以完成财务终审。")
+            approval_source = "finance_approval"
+            event_type = "expense_claim_finance_approval"
+            label = "财务审核通过"
+            next_status = "approved"
+            next_stage = ARCHIVE_ACCOUNTING_STAGE
+            default_message = "{operator} 已完成财务审核，进入归档入账。"
+        else:
+            raise ValueError("当前节点不支持审批通过。")
+
+        approval_opinion = str(opinion or "").strip()
+        if previous_stage in {DIRECT_MANAGER_APPROVAL_STAGE, BUDGET_MANAGER_APPROVAL_STAGE} and not approval_opinion:
+            approval_opinion = "同意"
+
+        before_json = self._serialize_claim(claim)
+        operator = self._access_policy.resolve_current_user_display_name(current_user)
+        budget_flags: list[dict[str, Any]] = []
+        if approval_source == "finance_approval" and not is_application_claim:
+            consumed_budget_flag = self._consume_budget_for_finance_approval(claim, current_user)
+            if consumed_budget_flag is not None:
+                budget_flags.append(consumed_budget_flag)
+        approval_flag = {
+            "source": approval_source,
+            "event_type": event_type,
+            "approval_event_id": str(uuid.uuid4()),
+            "severity": "info",
+            "label": label,
+            "message": approval_opinion or default_message.format(operator=operator, next_stage=next_stage),
+            "opinion": approval_opinion,
+            "operator": operator,
+            "operator_username": current_user.username,
+            "operator_role_codes": [
+                str(item).strip().lower()
+                for item in current_user.role_codes
+                if str(item).strip()
+            ],
+            "previous_status": str(claim.status or "").strip(),
+            "previous_approval_stage": previous_stage,
+            "next_status": next_status,
+            "next_approval_stage": next_stage,
+            "created_at": datetime.now(UTC).isoformat(),
+        }
+        if next_budget_manager is not None:
+            approval_flag.update(
+                {
+                    "next_approver_name": str(next_budget_manager.name or "").strip(),
+                    "next_approver_employee_id": next_budget_manager.id,
+                    "next_approver_grade": str(next_budget_manager.grade or "").strip(),
+                    "next_approver_role_code": "budget_monitor",
+                }
+            )
+
+        claim.status = next_status
+        claim.approval_stage = next_stage
+        if claim.submitted_at is None:
+            claim.submitted_at = datetime.now(UTC)
+        if is_application_claim and previous_stage == BUDGET_MANAGER_APPROVAL_STAGE:
+            approval_flag["leader_opinion"] = self._resolve_latest_approval_opinion(
+                claim,
+                source="manual_approval",
+            )
+            approval_flag["budget_opinion"] = approval_opinion
+            generated_draft = self._create_reimbursement_draft_from_application(
+                application_claim=claim,
+                approval_flag=approval_flag,
+                operator=operator,
+            )
+            transferred_budget_flag = self._transfer_application_budget_to_reimbursement(
+                application_claim=claim,
+                draft_claim=generated_draft,
+                current_user=current_user,
+            )
+            if transferred_budget_flag is not None:
+                budget_flags.append(transferred_budget_flag)
+                generated_draft.risk_flags_json = self._append_budget_flags(
+                    generated_draft.risk_flags_json,
+                    transferred_budget_flag,
+                )
+        claim.risk_flags_json = self._append_budget_flags(
+            [*list(claim.risk_flags_json or []), approval_flag],
+            budget_flags,
+        )
+
+        self.db.commit()
+        self.db.refresh(claim)
+
+        self.audit_service.log_action(
+            actor=operator,
+            action="expense_claim.approve",
+            resource_type="expense_claim",
+            resource_id=claim.id,
+            before_json=before_json,
+            after_json=self._serialize_claim(claim),
+        )
+
+        return claim
+
+    @staticmethod
+    def _resolve_latest_approval_opinion(claim, *, source: str) -> str:
+        for flag in reversed(list(claim.risk_flags_json or [])):
+            if not isinstance(flag, dict):
+                continue
+            if str(flag.get("source") or "").strip() != source:
+                continue
+            opinion = str(flag.get("opinion") or flag.get("message") or "").strip()
+            if opinion:
+                return opinion
+        return ""
diff --git a/server/src/app/services/expense_claim_constants.py b/server/src/app/services/expense_claim_constants.py
index 3f5fba4..1ea7f84 100644
--- a/server/src/app/services/expense_claim_constants.py
+++ b/server/src/app/services/expense_claim_constants.py
@@ -225,6 +225,7 @@ RETURN_REASON_OPTIONS = {
     "application_budget_basis_missing": "预算测算依据不足",
     "application_policy_mismatch": "制度口径不匹配",
     "application_attachment_needed": "前置材料需补充",
+    "application_other": "其他",
 }
 MAX_CLAIM_NO_RETRY_ATTEMPTS = 3
 DOCUMENT_DATE_PATTERN = re.compile(
diff --git a/server/src/app/services/expense_claim_item_sync.py b/server/src/app/services/expense_claim_item_sync.py
index 8af4bf7..c3650fd 100644
--- a/server/src/app/services/expense_claim_item_sync.py
+++ b/server/src/app/services/expense_claim_item_sync.py
@@ -12,6 +12,7 @@ from sqlalchemy import inspect as sqlalchemy_inspect
 from app.api.deps import CurrentUserContext
 from app.core.agent_enums import AgentAssetDomain, AgentAssetStatus, AgentAssetType
 from app.models.agent_asset import AgentAsset
+from app.models.employee import Employee
 from app.models.financial_record import ExpenseClaim, ExpenseClaimItem
 from app.schemas.reimbursement import TravelReimbursementCalculatorRequest
 from app.services.agent_asset_rule_library import AgentAssetRuleLibraryManager
@@ -50,7 +51,7 @@ class ExpenseClaimItemSyncMixin:
                 self._discard_claim_item(claim, item)
             return
 
-        grade = str(claim.employee_grade or "").strip()
+        grade = self._resolve_claim_employee_grade(claim)
         if not grade:
             return
 
@@ -115,6 +116,16 @@ class ExpenseClaimItemSyncMixin:
         item.item_amount = allowance_amount
         item.invoice_id = None
 
+    def _resolve_claim_employee_grade(self, claim: ExpenseClaim) -> str:
+        grade = str(claim.employee_grade or "").strip()
+        if grade:
+            return grade
+        employee_id = str(claim.employee_id or "").strip()
+        if not employee_id:
+            return ""
+        employee = self.db.get(Employee, employee_id)
+        return str(employee.grade if employee is not None and employee.grade else "").strip()
+
     def _discard_claim_item(self, claim: ExpenseClaim, item: ExpenseClaimItem) -> None:
         if item in claim.items:
             claim.items.remove(item)
diff --git a/server/src/app/services/expense_claim_read_model.py b/server/src/app/services/expense_claim_read_model.py
index 89cff40..5bed930 100644
--- a/server/src/app/services/expense_claim_read_model.py
+++ b/server/src/app/services/expense_claim_read_model.py
@@ -204,6 +204,8 @@ class ExpenseClaimReadModelMixin:
         normalized = str(stage or "").strip()
         if "直属" in normalized or "领导" in normalized or "负责人" in normalized:
             return "direct_manager"
+        if "预算" in normalized:
+            return "budget"
         if "财务" in normalized:
             return "finance"
         if "AI" in normalized or "预审" in normalized:
diff --git a/server/src/app/services/expense_claim_workflow_constants.py b/server/src/app/services/expense_claim_workflow_constants.py
new file mode 100644
index 0000000..74d3ad7
--- /dev/null
+++ b/server/src/app/services/expense_claim_workflow_constants.py
@@ -0,0 +1,6 @@
+DIRECT_MANAGER_APPROVAL_STAGE = "直属领导审批"
+BUDGET_MANAGER_APPROVAL_STAGE = "预算管理者审批"
+FINANCE_APPROVAL_STAGE = "财务审批"
+APPROVAL_DONE_STAGE = "审批完成"
+ARCHIVE_ACCOUNTING_STAGE = "归档入账"
+
diff --git a/server/src/app/services/expense_claims.py b/server/src/app/services/expense_claims.py
index 274930f..850f5e6 100644
--- a/server/src/app/services/expense_claims.py
+++ b/server/src/app/services/expense_claims.py
@@ -35,6 +35,7 @@ from app.services.audit import AuditLogService
 from app.services.document_intelligence import build_document_insight
 from app.services.document_numbering import is_application_claim_no
 from app.services.expense_claim_access_policy import ExpenseClaimAccessPolicy
+from app.services.expense_claim_approval_flow import ExpenseClaimApprovalFlowMixin
 from app.services.expense_claim_attachment_presentation import ExpenseClaimAttachmentPresentation
 from app.services.expense_claim_attachment_storage import ExpenseClaimAttachmentStorage
 from app.services.expense_claim_application_handoff import ExpenseClaimApplicationHandoffMixin
@@ -127,6 +128,7 @@ from app.services.ocr import OcrService
 
 
 class ExpenseClaimService(
+    ExpenseClaimApprovalFlowMixin,
     ExpenseClaimApplicationHandoffMixin,
     ExpenseClaimBudgetFlowMixin,
     ExpenseClaimAttachmentOperationsMixin,
@@ -234,6 +236,18 @@ class ExpenseClaimService(
         stmt = self._access_policy.apply_claim_scope(stmt, current_user, include_approval_scope=True)
         return self.db.scalar(stmt)
 
+    def can_view_budget_analysis(self, current_user: CurrentUserContext, claim: ExpenseClaim | None = None) -> bool:
+        if claim is None:
+            return self._access_policy.is_budget_manager_user(current_user)
+        if current_user.is_admin:
+            return True
+        role_codes = self._access_policy.normalize_role_codes(current_user)
+        if "executive" in role_codes:
+            return True
+        if self._access_policy.is_claim_owned_by_current_user(claim, current_user):
+            return False
+        return self._access_policy.is_department_p8_budget_monitor(current_user, claim)
+
     def update_claim(
         self,
         *,
@@ -562,9 +576,6 @@ class ExpenseClaimService(
         if claim is None:
             return None
 
-        if not self._access_policy.can_return_claim(current_user, claim):
-            raise ValueError("只有财务人员、高级财务人员或当前审批人可以退回报销单。")
-
         normalized_status = str(claim.status or "").strip().lower()
         if normalized_status == "draft":
             raise ValueError("草稿状态无需退回。")
@@ -573,6 +584,9 @@ class ExpenseClaimService(
         if normalized_status in {"approved", "completed", "paid"}:
             raise ValueError("已完成单据不允许退回。")
 
+        if not self._access_policy.can_return_claim(current_user, claim):
+            raise ValueError("只有财务人员、高级财务人员或当前审批人可以退回报销单。")
+
         before_json = self._serialize_claim(claim)
         operator = self._access_policy.resolve_current_user_display_name(current_user)
         previous_status = str(claim.status or "").strip()
@@ -580,21 +594,25 @@ class ExpenseClaimService(
         previous_stage_key = self._normalize_return_stage_key(previous_stage)
         is_application_claim = self._is_expense_application_claim(claim)
         is_direct_manager_return = previous_stage_key == "direct_manager"
+        is_budget_return = previous_stage_key == "budget"
+        is_application_return = is_application_claim and (is_direct_manager_return or is_budget_return)
         return_event_type = (
             "expense_application_return"
-            if is_application_claim and is_direct_manager_return
+            if is_application_return
             else "expense_claim_return"
         )
         return_label = (
             "领导退回"
             if is_application_claim and is_direct_manager_return
+            else "预算退回"
+            if is_application_claim and is_budget_return
             else "人工退回"
         )
         return_reason = str(reason or "").strip()
         reason_code_payload = self._normalize_return_reason_code_payload(reason_codes)
         normalized_reason_codes = reason_code_payload["reason_codes"]
         unknown_reason_codes = reason_code_payload["unknown_reason_codes"]
-        if is_application_claim and is_direct_manager_return and not any(
+        if is_application_return and not any(
             code.startswith("application_") for code in normalized_reason_codes
         ):
             raise ValueError("申请单退回必须选择至少一个退单类型。")
@@ -627,6 +645,7 @@ class ExpenseClaimService(
             "reason": return_reason,
             "opinion": message,
             "leader_opinion": message if is_application_claim and is_direct_manager_return else "",
+            "budget_opinion": message if is_application_claim and is_budget_return else "",
             "reason_codes": normalized_reason_codes,
             "risk_points": risk_points,
             "operator": operator,
@@ -676,204 +695,6 @@ class ExpenseClaimService(
 
         return claim
 
-    def approve_claim(
-        self,
-        claim_id: str,
-        current_user: CurrentUserContext,
-        *,
-        opinion: str | None = None,
-    ) -> ExpenseClaim | None:
-        claim = self.get_claim(claim_id, current_user)
-        if claim is None:
-            return None
-
-        normalized_status = str(claim.status or "").strip().lower()
-        if normalized_status != "submitted":
-            raise ValueError("只有审批中的单据可以审批通过。")
-
-        previous_stage = str(claim.approval_stage or "").strip()
-        is_application_claim = self._is_expense_application_claim(claim)
-        if previous_stage == "直属领导审批":
-            if not self._access_policy.can_approve_claim(current_user, claim):
-                raise ValueError("只有当前直属领导审批人可以审批通过该单据。")
-            approval_source = "manual_approval"
-            if is_application_claim:
-                event_type = "expense_application_approval"
-                label = "领导审批通过"
-                next_status = "approved"
-                next_stage = "审批完成"
-                default_message = "{operator} 已确认审核，申请流程完成并生成报销草稿。"
-            else:
-                event_type = "expense_claim_approval"
-                label = "领导审批通过"
-                next_status = "submitted"
-                next_stage = "财务审批"
-                default_message = "{operator} 已审批通过，流转至{next_stage}。"
-        elif previous_stage == "财务审批":
-            if is_application_claim:
-                raise ValueError("费用申请无需财务审批，直属领导审批通过后即完成。")
-            if not self._access_policy.can_approve_claim(current_user, claim):
-                raise ValueError("只有财务人员可以完成财务终审。")
-            approval_source = "finance_approval"
-            event_type = "expense_claim_finance_approval"
-            label = "财务审核通过"
-            next_status = "approved"
-            next_stage = "归档入账"
-            default_message = "{operator} 已完成财务审核，进入归档入账。"
-        else:
-            raise ValueError("当前节点不支持审批通过。")
-
-        approval_opinion = str(opinion or "").strip()
-        if previous_stage == "直属领导审批" and not approval_opinion:
-            raise ValueError("领导审核意见不能为空，请填写意见后再确认审核。")
-
-        before_json = self._serialize_claim(claim)
-        operator = self._access_policy.resolve_current_user_display_name(current_user)
-        budget_flags: list[dict[str, Any]] = []
-        if approval_source == "finance_approval" and not is_application_claim:
-            consumed_budget_flag = self._consume_budget_for_finance_approval(claim, current_user)
-            if consumed_budget_flag is not None:
-                budget_flags.append(consumed_budget_flag)
-        approval_flag = {
-            "source": approval_source,
-            "event_type": event_type,
-            "approval_event_id": str(uuid.uuid4()),
-            "severity": "info",
-            "label": label,
-            "message": approval_opinion or default_message.format(operator=operator, next_stage=next_stage),
-            "opinion": approval_opinion,
-            "operator": operator,
-            "operator_username": current_user.username,
-            "operator_role_codes": [
-                str(item).strip().lower()
-                for item in current_user.role_codes
-                if str(item).strip()
-            ],
-            "previous_status": str(claim.status or "").strip(),
-            "previous_approval_stage": previous_stage,
-            "next_status": next_status,
-            "next_approval_stage": next_stage,
-            "created_at": datetime.now(UTC).isoformat(),
-        }
-
-        claim.status = next_status
-        claim.approval_stage = next_stage
-        if claim.submitted_at is None:
-            claim.submitted_at = datetime.now(UTC)
-        if is_application_claim and previous_stage == "直属领导审批":
-            generated_draft = self._create_reimbursement_draft_from_application(
-                application_claim=claim,
-                approval_flag=approval_flag,
-                operator=operator,
-            )
-            transferred_budget_flag = self._transfer_application_budget_to_reimbursement(
-                application_claim=claim,
-                draft_claim=generated_draft,
-                current_user=current_user,
-            )
-            if transferred_budget_flag is not None:
-                budget_flags.append(transferred_budget_flag)
-                generated_draft.risk_flags_json = self._append_budget_flags(
-                    generated_draft.risk_flags_json,
-                    transferred_budget_flag,
-                )
-        claim.risk_flags_json = self._append_budget_flags(
-            [*list(claim.risk_flags_json or []), approval_flag],
-            budget_flags,
-        )
-
-        self.db.commit()
-        self.db.refresh(claim)
-
-        self.audit_service.log_action(
-            actor=operator,
-            action="expense_claim.approve",
-            resource_type="expense_claim",
-            resource_id=claim.id,
-            before_json=before_json,
-            after_json=self._serialize_claim(claim),
-        )
-
-        return claim
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
 
 
 
diff --git a/server/src/app/services/knowledge.py b/server/src/app/services/knowledge.py
index 48cc4a9..d4dc797 100644
--- a/server/src/app/services/knowledge.py
+++ b/server/src/app/services/knowledge.py
@@ -602,7 +602,15 @@ class KnowledgeService:
             status_payload = status_map.get(document_id) or {}
             rag_status = str(status_payload.get("status") or "").strip().lower()
             linked_run_status = resolve_linked_ingest_run_status(entry, db=self.db)
-            if linked_run_status == AgentRunStatus.FAILED.value and rag_status in {
+            if not status_payload:
+                if (
+                    current_status == KNOWLEDGE_INGEST_STATUS_SYNCING
+                    and linked_run_status == AgentRunStatus.FAILED.value
+                ):
+                    desired_status = KNOWLEDGE_INGEST_STATUS_FAILED
+                else:
+                    continue
+            elif linked_run_status == AgentRunStatus.FAILED.value and rag_status in {
                 "pending",
                 "processing",
                 "preprocessed",
diff --git a/server/src/app/services/knowledge_rag.py b/server/src/app/services/knowledge_rag.py
index 684aaf7..2d36fd9 100644
--- a/server/src/app/services/knowledge_rag.py
+++ b/server/src/app/services/knowledge_rag.py
@@ -4,8 +4,9 @@ import os
 import re
 import socket
 import threading
+from concurrent.futures import ThreadPoolExecutor
 from pathlib import Path
-from typing import Any
+from typing import Any, Callable
 
 from sqlalchemy.orm import Session
 
@@ -89,8 +90,10 @@ STRUCTURED_APPENDIX_LEADING_MARKERS = (
 )
 STRUCTURED_APPENDIX_LEADING_WINDOW = 220
 _runtime_lock = threading.RLock()
-_runtime_instances: dict[int, _LightRagRuntime] = {}
-_runtime_signatures: dict[int, tuple[Any, ...]] = {}
+_runtime_executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="knowledge-rag-runtime")
+_runtime_instances: dict[str, _LightRagRuntime] = {}
+_runtime_signatures: dict[str, tuple[Any, ...]] = {}
+_RUNTIME_CACHE_KEY = "lightrag"
 
 
 class KnowledgeRagService:
@@ -133,21 +136,26 @@ class KnowledgeRagService:
 
         runtime_hits: list[dict[str, Any]] = []
         runtime_references: list[str] = []
-        try:
-            runtime = self._get_runtime()
-            raw = runtime.query_data(rewritten_query, conversation_history=conversation_history)
-            data = raw.get("data") if isinstance(raw, dict) else {}
-            chunks = list(data.get("chunks") or []) if isinstance(data, dict) else []
-            entities = list(data.get("entities") or []) if isinstance(data, dict) else []
-            runtime_references = list(data.get("references") or []) if isinstance(data, dict) else []
-            runtime_hits = self._build_hits_from_query_data(
-                query=rewritten_query,
-                chunks=chunks,
-                entities=entities,
-                limit=limit,
-            )
-        except Exception as exc:
-            logger.warning("Knowledge query failed: %s", exc)
+        if not local_result.confident:
+            try:
+                raw = self._run_runtime_operation(
+                    lambda runtime: runtime.query_data(
+                        rewritten_query,
+                        conversation_history=conversation_history,
+                    )
+                )
+                data = raw.get("data") if isinstance(raw, dict) else {}
+                chunks = list(data.get("chunks") or []) if isinstance(data, dict) else []
+                entities = list(data.get("entities") or []) if isinstance(data, dict) else []
+                runtime_references = list(data.get("references") or []) if isinstance(data, dict) else []
+                runtime_hits = self._build_hits_from_query_data(
+                    query=rewritten_query,
+                    chunks=chunks,
+                    entities=entities,
+                    limit=limit,
+                )
+            except Exception as exc:
+                logger.warning("Knowledge query failed: %s", exc)
 
         all_hits: dict[str, dict[str, Any]] = {}
         for hit in local_result.hits:
@@ -189,7 +197,7 @@ class KnowledgeRagService:
             ],
             "raw_references": runtime_references,
             "metadata": {
-                "retrieval_strategy": "fusion",
+                "retrieval_strategy": "fusion" if runtime_hits else "local_text_chunks",
                 "local_total_chunks": local_result.total_chunks,
                 "local_best_score": local_result.best_score,
             },
@@ -244,14 +252,17 @@ class KnowledgeRagService:
         file_paths: list[str] = []
         document_summaries: list[dict[str, Any]] = []
 
-        runtime = self._get_runtime()
-        existing_statuses = runtime.get_document_statuses(normalized_ids)
+        existing_statuses = self._run_runtime_operation(
+            lambda runtime: runtime.get_document_statuses(normalized_ids)
+        )
 
         for document_id in normalized_ids:
             entry = knowledge_service.get_document_entry(document_id)
             if force and document_id in existing_statuses:
                 try:
-                    runtime.delete_document(document_id)
+                    self._run_runtime_operation(
+                        lambda runtime, target_id=document_id: runtime.delete_document(target_id)
+                    )
                 except Exception as exc:
                     logger.warning(
                         "Delete existing LightRAG document failed doc_id=%s: %s", document_id, exc
@@ -277,13 +288,17 @@ class KnowledgeRagService:
                 )
             )
 
-        track_id = runtime.insert_documents(
-            texts=texts,
-            document_ids=normalized_ids,
-            file_paths=file_paths,
+        track_id = self._run_runtime_operation(
+            lambda runtime: runtime.insert_documents(
+                texts=texts,
+                document_ids=normalized_ids,
+                file_paths=file_paths,
+            )
         )
 
-        statuses = runtime.get_document_statuses(normalized_ids)
+        statuses = self._run_runtime_operation(
+            lambda runtime: runtime.get_document_statuses(normalized_ids)
+        )
         succeeded_document_ids: list[str] = []
         failed_documents: list[dict[str, str]] = []
         summary_by_id = {
@@ -344,7 +359,9 @@ class KnowledgeRagService:
         if not target_ids:
             return {}
         try:
-            statuses = self._get_runtime().get_document_statuses(target_ids)
+            statuses = self._run_runtime_operation(
+                lambda runtime: runtime.get_document_statuses(target_ids)
+            )
         except Exception as exc:
             logger.warning("Load LightRAG document statuses failed: %s", exc)
             return {}
@@ -358,16 +375,40 @@ class KnowledgeRagService:
         if not normalized_id:
             return
         try:
-            self._get_runtime().delete_document(normalized_id)
+            self._run_runtime_operation(
+                lambda runtime: runtime.delete_document(normalized_id)
+            )
         except Exception as exc:
             logger.warning("Delete LightRAG document ignored doc_id=%s: %s", normalized_id, exc)
 
-    def _get_runtime(self) -> _LightRagRuntime:
+    def _run_runtime_operation(self, operation: Callable[[_LightRagRuntime], Any]) -> Any:
         signature, runtime_kwargs = self._build_runtime_signature()
-        thread_id = threading.get_ident()
+        return _runtime_executor.submit(
+            self._execute_runtime_operation,
+            signature,
+            runtime_kwargs,
+            operation,
+        ).result()
+
+    def _execute_runtime_operation(
+        self,
+        signature: tuple[Any, ...],
+        runtime_kwargs: dict[str, Any],
+        operation: Callable[[_LightRagRuntime], Any],
+    ) -> Any:
+        return operation(self._get_runtime(signature=signature, runtime_kwargs=runtime_kwargs))
+
+    def _get_runtime(
+        self,
+        *,
+        signature: tuple[Any, ...] | None = None,
+        runtime_kwargs: dict[str, Any] | None = None,
+    ) -> _LightRagRuntime:
+        if signature is None or runtime_kwargs is None:
+            signature, runtime_kwargs = self._build_runtime_signature()
         with _runtime_lock:
-            runtime = _runtime_instances.get(thread_id)
-            if runtime is not None and _runtime_signatures.get(thread_id) == signature:
+            runtime = _runtime_instances.get(_RUNTIME_CACHE_KEY)
+            if runtime is not None and _runtime_signatures.get(_RUNTIME_CACHE_KEY) == signature:
                 return runtime
 
             if runtime is not None:
@@ -377,8 +418,8 @@ class KnowledgeRagService:
                     logger.warning("Finalize previous LightRAG runtime failed: %s", exc)
 
             runtime = _LightRagRuntime(**runtime_kwargs)
-            _runtime_instances[thread_id] = runtime
-            _runtime_signatures[thread_id] = signature
+            _runtime_instances[_RUNTIME_CACHE_KEY] = runtime
+            _runtime_signatures[_RUNTIME_CACHE_KEY] = signature
             return runtime
 
     def _build_runtime_signature(self) -> tuple[tuple[Any, ...], dict[str, Any]]:
@@ -633,6 +674,10 @@ class KnowledgeRagService:
 
 
 def shutdown_knowledge_rag_runtime() -> None:
+    _runtime_executor.submit(_shutdown_runtime_instances).result()
+
+
+def _shutdown_runtime_instances() -> None:
     with _runtime_lock:
         for runtime in list(_runtime_instances.values()):
             try:
diff --git a/server/src/app/services/knowledge_rag_runtime.py b/server/src/app/services/knowledge_rag_runtime.py
index dea2222..6eeb269 100644
--- a/server/src/app/services/knowledge_rag_runtime.py
+++ b/server/src/app/services/knowledge_rag_runtime.py
@@ -229,7 +229,13 @@ class _LightRagRuntime:
             raise KnowledgeRagError(str(getattr(result, "message", "") or "LightRAG 删除文档失败。"))
 
     def _probe_embedding_dimension(self, config: RuntimeModelConfig) -> int:
-        vectors = self._request_embeddings(config, ["dimension probe"])
+        try:
+            vectors = self._request_embeddings(config, ["dimension probe"])
+        except Exception as exc:
+            raise KnowledgeRagError(
+                "Embedding model probe failed "
+                f"(slot={config.slot}, provider={config.provider}, model={config.model}): {exc}"
+            ) from exc
         if not vectors or not isinstance(vectors[0], list):
             raise KnowledgeRagError("无法从 embedding 模型返回结果中解析向量维度。")
         dimension = len(vectors[0])
diff --git a/server/src/app/services/settings.py b/server/src/app/services/settings.py
index c5131a9..33b4d7d 100644
--- a/server/src/app/services/settings.py
+++ b/server/src/app/services/settings.py
@@ -335,7 +335,12 @@ class SettingsService:
         for model_row in model_rows.values():
             self.db.refresh(model_row)
 
-        return self._serialize(settings_row, secrets_row, model_rows)
+        return self._serialize(
+            settings_row,
+            secrets_row,
+            model_rows,
+            self._build_hermes_form_snapshot(),
+        )
 
     def load_saved_model_api_key(self, slot: str | None) -> str:
         if not slot or slot not in MODEL_SLOT_CONFIGS:
@@ -748,7 +753,11 @@ class SettingsService:
             hermesForm=hermes_form,
             llmForm={
                 "mainProvider": main_model.provider,
-                "backupProvider": backup_model.provider,
+                "mainModel": main_model.model_name,
+                "mainEndpoint": main_model.endpoint,
+                "mainApiKey": "",
+                "mainApiKeyConfigured": bool(main_model.api_key_encrypted),
+                "backupProvider": backup_model.provider,
                 "backupModel": backup_model.model_name,
                 "backupEndpoint": backup_model.endpoint,
                 "backupApiKey": "",
diff --git a/server/src/app/services/user_agent.py b/server/src/app/services/user_agent.py
index 295ba81..7630458 100644
--- a/server/src/app/services/user_agent.py
+++ b/server/src/app/services/user_agent.py
@@ -148,22 +148,7 @@ class UserAgentService(
                 requires_confirmation=payload.requires_confirmation,
             )
 
-        fast_knowledge_answer = self._build_fast_knowledge_answer(
-            payload,
-            citations=citations,
-        )
-        if fast_knowledge_answer:
-            return UserAgentResponse(
-                answer=fast_knowledge_answer,
-                citations=citations,
-                suggested_actions=suggested_actions,
-                query_payload=query_payload,
-                draft_payload=draft_payload,
-                review_payload=review_payload,
-                risk_flags=risk_flags,
-                requires_confirmation=payload.requires_confirmation,
-            )
-
+        # 知识库问答必须优先让模型基于召回证据组织答案，避免片段渲染抢答导致答非所问。
         fallback_answer = self._build_fallback_answer(
             payload,
             citations=citations,
diff --git a/server/src/app/services/user_agent_knowledge.py b/server/src/app/services/user_agent_knowledge.py
index f338688..71043c2 100644
--- a/server/src/app/services/user_agent_knowledge.py
+++ b/server/src/app/services/user_agent_knowledge.py
@@ -86,7 +86,6 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
         *,
         citations: list[UserAgentCitation],
     ) -> str | None:
-        return None
         if payload.ontology.scenario != "knowledge":
             return None
         if str(payload.tool_payload.get("result_type") or "").strip() != "knowledge_search":
@@ -130,7 +129,10 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
         primary_heading = self._format_knowledge_heading_label(
             str(primary_item.get("heading") or "").strip()
         )
-        primary_lines = self._collect_direct_knowledge_answer_lines(ordered_evidence_items)
+        primary_lines = self._collect_direct_knowledge_answer_lines(
+            ordered_evidence_items,
+            query_terms=query_terms,
+        )
 
         lines: list[str] = []
         if user_name:
@@ -139,20 +141,42 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
         if primary_heading:
             source_prefix = f"{source_prefix}（{primary_heading}）"
 
+        conclusion_lines: list[str] = []
+        evidence_lines: list[str] = []
         if str(primary_item.get("kind") or "") == "table":
-            lines.append(f"{source_prefix}，当前能直接确认的是：")
-            lines.append(self._extract_relevant_table_preview(str(primary_item.get("content") or ""), query_terms))
+            table_content = str(primary_item.get("content") or "")
+            if self._question_requests_broad_knowledge_table(question):
+                table_preview = table_content.strip()
+            else:
+                table_preview = self._extract_relevant_table_preview(
+                    table_content,
+                    query_terms,
+                    preferred_terms=self._build_knowledge_table_preferred_terms(payload),
+                )
+            table_summary = self._summarize_knowledge_table_preview(table_preview)
+            conclusion_lines.append(f"{source_prefix}，{table_summary}")
+            evidence_lines.append(table_preview)
         else:
             if not primary_lines:
-                lines.append(
+                summary = self._summarize_knowledge_evidence_content(primary_item, query_terms)
+                conclusion_lines.append(
                     f"{source_prefix}，当前能直接确认的是："
-                    f"{self._summarize_knowledge_evidence_content(primary_item, query_terms)}"
+                    f"{summary}"
                 )
             elif len(primary_lines) == 1:
-                lines.append(f"{source_prefix}，当前能直接确认的是：{primary_lines[0].strip()}")
+                conclusion_lines.append(f"{source_prefix}，当前能直接确认的是：{primary_lines[0].strip()}")
+                evidence_lines.extend(primary_lines)
             else:
-                lines.append(f"{source_prefix}，当前能直接确认的是：")
-                lines.extend(primary_lines)
+                subject = self._build_knowledge_answer_subject(question, primary_heading)
+                summary = self._summarize_knowledge_lines_conclusion(
+                    primary_lines,
+                    heading=subject,
+                )
+                if summary:
+                    conclusion_lines.append(f"{source_prefix}，{summary}")
+                else:
+                    conclusion_lines.append(f"{source_prefix}，当前能直接确认的是：")
+                evidence_lines.extend(primary_lines)
 
         notes: list[str] = []
         location_note = self._build_missing_location_grounding_note(question, evidence_items)
@@ -161,14 +185,64 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
         if self._question_requires_explicit_condition(question) and not self._answer_evidence_has_numeric_or_condition(evidence_items):
             notes.append("当前命中的证据更偏规则说明或流程约束，还没有直接给出可立即套用的数值或完整条件。")
 
+        self._append_markdown_section(lines, "结论", conclusion_lines)
+        self._append_markdown_section(lines, "依据", evidence_lines)
         if notes:
-            lines.append("")
-            lines.append("说明：")
-            lines.extend(f"- {note}" for note in notes)
+            self._append_markdown_section(lines, "说明", [f"- {note}" for note in notes])
 
         return "\n".join(line for line in lines if line is not None).strip()
 
 
+    @staticmethod
+    def _append_markdown_section(lines: list[str], title: str, body_lines: list[str]) -> None:
+        cleaned = [str(line or "").rstrip() for line in body_lines if str(line or "").strip()]
+        if not cleaned:
+            return
+        if lines and lines[-1] != "":
+            lines.append("")
+        lines.append(f"## {title}")
+        lines.append("")
+        lines.extend(cleaned)
+
+
+    @staticmethod
+    def _build_knowledge_answer_subject(question: str, heading: str = "") -> str:
+        clean_heading = str(heading or "").strip()
+        if clean_heading and not any(
+            marker in clean_heading
+            for marker in ("问答线索补充", "结构化表格补充", "重点章节摘录", "章节导航")
+        ):
+            return clean_heading
+
+        normalized = re.sub(r"\s+", "", str(question or "").strip())
+        normalized = re.sub(r"[？?。.!！]+$", "", normalized)
+        normalized = re.sub(r"(是什么|有哪些|是多少|如何|怎么|吗|呢)$", "", normalized)
+        return normalized.strip("：:，,。.")
+
+
+    @staticmethod
+    def _build_knowledge_table_preferred_terms(payload: UserAgentRequest) -> list[str]:
+        terms: list[str] = []
+        context = payload.context_json or {}
+        for key in ("grade", "position", "job_grade", "rank", "level"):
+            value = str(context.get(key) or "").strip()
+            if value and value not in terms:
+                terms.append(value)
+
+        grade_match = re.fullmatch(r"[Pp](\d{1,2})", str(context.get("grade") or "").strip())
+        if grade_match:
+            grade = int(grade_match.group(1))
+            for start in range(max(0, grade - 4), grade + 1):
+                for end in range(grade, min(12, grade + 4) + 1):
+                    if start >= end:
+                        continue
+                    for separator in ("～", "~", "-", "至"):
+                        term = f"P{start}{separator}P{end}"
+                        if term not in terms:
+                            terms.append(term)
+        return terms
+
+
     @staticmethod
     def _resolve_knowledge_question(payload: UserAgentRequest) -> str:
         return str(payload.context_json.get("user_input_text") or payload.message or "").strip()
@@ -484,6 +558,8 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
     def _collect_direct_knowledge_answer_lines(
         self,
         ordered_evidence_items: list[dict[str, Any]],
+        *,
+        query_terms: list[str] | None = None,
     ) -> list[str]:
         if not ordered_evidence_items:
             return []
@@ -509,8 +585,18 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
         lines: list[str] = []
         seen: set[str] = set()
         for item in related_items:
-            rendered = self._render_knowledge_evidence_text(item)
-            for line in rendered.splitlines():
+            item_kind = str(item.get("kind") or "").strip()
+            item_content = str(item.get("content") or "")
+            if item_kind == "paragraph" or self._has_inline_numbered_knowledge_items(item_content):
+                rendered = self._focus_knowledge_segment_content(
+                    item_content,
+                    query_terms or [],
+                )
+                rendered_lines = self._split_inline_numbered_knowledge_items(rendered)
+            else:
+                rendered = self._render_knowledge_evidence_text(item)
+                rendered_lines = rendered.splitlines()
+            for line in rendered_lines:
                 normalized = str(line or "").strip()
                 if not normalized or normalized in seen:
                     continue
@@ -573,13 +659,21 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
             or "相关制度"
         ).strip()
         user_name = str(payload.context_json.get("name") or "").strip()
-        prefix = f"{user_name}，您好。\n" if user_name else ""
+        answer_lines: list[str] = []
+        if user_name:
+            answer_lines.append(f"{user_name}，您好。")
         if not hits:
-            return (
-                f"{prefix}我已经从《{title}》中检索到与你这次问题相关的制度依据，"
-                "但本次答案生成环节暂时没有成功返回。请稍后重试一次；如果仍然失败，"
-                "建议先检查主对话模型的连通性。"
+            self._append_markdown_section(
+                answer_lines,
+                "结论",
+                [f"当前没有拿到可用于回答这个问题的《{title}》知识库命中。"],
             )
+            self._append_markdown_section(
+                answer_lines,
+                "说明",
+                ["- 我不会用相似主题或外部常识硬凑答案；请补充更具体的关键词后再试一次。"],
+            )
+            return "\n".join(answer_lines).strip()
 
         evidence_lines: list[str] = []
         for item in evidence_items[:3]:
@@ -614,19 +708,28 @@ class UserAgentKnowledgeMixin(UserAgentKnowledgeHelpersMixin):
                 evidence_lines.append(f"- **《{item_title}》**：{excerpt}")
 
         if not evidence_lines:
-            return (
-                f"{prefix}当前《{title}》里可用于回答的关键条款还不够明确。"
-                "请补充费用类型、适用地区、职级或具体业务场景，我再继续帮你缩小范围。"
+            self._append_markdown_section(
+                answer_lines,
+                "结论",
+                [f"当前《{title}》里可用于回答这个问题的关键条款还不够明确。"],
             )
+            self._append_markdown_section(
+                answer_lines,
+                "说明",
+                ["- 请补充费用类型、适用地区、职级或具体业务场景，我再继续帮你缩小范围。"],
+            )
+            return "\n".join(answer_lines).strip()
 
-        return "\n".join(
-            [
-                f"{prefix}我先根据当前制度依据给出可以确认的部分。",
-                "",
-                "**依据**：",
-                *evidence_lines,
-                "",
-                "**说明**：以上只使用当前命中的知识库证据；没有在证据中出现的适用条件或金额，我不会替你默认补齐。",
-            ]
-        ).strip()
+        self._append_markdown_section(
+            answer_lines,
+            "结论",
+            ["我先根据当前制度依据给出可以确认的部分。"],
+        )
+        self._append_markdown_section(answer_lines, "依据", evidence_lines)
+        self._append_markdown_section(
+            answer_lines,
+            "说明",
+            ["- 以上只使用当前命中的知识库证据；没有在证据中出现的适用条件或金额，我不会替你默认补齐。"],
+        )
+        return "\n".join(answer_lines).strip()
 
diff --git a/server/src/app/services/user_agent_response.py b/server/src/app/services/user_agent_response.py
index d25dd66..a315f73 100644
--- a/server/src/app/services/user_agent_response.py
+++ b/server/src/app/services/user_agent_response.py
@@ -280,10 +280,15 @@ class UserAgentResponseMixin:
         if payload.ontology.scenario == "knowledge":
             answer_style_instruction = (
                 "你是财务制度知识问答助手。只能依据 facts.tool_payload.hits、facts.knowledge_answer_evidence、citations 与 conversation_history 回答，"
-                "不要扩展成通用助手。优先直接回答，不要复述思考过程，不要输出 JSON、代码块或 <think>。"
+                "不要扩展成通用助手。优先直接回答，不要复述思考过程，不要输出 JSON、代码块或可见思考过程。"
+                "禁止使用“已命中”“答案整理阶段”“稍后重试”。"
+                "最终答复必须使用 Markdown，优先包含“## 结论”“## 依据”“## 说明”这三个二级标题；"
+                "如果某一部分没有内容，可以省略该标题。"
                 "回答风格要像一位真正熟悉制度的财务伙伴：先直接回应用户的核心问题，再用一张简洁表格或短段落说明依据，"
                 "最后补充最重要的注意事项。不要写成“已检索到内容”的系统回执，也不要把命中片段连缀成答案。"
                 "必须优先回答用户当前这句话本身，不能把制度标题、制度全文或完整标准表当成主答案。"
+                "回答前先判断召回内容是否真的能回答当前问题；如果不能，必须明确说当前知识库没有找到直接依据，"
+                "不要改答相邻主题，也不要用相似条款硬凑答案。"
                 "如果用户问的是某次具体行程“一共能报多少”，就先给“当前已能确认的金额”，再用一张很短的表说明项目、"
                 "适用标准、计算式和结果；如果总额还缺少住宿晚数、实际票据或其他必要条件，就明确写出“暂不能确认的部分”。"
                 "只有用户明确在问“标准有哪些”或“制度全文怎么规定”时，才展开完整标准表。"
@@ -488,7 +493,7 @@ class UserAgentResponseMixin:
         citations: list[UserAgentCitation],
     ) -> str:
         if str(payload.tool_payload.get("result_type") or "").strip() == "knowledge_search":
-            if citations:
+            if citations or list(payload.tool_payload.get("hits") or []):
                 return self._build_knowledge_search_answer(payload, citations)
 
             tool_message = str(payload.tool_payload.get("message") or "").strip()
diff --git a/server/storage/knowledge/.index.json b/server/storage/knowledge/.index.json
index 47aeb04..c75a27c 100644
--- a/server/storage/knowledge/.index.json
+++ b/server/storage/knowledge/.index.json
@@ -14,13 +14,13 @@
       "updated_at": "2026-05-17T09:28:28.999515+00:00",
       "uploaded_by": "admin",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:08.579777+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-17T10:01:33.272539+00:00",
       "ingest_completed_at": "2026-05-17T10:01:33.272539+00:00",
       "ingest_document_name": "远光《公司支出管理办法（2024）》.pdf",
       "ingest_document_updated_at": "2026-05-17T09:28:28.999515+00:00",
       "ingest_document_sha256": "67a74538bce0dec71ccbb947256cc2c9c0e672d148de49406b967ae1379dbece",
-      "ingest_agent_run_id": "run_7236fb72747742a3"
+      "ingest_agent_run_id": "run_b5984bade5324755"
     },
     {
       "id": "c7601043d9944ef2bcf4d3f67ed253f7",
@@ -35,8 +35,8 @@
       "updated_at": "2026-05-22T07:00:22.328877+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:09.863684+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T09:22:25.565409+00:00",
       "ingest_completed_at": "2026-05-22T09:22:25.565409+00:00",
       "ingest_document_name": "远光软件会计科目使用说明.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.328877+00:00",
@@ -56,8 +56,8 @@
       "updated_at": "2026-05-22T07:00:22.011016+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:50.652735+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-23T14:30:33.605531+00:00",
       "ingest_completed_at": "2026-05-23T14:30:33.605531+00:00",
       "ingest_document_name": "远光软件财务基础知识手册.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.011016+00:00",
@@ -77,8 +77,8 @@
       "updated_at": "2026-05-22T07:00:22.352133+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:51.908821+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T09:23:11.334499+00:00",
       "ingest_completed_at": "2026-05-22T09:23:11.334499+00:00",
       "ingest_document_name": "远光软件财务术语解释手册.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.352133+00:00",
@@ -98,8 +98,8 @@
       "updated_at": "2026-05-22T07:00:22.304623+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:13.581834+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T09:24:18.933073+00:00",
       "ingest_completed_at": "2026-05-22T09:24:18.933073+00:00",
       "ingest_document_name": "远光软件高新技术企业税收优惠政策汇总.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:22.304623+00:00",
@@ -119,8 +119,8 @@
       "updated_at": "2026-05-22T07:00:18.153373+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:53.906324+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:01:43.168774+00:00",
       "ingest_completed_at": "2026-05-22T16:01:43.168774+00:00",
       "ingest_document_name": "远光软件公司内部控制基本规范.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:18.153373+00:00",
@@ -140,8 +140,8 @@
       "updated_at": "2026-05-22T07:00:18.190399+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:55.339114+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:03:00.735908+00:00",
       "ingest_completed_at": "2026-05-22T16:03:00.735908+00:00",
       "ingest_document_name": "远光软件公司合同管理制度.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.190399+00:00",
@@ -161,8 +161,8 @@
       "updated_at": "2026-05-22T07:00:17.798679+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:56.741808+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:03:46.921675+00:00",
       "ingest_completed_at": "2026-05-22T16:03:46.921675+00:00",
       "ingest_document_name": "远光软件公司财务管理制度总则.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:17.798679+00:00",
@@ -182,8 +182,8 @@
       "updated_at": "2026-05-22T07:00:18.531598+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:19.014702+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:04:58.719410+00:00",
       "ingest_completed_at": "2026-05-22T16:04:58.719410+00:00",
       "ingest_document_name": "远光软件公司资产管理制度.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:18.531598+00:00",
@@ -203,8 +203,8 @@
       "updated_at": "2026-05-22T07:00:18.221073+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:00:59.485821+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:06:08.172318+00:00",
       "ingest_completed_at": "2026-05-22T16:06:08.172318+00:00",
       "ingest_document_name": "远光软件公司采购管理办法.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.221073+00:00",
@@ -224,8 +224,8 @@
       "updated_at": "2026-05-22T07:00:19.734422+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:00.774887+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:06:48.466110+00:00",
       "ingest_completed_at": "2026-05-22T16:06:48.466110+00:00",
       "ingest_document_name": "远光软件公司差旅费管理办法.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:19.734422+00:00",
@@ -245,8 +245,8 @@
       "updated_at": "2026-05-22T07:00:20.095824+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:02.037101+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:07:23.262328+00:00",
       "ingest_completed_at": "2026-05-22T16:07:23.262328+00:00",
       "ingest_document_name": "远光软件出差审批流程说明.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:20.095824+00:00",
@@ -266,8 +266,8 @@
       "updated_at": "2026-05-22T07:00:20.128471+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:24.076574+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:08:02.190081+00:00",
       "ingest_completed_at": "2026-05-22T16:08:02.190081+00:00",
       "ingest_document_name": "远光软件国际出差管理规定.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:20.128471+00:00",
@@ -287,8 +287,8 @@
       "updated_at": "2026-05-22T07:00:19.759954+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:25.270086+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:09:23.091744+00:00",
       "ingest_completed_at": "2026-05-22T16:09:23.091744+00:00",
       "ingest_document_name": "远光软件差旅费标准速查表.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:19.759954+00:00",
@@ -308,8 +308,8 @@
       "updated_at": "2026-05-22T07:00:18.922298+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:26.510710+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:11:04.764727+00:00",
       "ingest_completed_at": "2026-05-22T16:11:04.764727+00:00",
       "ingest_document_name": "远光软件公司发票审核标准.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.922298+00:00",
@@ -329,8 +329,8 @@
       "updated_at": "2026-05-22T07:00:18.560177+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:06.719118+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:11:54.017817+00:00",
       "ingest_completed_at": "2026-05-22T16:11:54.017817+00:00",
       "ingest_document_name": "远光软件公司发票管理规范.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.560177+00:00",
@@ -350,8 +350,8 @@
       "updated_at": "2026-05-22T07:00:18.888128+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:28.865726+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:12:23.821434+00:00",
       "ingest_completed_at": "2026-05-22T16:12:23.821434+00:00",
       "ingest_document_name": "远光软件公司增值税发票操作指南.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:18.888128+00:00",
@@ -371,8 +371,8 @@
       "updated_at": "2026-05-22T07:00:18.953110+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:30.095619+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:13:15.450300+00:00",
       "ingest_completed_at": "2026-05-22T16:13:15.450300+00:00",
       "ingest_document_name": "远光软件公司电子发票管理办法.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.953110+00:00",
@@ -392,8 +392,8 @@
       "updated_at": "2026-05-22T07:00:21.585718+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:09.790447+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:13:44.636629+00:00",
       "ingest_completed_at": "2026-05-22T16:13:44.636629+00:00",
       "ingest_document_name": "远光软件企业所得税汇算清缴操作手册.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:21.585718+00:00",
@@ -413,8 +413,8 @@
       "updated_at": "2026-05-22T07:00:20.881351+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:11.027818+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:14:50.092490+00:00",
       "ingest_completed_at": "2026-05-22T16:14:50.092490+00:00",
       "ingest_document_name": "远光软件公司税务管理制度.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:20.881351+00:00",
@@ -434,8 +434,8 @@
       "updated_at": "2026-05-22T07:00:21.606227+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:33.826025+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:15:56.676286+00:00",
       "ingest_completed_at": "2026-05-22T16:15:56.676286+00:00",
       "ingest_document_name": "远光软件研发费用加计扣除管理办法.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:21.606227+00:00",
@@ -455,8 +455,8 @@
       "updated_at": "2026-05-22T07:00:21.202633+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:13.991763+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:16:06.540773+00:00",
       "ingest_completed_at": "2026-05-22T16:16:06.540773+00:00",
       "ingest_document_name": "远光软件软件产品增值税即征即退操作指南.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:21.202633+00:00",
@@ -476,8 +476,8 @@
       "updated_at": "2026-05-22T07:00:22.379307+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:15.257700+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:23:24.252614+00:00",
       "ingest_completed_at": "2026-05-22T16:23:24.252614+00:00",
       "ingest_document_name": "远光软件公司预算管理制度.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.379307+00:00",
@@ -497,8 +497,8 @@
       "updated_at": "2026-05-22T07:00:22.760169+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:16.510610+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:23:29.997956+00:00",
       "ingest_completed_at": "2026-05-22T16:23:29.997956+00:00",
       "ingest_document_name": "远光软件年度预算编制指南.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:22.760169+00:00",
@@ -518,8 +518,8 @@
       "updated_at": "2026-05-22T07:00:22.848272+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:38.728430+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:24:37.382612+00:00",
       "ingest_completed_at": "2026-05-22T16:24:37.382612+00:00",
       "ingest_document_name": "远光软件预算执行分析报告模板.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.848272+00:00",
@@ -539,8 +539,8 @@
       "updated_at": "2026-05-22T07:00:22.803708+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:19.050297+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:24:45.161319+00:00",
       "ingest_completed_at": "2026-05-22T16:24:45.161319+00:00",
       "ingest_document_name": "远光软件预算编制模板.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:22.803708+00:00",
@@ -560,8 +560,8 @@
       "updated_at": "2026-05-22T07:00:21.971983+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:20.323058+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:25:33.968414+00:00",
       "ingest_completed_at": "2026-05-22T16:25:33.968414+00:00",
       "ingest_document_name": "远光软件财务共享服务SLA标准.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:21.971983+00:00",
@@ -581,8 +581,8 @@
       "updated_at": "2026-05-22T07:00:21.634300+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:21.585474+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:26:05.301987+00:00",
       "ingest_completed_at": "2026-05-22T16:26:05.301987+00:00",
       "ingest_document_name": "远光软件财务共享服务中心运营管理办法.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:21.634300+00:00",
@@ -602,8 +602,8 @@
       "updated_at": "2026-05-22T07:00:21.945868+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:43.752235+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:26:54.048075+00:00",
       "ingest_completed_at": "2026-05-22T16:26:54.048075+00:00",
       "ingest_document_name": "远光软件财务共享服务操作手册.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:21.945868+00:00",
@@ -623,8 +623,8 @@
       "updated_at": "2026-05-22T07:00:19.662743+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:24.093834+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:27:31.775974+00:00",
       "ingest_completed_at": "2026-05-22T16:27:31.775974+00:00",
       "ingest_document_name": "远光软件报销流程培训手册.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:19.662743+00:00",
@@ -644,8 +644,8 @@
       "updated_at": "2026-05-22T07:00:19.323921+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:25.246857+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:27:44.244066+00:00",
       "ingest_completed_at": "2026-05-22T16:27:44.244066+00:00",
       "ingest_document_name": "远光软件新员工财务培训课件.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:19.323921+00:00",
@@ -665,8 +665,8 @@
       "updated_at": "2026-05-22T07:00:18.988700+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:26.471932+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:28:24.573683+00:00",
       "ingest_completed_at": "2026-05-22T16:28:24.573683+00:00",
       "ingest_document_name": "远光软件财务制度培训手册.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:18.988700+00:00",
@@ -686,8 +686,8 @@
       "updated_at": "2026-05-22T07:00:19.686485+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:48.525207+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:29:03.349502+00:00",
       "ingest_completed_at": "2026-05-22T16:29:03.349502+00:00",
       "ingest_document_name": "远光软件财务培训课程安排.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:19.686485+00:00",
@@ -707,8 +707,8 @@
       "updated_at": "2026-05-22T07:00:20.476077+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:49.746825+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:29:29.050791+00:00",
       "ingest_completed_at": "2026-05-22T16:29:29.050791+00:00",
       "ingest_document_name": "远光软件报销问题处理指引.xlsx",
       "ingest_document_updated_at": "2026-05-22T07:00:20.476077+00:00",
@@ -728,8 +728,8 @@
       "updated_at": "2026-05-22T07:00:20.453567+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:30.343781+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:35:03.548506+00:00",
       "ingest_completed_at": "2026-05-22T16:35:03.548506+00:00",
       "ingest_document_name": "远光软件财务制度问答汇总.pdf",
       "ingest_document_updated_at": "2026-05-22T07:00:20.453567+00:00",
@@ -749,8 +749,8 @@
       "updated_at": "2026-05-22T07:00:20.158497+00:00",
       "uploaded_by": "系统导入",
       "version_number": 1,
-      "ingest_status": 4,
-      "ingest_status_updated_at": "2026-05-26T16:01:31.573128+00:00",
+      "ingest_status": 3,
+      "ingest_status_updated_at": "2026-05-22T16:35:27.056080+00:00",
       "ingest_completed_at": "2026-05-22T16:35:27.056080+00:00",
       "ingest_document_name": "远光软件财务报销常见问题解答.docx",
       "ingest_document_updated_at": "2026-05-22T07:00:20.158497+00:00",
diff --git a/server/tests/test_agent_runs_service.py b/server/tests/test_agent_runs_service.py
index 3cc8fdd..fd4e542 100644
--- a/server/tests/test_agent_runs_service.py
+++ b/server/tests/test_agent_runs_service.py
@@ -49,6 +49,27 @@ def test_agent_run_service_marks_stale_knowledge_sync_run_failed_on_read() -> No
         assert all(item.run_id != created.run_id for item in running_runs)
 
 
+def test_agent_run_service_marks_stale_llm_wiki_run_failed_on_read() -> None:
+    with build_session() as db:
+        service = AgentRunService(db)
+        created = service.create_run(
+            agent=AgentName.HERMES.value,
+            source=AgentRunSource.SCHEDULE.value,
+            status=AgentRunStatus.RUNNING.value,
+            route_json={
+                "job_type": "llm_wiki_sync",
+                "heartbeat_at": (datetime.now(UTC) - timedelta(minutes=31)).isoformat(),
+                "requested_document_ids": [],
+            },
+        )
+
+        fetched = service.get_run(created.run_id)
+
+        assert fetched is not None
+        assert fetched.status == AgentRunStatus.FAILED.value
+        assert fetched.error_message == "Knowledge index heartbeat timed out."
+
+
 def test_agent_run_service_updates_existing_tool_call() -> None:
     with build_session() as db:
         service = AgentRunService(db)
diff --git a/server/tests/test_applicant_expense_profile_algorithm.py b/server/tests/test_applicant_expense_profile_algorithm.py
new file mode 100644
index 0000000..fd6d8c4
--- /dev/null
+++ b/server/tests/test_applicant_expense_profile_algorithm.py
@@ -0,0 +1,74 @@
+from decimal import Decimal
+
+from app.algorithem import ApplicantExpenseProfileInput, evaluate_applicant_expense_profile
+
+
+def test_applicant_profile_recommends_review_and_days_cap() -> None:
+    result = evaluate_applicant_expense_profile(
+        ApplicantExpenseProfileInput(
+            applicant_claim_count_90d=6,
+            peer_claim_count_p75_90d=4,
+            applicant_amount_90d=Decimal("42800"),
+            available_peer_budget_90d=Decimal("150000"),
+            amount_percentile=Decimal("88"),
+            peer_amount_median_90d=Decimal("25000"),
+            adjusted_or_returned_count_180d=3,
+            approved_claim_count_180d=12,
+            requested_days=Decimal("5"),
+            peer_travel_days_p75=Decimal("3"),
+            business_buffer_days=Decimal("1"),
+            claim_amount=Decimal("8000"),
+            peer_daily_cost_baseline=Decimal("1400"),
+            tolerance_factor=Decimal("1.20"),
+        )
+    )
+
+    assert result.profile_score == 68
+    assert result.profile_level == "review"
+    assert result.recommendation_level == "review"
+    assert result.travel_days_ratio == Decimal("1.6667")
+    assert result.suggested_days == Decimal("4.0000")
+    assert result.suggested_amount_cap == Decimal("6720.00")
+    assert "applicant.amount_percentile.p85" in result.basis_codes
+    assert "travel.days_ratio.review" in result.basis_codes
+
+
+def test_applicant_profile_handles_missing_baselines_without_false_risk() -> None:
+    result = evaluate_applicant_expense_profile(
+        ApplicantExpenseProfileInput(
+            applicant_claim_count_90d=1,
+            applicant_amount_90d=Decimal("800"),
+            requested_days=Decimal("2"),
+            claim_amount=Decimal("800"),
+        )
+    )
+
+    assert result.profile_score == 0
+    assert result.profile_level == "normal"
+    assert result.recommendation_level == "normal"
+    assert result.frequency_ratio == Decimal("0")
+    assert result.daily_cost_ratio == Decimal("0")
+    assert result.suggested_days == Decimal("2.0000")
+    assert result.suggested_amount_cap is None
+    assert result.basis_codes == []
+
+
+def test_entertainment_deviation_can_escalate_recommendation() -> None:
+    result = evaluate_applicant_expense_profile(
+        ApplicantExpenseProfileInput(
+            entertainment_amount=Decimal("3000"),
+            attendee_count=3,
+            entertainment_standard_cap=Decimal("600"),
+            same_customer_frequency_90d=3,
+            applicant_entertainment_percentile=Decimal("96"),
+        )
+    )
+
+    assert result.entertainment_deviation_score == 100
+    assert result.current_claim_deviation_score == 100
+    assert result.profile_score == 15
+    assert result.profile_level == "normal"
+    assert result.recommendation_level == "escalation"
+    assert "entertainment.per_capita.escalation" in result.basis_codes
+    assert "entertainment.same_customer.frequency_review" in result.basis_codes
+    assert "entertainment.amount_percentile.p95" in result.basis_codes
diff --git a/server/tests/test_budget_endpoints.py b/server/tests/test_budget_endpoints.py
index 46e65dd..dd6a6fe 100644
--- a/server/tests/test_budget_endpoints.py
+++ b/server/tests/test_budget_endpoints.py
@@ -13,6 +13,10 @@ from app.api.deps import get_db
 from app.db.base import Base
 from app.main import create_app
 from app.models.budget import BudgetAllocation, BudgetTransaction
+from app.models.employee import Employee
+from app.models.financial_record import ExpenseClaim
+from app.models.organization import OrganizationUnit
+from app.models.role import Role
 
 
 def build_session_factory() -> sessionmaker[Session]:
@@ -109,6 +113,27 @@ def seed_budget_allocations(db: Session) -> None:
     db.commit()
 
 
+def seed_market_budget_monitor(db: Session) -> tuple[Role, OrganizationUnit]:
+    role = Role(role_code="budget_monitor", name="预算监控员")
+    department = OrganizationUnit(
+        id="dept-market",
+        unit_code="MARKET-DEPT",
+        name="市场部",
+        unit_type="department",
+    )
+    employee = Employee(
+        employee_no="E-BUDGET-MARKET-P8",
+        name="赵预算",
+        email="budget-monitor@example.com",
+        grade="P8",
+        organization_unit=department,
+        roles=[role],
+    )
+    db.add_all([role, department, employee])
+    db.flush()
+    return role, department
+
+
 def test_admin_can_view_all_budget_allocations_without_is_admin_header() -> None:
     client, session_factory = build_client()
     with session_factory() as db:
@@ -281,3 +306,69 @@ def test_budget_monitor_cannot_edit_and_admin_can_edit() -> None:
     )
     assert admin_response.status_code == 201
     assert admin_response.json()["department_name"] == "销售部"
+
+
+def test_budget_analysis_endpoint_is_limited_to_budget_roles() -> None:
+    client, session_factory = build_client()
+    with session_factory() as db:
+        seed_budget_allocations(db)
+        budget_role, market_department = seed_market_budget_monitor(db)
+        p6_budget_monitor = Employee(
+            employee_no="E-BUDGET-MARKET-P6",
+            name="低级预算",
+            email="p6-budget-monitor@example.com",
+            grade="P6",
+            organization_unit=market_department,
+            roles=[budget_role],
+        )
+        db.add(p6_budget_monitor)
+        db.flush()
+        claim = ExpenseClaim(
+            claim_no="APP-BUDGET-ANALYSIS-001",
+            employee_id=p6_budget_monitor.id,
+            employee_name="低级预算",
+            department_id="dept-market",
+            department_name="市场部",
+            project_code=None,
+            expense_type="travel_application",
+            reason="客户现场交付预算申请",
+            location="上海",
+            amount=Decimal("12000.00"),
+            currency="CNY",
+            invoice_count=0,
+            occurred_at=datetime(2026, 5, 12, 9, 0, tzinfo=UTC),
+            submitted_at=datetime(2026, 5, 12, 10, 0, tzinfo=UTC),
+            status="submitted",
+            approval_stage="预算管理者审批",
+            risk_flags_json=[],
+        )
+        db.add(claim)
+        db.commit()
+        claim_id = claim.id
+
+    ordinary_response = client.get(
+        f"/api/v1/reimbursements/claims/{claim_id}/budget-analysis",
+        headers={
+            "x-auth-username": "zhangsan@example.com",
+            "x-auth-role-codes": "employee",
+        },
+    )
+    monitor_response = client.get(
+        f"/api/v1/reimbursements/claims/{claim_id}/budget-analysis",
+        headers={
+            "x-auth-username": "budget-monitor@example.com",
+            "x-auth-role-codes": "budget_monitor",
+        },
+    )
+    p6_monitor_response = client.get(
+        f"/api/v1/reimbursements/claims/{claim_id}/budget-analysis",
+        headers={
+            "x-auth-username": "p6-budget-monitor@example.com",
+            "x-auth-role-codes": "budget_monitor",
+        },
+    )
+
+    assert ordinary_response.status_code == 403
+    assert p6_monitor_response.status_code == 403
+    assert monitor_response.status_code == 200
+    assert Decimal(monitor_response.json()["metrics"]["claim_amount_ratio"]) == Decimal("24.00")
diff --git a/server/tests/test_expense_claim_service.py b/server/tests/test_expense_claim_service.py
index 87b9526..2515ef2 100644
--- a/server/tests/test_expense_claim_service.py
+++ b/server/tests/test_expense_claim_service.py
@@ -16,10 +16,12 @@ from app.models.budget import BudgetAllocation, BudgetReservation, BudgetTransac
 from app.models.employee import Employee
 from app.models.financial_record import ExpenseClaim, ExpenseClaimItem
 from app.models.organization import OrganizationUnit
+from app.models.role import Role
 from app.schemas.ontology import OntologyParseRequest
 from app.schemas.ocr import OcrRecognizeBatchRead, OcrRecognizeDocumentRead
 from app.schemas.reimbursement import ExpenseClaimItemCreate, ExpenseClaimItemUpdate, ExpenseClaimUpdate
 from app.services.agent_conversations import AgentConversationService
+from app.services.budget import BudgetService
 from app.services.expense_claim_attachment_storage import ExpenseClaimAttachmentStorage
 from app.services.expense_claims import ExpenseClaimService
 from app.services.ontology import SemanticOntologyService
@@ -108,6 +110,16 @@ def _seed_budget_allocation(
     return allocation
 
 
+def _seed_budget_monitor_role(db: Session) -> Role:
+    role = db.query(Role).filter(Role.role_code == "budget_monitor").one_or_none()
+    if role is not None:
+        return role
+    role = Role(role_code="budget_monitor", name="预算监控员")
+    db.add(role)
+    db.flush()
+    return role
+
+
 def test_validate_claim_for_submission_allows_office_claim_without_location() -> None:
     service = ExpenseClaimService.__new__(ExpenseClaimService)
     claim = build_claim(expense_type="office", location="待补充")
@@ -3320,32 +3332,55 @@ def test_application_submit_skips_budget_for_non_demo_subject() -> None:
         )
 
 
-def test_direct_manager_can_approve_application_claim_to_reimbursement_draft() -> None:
-    current_user = CurrentUserContext(
+def test_direct_manager_can_route_application_claim_to_budget_approval_then_budget_manager_creates_draft() -> None:
+    manager_user = CurrentUserContext(
         username="manager-application-approve@example.com",
         name="李经理",
         role_codes=["manager"],
         is_admin=False,
     )
+    budget_user = CurrentUserContext(
+        username="budget-p8-application-approve@example.com",
+        name="赵预算",
+        role_codes=["budget_monitor"],
+        is_admin=False,
+    )
 
     with build_session() as db:
+        budget_role = _seed_budget_monitor_role(db)
+        department = OrganizationUnit(
+            unit_code="DELIVERY-BUDGET-APPROVE",
+            name="交付部",
+            unit_type="department",
+        )
         manager = Employee(
             employee_no="E8112",
             name="李经理",
             email="manager-application-approve@example.com",
+            organization_unit=department,
+        )
+        budget_manager = Employee(
+            employee_no="E8112-BUDGET",
+            name="赵预算",
+            email="budget-p8-application-approve@example.com",
+            grade="P8",
+            organization_unit=department,
+            roles=[budget_role],
         )
         employee = Employee(
             employee_no="E8113",
             name="张三",
             email="zhangsan-application-approve@example.com",
             manager=manager,
+            organization_unit=department,
         )
-        db.add_all([manager, employee])
+        db.add_all([department, manager, budget_manager, employee])
         db.flush()
         claim = ExpenseClaim(
             claim_no="APP-20260525-APPROVE",
             employee_id=employee.id,
             employee_name="张三",
+            department_id=department.id,
             department_name="交付部",
             project_code="PRJ-A",
             expense_type="travel_application",
@@ -3364,10 +3399,33 @@ def test_direct_manager_can_approve_application_claim_to_reimbursement_draft() -
         db.commit()
         claim_id = claim.id
 
+        leader_approved = ExpenseClaimService(db).approve_claim(
+            claim_id,
+            manager_user,
+            opinion="业务必要，同意申请。",
+        )
+
+        assert leader_approved is not None
+        assert leader_approved.status == "submitted"
+        assert leader_approved.approval_stage == "预算管理者审批"
+        assert db.query(ExpenseClaim).filter(ExpenseClaim.claim_no.like("RE-%")).count() == 0
+        assert any(
+            isinstance(flag, dict)
+            and flag.get("source") == "manual_approval"
+            and flag.get("event_type") == "expense_application_approval"
+            and flag.get("opinion") == "业务必要，同意申请。"
+            and flag.get("previous_approval_stage") == "直属领导审批"
+            and flag.get("next_status") == "submitted"
+            and flag.get("next_approval_stage") == "预算管理者审批"
+            and flag.get("next_approver_name") == "赵预算"
+            and flag.get("next_approver_grade") == "P8"
+            for flag in leader_approved.risk_flags_json
+        )
+
         approved = ExpenseClaimService(db).approve_claim(
             claim_id,
-            current_user,
-            opinion="业务必要，同意申请。",
+            budget_user,
+            opinion="预算额度可承接，同意。",
         )
 
         assert approved is not None
@@ -3400,14 +3458,15 @@ def test_direct_manager_can_approve_application_claim_to_reimbursement_draft() -
             and flag.get("event_type") == "expense_application_to_reimbursement_draft"
             and flag.get("application_claim_no") == "APP-20260525-APPROVE"
             and flag.get("leader_opinion") == "业务必要，同意申请。"
+            and flag.get("budget_opinion") == "预算额度可承接，同意。"
             for flag in generated_draft.risk_flags_json
         )
         assert any(
             isinstance(flag, dict)
-            and flag.get("source") == "manual_approval"
-            and flag.get("event_type") == "expense_application_approval"
-            and flag.get("opinion") == "业务必要，同意申请。"
-            and flag.get("previous_approval_stage") == "直属领导审批"
+            and flag.get("source") == "budget_approval"
+            and flag.get("event_type") == "expense_application_budget_approval"
+            and flag.get("opinion") == "预算额度可承接，同意。"
+            and flag.get("previous_approval_stage") == "预算管理者审批"
             and flag.get("next_status") == "approved"
             and flag.get("next_approval_stage") == "审批完成"
             and flag.get("generated_draft_claim_id") == generated_draft.id
@@ -3473,7 +3532,7 @@ def test_direct_manager_return_application_claim_records_return_node_and_opinion
             claim.id,
             manager_user,
             reason="预算说明不够清楚，请补充项目必要性。",
-            reason_codes=["application_business_need_unclear", "application_budget_basis_missing"],
+            reason_codes=["application_budget_basis_missing"],
         )
 
         assert returned is not None
@@ -3493,11 +3552,8 @@ def test_direct_manager_return_application_claim_records_return_node_and_opinion
         assert return_event["leader_opinion"] == "预算说明不够清楚，请补充项目必要性。"
         assert return_event["return_stage"] == "直属领导审批"
         assert return_event["return_stage_key"] == "direct_manager"
-        assert return_event["reason_codes"] == [
-            "application_business_need_unclear",
-            "application_budget_basis_missing",
-        ]
-        assert return_event["risk_points"] == ["业务必要性说明不足", "预算测算依据不足"]
+        assert return_event["reason_codes"] == ["application_budget_basis_missing"]
+        assert return_event["risk_points"] == ["预算测算依据不足"]
         assert return_event["next_status"] == "returned"
         assert return_event["next_approval_stage"] == "待提交"
 
@@ -3515,20 +3571,43 @@ def test_application_approval_transfers_budget_reservation_to_reimbursement_draf
         role_codes=["manager"],
         is_admin=False,
     )
+    budget_user = CurrentUserContext(
+        username="budget-p8-transfer@example.com",
+        name="赵预算",
+        role_codes=["budget_monitor"],
+        is_admin=False,
+    )
 
     with build_session() as db:
+        budget_role = _seed_budget_monitor_role(db)
+        department = OrganizationUnit(
+            id="dept-budget-transfer",
+            unit_code="DELIVERY-BUDGET-TRANSFER",
+            name="交付部",
+            unit_type="department",
+        )
         manager = Employee(
             employee_no="M-BUDGET-APP",
             name="李经理",
             email="manager-application-budget@example.com",
+            organization_unit=department,
+        )
+        budget_manager = Employee(
+            employee_no="P8-BUDGET-APP",
+            name="赵预算",
+            email="budget-p8-transfer@example.com",
+            grade="P8",
+            organization_unit=department,
+            roles=[budget_role],
         )
         employee = Employee(
             employee_no="E-BUDGET-APP",
             name="张三",
             email="application-budget-owner-approve@example.com",
             manager=manager,
+            organization_unit=department,
         )
-        db.add_all([manager, employee])
+        db.add_all([department, manager, budget_manager, employee])
         db.flush()
         _seed_budget_allocation(
             db,
@@ -3560,10 +3639,19 @@ def test_application_approval_transfers_budget_reservation_to_reimbursement_draf
         service = ExpenseClaimService(db)
 
         service.submit_claim(claim.id, owner)
-        approved = service.approve_claim(claim.id, manager_user, opinion="同意申请")
-        generated_draft = db.query(ExpenseClaim).filter(ExpenseClaim.claim_no.like("RE-%")).one()
+        leader_approved = service.approve_claim(claim.id, manager_user, opinion="同意申请")
         reservation = db.query(BudgetReservation).one()
 
+        assert leader_approved is not None
+        assert leader_approved.approval_stage == "预算管理者审批"
+        assert reservation.source_type == "application"
+        assert reservation.source_id == claim.id
+        assert db.query(ExpenseClaim).filter(ExpenseClaim.claim_no.like("RE-%")).count() == 0
+
+        approved = service.approve_claim(claim.id, budget_user, opinion="预算通过")
+        generated_draft = db.query(ExpenseClaim).filter(ExpenseClaim.claim_no.like("RE-%")).one()
+        db.refresh(reservation)
+
         assert approved is not None
         assert reservation.source_type == "claim"
         assert reservation.source_id == generated_draft.id
@@ -3576,7 +3664,7 @@ def test_application_approval_transfers_budget_reservation_to_reimbursement_draf
         )
 
 
-def test_direct_manager_approval_requires_leader_opinion() -> None:
+def test_direct_manager_approval_defaults_blank_opinion_to_agree() -> None:
     current_user = CurrentUserContext(
         username="manager-application-required-opinion@example.com",
         name="李经理",
@@ -3620,19 +3708,79 @@ def test_direct_manager_approval_requires_leader_opinion() -> None:
         db.commit()
         claim_id = claim.id
 
-        with pytest.raises(ValueError, match="领导审核意见不能为空"):
-            ExpenseClaimService(db).approve_claim(
-                claim_id,
-                current_user,
-                opinion="   ",
-            )
+        approved = ExpenseClaimService(db).approve_claim(
+            claim_id,
+            current_user,
+            opinion="   ",
+        )
 
-        db.refresh(claim)
-        assert claim.status == "submitted"
-        assert claim.approval_stage == "直属领导审批"
+        assert approved is not None
+        assert approved.status == "submitted"
+        assert approved.approval_stage == "预算管理者审批"
+        assert any(
+            isinstance(flag, dict)
+            and flag.get("event_type") == "expense_application_approval"
+            and flag.get("opinion") == "同意"
+            and flag.get("next_approval_stage") == "预算管理者审批"
+            for flag in approved.risk_flags_json
+        )
         assert db.query(ExpenseClaim).filter(ExpenseClaim.claim_no.like("RE-%")).count() == 0
 
 
+def test_budget_analysis_uses_current_application_reservation_without_double_counting() -> None:
+    owner = CurrentUserContext(
+        username="application-budget-analysis-owner@example.com",
+        name="张三",
+        role_codes=["employee"],
+        is_admin=False,
+    )
+
+    with build_session() as db:
+        employee = Employee(
+            employee_no="E-BUDGET-ANALYSIS",
+            name="张三",
+            email="application-budget-analysis-owner@example.com",
+        )
+        db.add(employee)
+        db.flush()
+        _seed_budget_allocation(
+            db,
+            department_id="dept-budget-analysis",
+            department_name="交付部",
+            amount=Decimal("50000.00"),
+        )
+        claim = ExpenseClaim(
+            claim_no="APP-20260525-ANALYSIS",
+            employee_id=employee.id,
+            employee_name="张三",
+            department_id="dept-budget-analysis",
+            department_name="交付部",
+            project_code="PRJ-A",
+            expense_type="travel_application",
+            reason="客户现场交付",
+            location="上海",
+            amount=Decimal("12000.00"),
+            currency="CNY",
+            invoice_count=0,
+            occurred_at=datetime(2026, 5, 25, 9, 0, tzinfo=UTC),
+            submitted_at=None,
+            status="draft",
+            approval_stage="待提交",
+            risk_flags_json=[],
+        )
+        db.add(claim)
+        db.commit()
+
+        ExpenseClaimService(db).submit_claim(claim.id, owner)
+        analysis = BudgetService(db).analyze_claim_budget(claim)
+
+        assert analysis["metrics"]["claim_amount_ratio"] == "24.00"
+        assert analysis["metrics"]["after_usage_rate"] == "24.00"
+        assert analysis["budget_context"]["current_reserved_amount"] == "12000.00"
+        assert analysis["score"] >= 70
+        assert any("本次申请金额 12000.00 元，占预算 24.00%" in item for item in analysis["basis"])
+
+
 def test_finance_approve_reimbursement_consumes_budget_reservation() -> None:
     current_user = CurrentUserContext(
         username="finance-budget-approve@example.com",
@@ -4207,3 +4355,121 @@ def test_list_approval_claims_limits_finance_to_finance_stage_claims() -> None:
         claims = ExpenseClaimService(db).list_approval_claims(current_user)
 
         assert [claim.claim_no for claim in claims] == ["EXP-FIN-LIST-202"]
+
+
+def test_list_approval_claims_allows_budget_monitor_to_view_budget_stage_applications() -> None:
+    current_user = CurrentUserContext(
+        username="budget-p8-list@example.com",
+        name="赵预算",
+        role_codes=["budget_monitor"],
+        is_admin=False,
+    )
+    p8_without_budget_role = CurrentUserContext(
+        username="budget-p8-list@example.com",
+        name="budget manager",
+        role_codes=["manager"],
+        is_admin=False,
+    )
+
+    with build_session() as db:
+        budget_role = _seed_budget_monitor_role(db)
+        delivery_department = OrganizationUnit(
+            unit_code="DELIVERY-BUDGET-LIST",
+            name="交付部",
+            unit_type="department",
+        )
+        market_department = OrganizationUnit(
+            unit_code="MARKET-BUDGET-LIST",
+            name="市场部",
+            unit_type="department",
+        )
+        budget_manager = Employee(
+            employee_no="E-P8-BUDGET-LIST",
+            name="赵预算",
+            email="budget-p8-list@example.com",
+            grade="P8",
+            organization_unit=delivery_department,
+            roles=[budget_role],
+        )
+        employee = Employee(
+            employee_no="E-BUDGET-LIST-OWNER",
+            name="张三",
+            email="budget-list-owner@example.com",
+            organization_unit=delivery_department,
+        )
+        market_employee = Employee(
+            employee_no="E-BUDGET-LIST-MARKET",
+            name="王五",
+            email="budget-list-market@example.com",
+            organization_unit=market_department,
+        )
+        db.add_all([delivery_department, market_department, budget_manager, employee, market_employee])
+        db.flush()
+        db.add_all(
+            [
+                ExpenseClaim(
+                    claim_no="APP-BUDGET-LIST-201",
+                    employee_id=employee.id,
+                    employee_name="张三",
+                    department_id=delivery_department.id,
+                    department_name="交付部",
+                    project_code="PRJ-BUDGET",
+                    expense_type="travel_application",
+                    reason="预算待审申请",
+                    location="上海",
+                    amount=Decimal("12000.00"),
+                    currency="CNY",
+                    invoice_count=0,
+                    occurred_at=datetime(2026, 5, 12, 9, 0, tzinfo=UTC),
+                    submitted_at=datetime(2026, 5, 12, 10, 0, tzinfo=UTC),
+                    status="submitted",
+                    approval_stage="预算管理者审批",
+                    risk_flags_json=[],
+                ),
+                ExpenseClaim(
+                    claim_no="APP-BUDGET-LIST-OTHER-DEPT",
+                    employee_id=market_employee.id,
+                    employee_name="王五",
+                    department_id=market_department.id,
+                    department_name="市场部",
+                    project_code="PRJ-BUDGET",
+                    expense_type="travel_application",
+                    reason="其他部门预算待审申请",
+                    location="上海",
+                    amount=Decimal("13000.00"),
+                    currency="CNY",
+                    invoice_count=0,
+                    occurred_at=datetime(2026, 5, 12, 9, 0, tzinfo=UTC),
+                    submitted_at=datetime(2026, 5, 12, 10, 0, tzinfo=UTC),
+                    status="submitted",
+                    approval_stage="预算管理者审批",
+                    risk_flags_json=[],
+                ),
+                ExpenseClaim(
+                    claim_no="EXP-BUDGET-LIST-202",
+                    employee_id=employee.id,
+                    employee_name="张三",
+                    department_id=delivery_department.id,
+                    department_name="交付部",
+                    project_code="PRJ-BUDGET",
+                    expense_type="transport",
+                    reason="财务待审报销",
+                    location="上海",
+                    amount=Decimal("88.00"),
+                    currency="CNY",
+                    invoice_count=1,
+                    occurred_at=datetime(2026, 5, 12, 9, 0, tzinfo=UTC),
+                    submitted_at=datetime(2026, 5, 12, 10, 0, tzinfo=UTC),
+                    status="submitted",
+                    approval_stage="财务审批",
+                    risk_flags_json=[],
+                ),
+            ]
+        )
+        db.commit()
+
+        claims = ExpenseClaimService(db).list_approval_claims(current_user)
+
+        assert [claim.claim_no for claim in claims] == ["APP-BUDGET-LIST-201"]
+        claims_without_budget_role = ExpenseClaimService(db).list_approval_claims(p8_without_budget_role)
+        assert [claim.claim_no for claim in claims_without_budget_role] == []
diff --git a/server/tests/test_knowledge_rag_runtime.py b/server/tests/test_knowledge_rag_runtime.py
new file mode 100644
index 0000000..392925b
--- /dev/null
+++ b/server/tests/test_knowledge_rag_runtime.py
@@ -0,0 +1,35 @@
+from __future__ import annotations
+
+import pytest
+
+from app.services.knowledge_rag_runtime import (
+    KnowledgeRagError,
+    RuntimeModelConfig,
+    _LightRagRuntime,
+)
+
+
+def test_embedding_probe_error_includes_model_context(monkeypatch) -> None:
+    runtime = _LightRagRuntime.__new__(_LightRagRuntime)
+    config = RuntimeModelConfig(
+        slot="embedding",
+        provider="GLM",
+        model="Embedding-3",
+        endpoint="https://open.bigmodel.cn/api/paas/v4/",
+        api_key="token",
+        capability="embedding",
+    )
+
+    def fail_embeddings(*_args, **_kwargs):
+        raise KnowledgeRagError("token expired")
+
+    monkeypatch.setattr(runtime, "_request_embeddings", fail_embeddings)
+
+    with pytest.raises(KnowledgeRagError) as exc_info:
+        runtime._probe_embedding_dimension(config)
+
+    message = str(exc_info.value)
+    assert "slot=embedding" in message
+    assert "provider=GLM" in message
+    assert "model=Embedding-3" in message
+    assert "token expired" in message
diff --git a/server/tests/test_knowledge_rag_service.py b/server/tests/test_knowledge_rag_service.py
index f18ee0c..fd2941d 100644
--- a/server/tests/test_knowledge_rag_service.py
+++ b/server/tests/test_knowledge_rag_service.py
@@ -238,7 +238,7 @@ def test_resolve_default_qdrant_url_falls_back_to_loopback(monkeypatch) -> None:
     assert knowledge_rag_module._resolve_default_qdrant_url() == "http://127.0.0.1:6333"
 
 
-def test_runtime_cache_is_isolated_by_thread(monkeypatch) -> None:
+def test_runtime_cache_uses_dedicated_instance_across_calling_threads(monkeypatch) -> None:
     knowledge_rag_module.shutdown_knowledge_rag_runtime()
     created_runtimes = []
 
@@ -270,8 +270,8 @@ def test_runtime_cache_is_isolated_by_thread(monkeypatch) -> None:
     thread.start()
     thread.join(timeout=5)
 
-    assert len(created_runtimes) == 2
-    assert worker_runtimes[0] is not main_runtime
+    assert len(created_runtimes) == 1
+    assert worker_runtimes[0] is main_runtime
 
     knowledge_rag_module.shutdown_knowledge_rag_runtime()
     assert all(runtime.finalized for runtime in created_runtimes)
diff --git a/server/tests/test_knowledge_service.py b/server/tests/test_knowledge_service.py
index 7141e83..0867038 100644
--- a/server/tests/test_knowledge_service.py
+++ b/server/tests/test_knowledge_service.py
@@ -12,6 +12,7 @@ from app.db.base import Base
 from app.services.agent_runs import AgentRunService
 from app.services.knowledge import (
     KNOWLEDGE_INGEST_STATUS_FAILED,
+    KNOWLEDGE_INGEST_STATUS_INGESTED,
     KNOWLEDGE_INGEST_STATUS_SYNCING,
     KnowledgeService,
 )
@@ -88,3 +89,41 @@ def test_reconcile_document_ingest_status_keeps_failed_when_linked_run_failed(
         entry = next(item for item in index["documents"] if item["id"] == uploaded.id)
         assert changed is True
         assert entry["ingest_status"] == KNOWLEDGE_INGEST_STATUS_FAILED
+
+
+def test_reconcile_document_ingest_status_preserves_ingested_when_status_map_missing(
+    tmp_path,
+    monkeypatch,
+) -> None:
+    service = KnowledgeService(storage_root=tmp_path)
+    uploaded = service.upload_document(
+        "报销制度",
+        "demo.txt",
+        b"hello",
+        CurrentUserContext(
+            username="admin",
+            name="管理员",
+            role_codes=["manager"],
+            is_admin=True,
+        ),
+    )
+    service.set_document_ingest_statuses(
+        [uploaded.id],
+        KNOWLEDGE_INGEST_STATUS_INGESTED,
+        agent_run_id="run_missing_status_map",
+    )
+    monkeypatch.setattr(
+        "app.services.knowledge_rag.KnowledgeRagService.get_document_status_map",
+        lambda self, _document_ids: {},
+    )
+
+    index = service._load_index()
+    changed = service._reconcile_document_ingest_statuses(
+        index,
+        document_ids=[uploaded.id],
+        preserve_syncing=False,
+    )
+
+    entry = next(item for item in index["documents"] if item["id"] == uploaded.id)
+    assert changed is False
+    assert entry["ingest_status"] == KNOWLEDGE_INGEST_STATUS_INGESTED
diff --git a/server/tests/test_knowledge_sync.py b/server/tests/test_knowledge_sync.py
new file mode 100644
index 0000000..a43c30b
--- /dev/null
+++ b/server/tests/test_knowledge_sync.py
@@ -0,0 +1,65 @@
+from __future__ import annotations
+
+from sqlalchemy import create_engine
+from sqlalchemy.orm import Session, sessionmaker
+from sqlalchemy.pool import StaticPool
+
+from app.api.deps import CurrentUserContext
+from app.db.base import Base
+from app.services.knowledge import KNOWLEDGE_INGEST_STATUS_INGESTED, KnowledgeService
+from app.services.knowledge_sync import KnowledgeSyncDispatchService
+
+
+def build_session() -> Session:
+    engine = create_engine(
+        "sqlite+pysqlite:///:memory:",
+        connect_args={"check_same_thread": False},
+        poolclass=StaticPool,
+    )
+    Base.metadata.create_all(bind=engine)
+    session_factory = sessionmaker(bind=engine, autoflush=False, autocommit=False)
+    return session_factory()
+
+
+def test_force_sync_queues_ingested_documents_and_creates_hermes_run(tmp_path, monkeypatch) -> None:
+    submitted: list[dict[str, object]] = []
+    user = CurrentUserContext(
+        username="admin",
+        name="管理员",
+        role_codes=["manager"],
+        is_admin=True,
+    )
+
+    with build_session() as db:
+        knowledge_service = KnowledgeService(storage_root=tmp_path, db=db)
+        uploaded = knowledge_service.upload_document("报销制度", "demo.txt", b"hello", user)
+        document_id = uploaded.id
+        knowledge_service.set_document_ingest_statuses(
+            [document_id],
+            KNOWLEDGE_INGEST_STATUS_INGESTED,
+            agent_run_id="run_previous",
+        )
+        monkeypatch.setattr(
+            "app.services.knowledge_rag.KnowledgeRagService.get_document_status_map",
+            lambda self, _document_ids: {},
+        )
+        monkeypatch.setattr(
+            "app.services.knowledge_sync.knowledge_index_task_manager.submit_sync",
+            lambda **kwargs: submitted.append(kwargs),
+        )
+
+        dispatch_service = KnowledgeSyncDispatchService(db)
+        dispatch_service.knowledge_service = knowledge_service
+
+        result = dispatch_service.queue_sync(
+            current_user=user,
+            folder=None,
+            document_ids=[document_id],
+            force=True,
+            changed_only=True,
+        )
+
+        assert result.agent_run_id.startswith("run_")
+        assert document_id in result.document_ids
+        assert submitted
+        assert submitted[0]["agent_run_id"] == result.agent_run_id
diff --git a/server/tests/test_settings_persistence.py b/server/tests/test_settings_persistence.py
index 92c1fc7..66d1eb1 100644
--- a/server/tests/test_settings_persistence.py
+++ b/server/tests/test_settings_persistence.py
@@ -14,6 +14,7 @@ from app.core import admin_secret
 from app.core import secret_box
 from app.core.secret_box import encrypt_secret
 from app.db.base import Base
+from app.models.hermes_config import HermesTaskConfig, HermesTaskExecutionLog
 from app.models.system_model_setting import SystemModelSetting
 from app.models.system_setting import SystemSetting
 from app.models.system_setting_secret import SystemSettingSecret
@@ -27,9 +28,11 @@ def build_session(db_file: Path) -> Session:
         f"sqlite+pysqlite:///{db_file.as_posix()}",
         connect_args={"check_same_thread": False},
     )
-    SystemSetting.__table__.create(bind=engine)
-    SystemSettingSecret.__table__.create(bind=engine)
-    SystemModelSetting.__table__.create(bind=engine)
+    SystemSetting.__table__.create(bind=engine)
+    SystemSettingSecret.__table__.create(bind=engine)
+    SystemModelSetting.__table__.create(bind=engine)
+    HermesTaskConfig.__table__.create(bind=engine)
+    HermesTaskExecutionLog.__table__.create(bind=engine)
     session_factory = sessionmaker(bind=engine, autoflush=False, autocommit=False)
     return session_factory()
 
@@ -45,9 +48,12 @@ def test_settings_service_persists_non_secret_and_secret_fields(monkeypatch) ->
     monkeypatch.setenv("HERMES_HOME", str(temp_dir / ".hermes"))
 
     with build_session(temp_dir / "settings.db") as db:
-        service = SettingsService(db)
-        initial_snapshot = service.get_settings_snapshot()
-        payload = initial_snapshot.model_dump()
+        service = SettingsService(db)
+        initial_snapshot = service.get_settings_snapshot()
+        assert initial_snapshot.llmForm.mainModel == "codex-mini-latest"
+        assert initial_snapshot.llmForm.mainEndpoint == "https://api.openai.com/v1"
+        assert initial_snapshot.llmForm.mainApiKey == ""
+        payload = initial_snapshot.model_dump()
 
         payload["companyForm"]["companyName"] = "YGSOFT"
         payload["companyForm"]["displayName"] = "云广软件"
diff --git a/server/tests/test_user_agent_service.py b/server/tests/test_user_agent_service.py
index d57d63e..7a3ff10 100644
--- a/server/tests/test_user_agent_service.py
+++ b/server/tests/test_user_agent_service.py
@@ -167,7 +167,9 @@ def test_user_agent_knowledge_prompt_enforces_knowledge_boundary() -> None:
         assert "不能用常识、外部知识或主观推断补齐缺失条件" in messages[0]["content"]
         assert "不能只依赖排在最前面的片段" in messages[0]["content"]
         assert "不能把第一列的数值直接套给后面的列名" in messages[0]["content"]
-        assert "最终答复必须像助手在认真回答问题" in messages[0]["content"]
+        assert "最终答复必须使用 Markdown" in messages[0]["content"]
+        assert "## 结论" in messages[0]["content"]
+        assert "如果不能，必须明确说当前知识库没有找到直接依据" in messages[0]["content"]
         assert "禁止使用“已命中”“答案整理阶段”“稍后重试”" in messages[0]["content"]
         assert "knowledge_evidence_blocks" in messages[0]["content"]
         assert '"knowledge_answer_evidence": []' in messages[1]["content"]
@@ -435,6 +437,9 @@ def test_user_agent_knowledge_fallback_is_honest_and_personalized() -> None:
         )
 
         assert answer.startswith("张三，您好。")
+        assert "## 结论" in answer
+        assert "## 依据" in answer
+        assert "## 说明" in answer
         assert "我先根据当前制度依据给出可以确认的部分" in answer
         assert "已命中" not in answer
         assert "答案整理阶段本轮没有及时返回" not in answer
@@ -477,8 +482,8 @@ def test_user_agent_knowledge_answer_generation_uses_fast_timeouts(monkeypatch)
         )
 
         assert answer == "测试回答"
-        assert captured["timeout_seconds"] == 5
-        assert captured["slot_timeouts"] == {"main": 3, "backup": 5}
+        assert captured["timeout_seconds"] == 30
+        assert captured["slot_timeouts"] == {"main": 20, "backup": 30}
         assert captured["max_attempts"] == 1
 
 
@@ -549,7 +554,7 @@ def test_user_agent_knowledge_terms_keep_accounting_subject_in_long_query() -> N
     assert "会计科目" in terms
 
 
-def test_user_agent_uses_fast_knowledge_answer_without_model(monkeypatch) -> None:
+def test_user_agent_knowledge_answer_uses_model_after_retrieval(monkeypatch) -> None:
     session_factory = build_session_factory()
     with session_factory() as db:
         ontology = SemanticOntologyService(db).parse(
@@ -560,11 +565,14 @@ def test_user_agent_uses_fast_knowledge_answer_without_model(monkeypatch) -> Non
             )
         )
         service = UserAgentService(db)
-        monkeypatch.setattr(
-            service,
-            "_generate_answer_with_model",
-            lambda *args, **kwargs: (_ for _ in ()).throw(AssertionError("model should not be called")),
-        )
+        captured: dict[str, object] = {}
+
+        def fake_generate_answer(payload, **kwargs):
+            captured["payload"] = payload
+            captured.update(kwargs)
+            return "## 结论\n\n员工应在费用发生后 30 日内提交报销申请。"
+
+        monkeypatch.setattr(service, "_generate_answer_with_model", fake_generate_answer)
 
         response = service.respond(
             UserAgentRequest(
@@ -593,10 +601,11 @@ def test_user_agent_uses_fast_knowledge_answer_without_model(monkeypatch) -> Non
             )
         )
 
-        assert response.answer.startswith("张三，您好。")
-        assert "**结论**" in response.answer
+        assert captured["payload"].ontology.scenario == "knowledge"
+        assert "费用报销制度" in captured["fallback_answer"]
+        assert "## 依据" in captured["fallback_answer"]
+        assert response.answer.startswith("## 结论")
         assert "30 日内提交报销申请" in response.answer
-        assert "## 依据" not in response.answer
         assert "答案整理阶段本轮没有及时返回" not in response.answer
 
 
@@ -804,7 +813,8 @@ def test_user_agent_fast_knowledge_answer_renders_relevant_table_preview() -> No
         assert "| 项目 | 港澳台 | 其他地区 | 国外 |" in answer
         assert "| 餐补 | 75 | 55 | 140 |" in answer
         assert "餐补的标准为" in answer
-        assert "## 依据" not in answer
+        assert "## 结论" in answer
+        assert "## 依据" in answer
 
 
 def test_user_agent_fast_knowledge_answer_uses_user_grade_for_table_row() -> None:
@@ -906,8 +916,8 @@ def test_user_agent_fast_knowledge_answer_notes_missing_location_grounding() ->
 
         assert answer is not None
         assert "没有直接写出“北京”对应的地区档位或映射关系" in answer
-        assert "**说明**" in answer
-        assert "## 依据" not in answer
+        assert "## 说明" in answer
+        assert "## 依据" in answer
 
 
 def test_user_agent_fast_knowledge_answer_expands_lead_in_list_items() -> None:
@@ -952,12 +962,12 @@ def test_user_agent_fast_knowledge_answer_expands_lead_in_list_items() -> None:
         )
 
         assert answer is not None
-        assert "**结论**" in answer
+        assert "## 结论" in answer
         assert "登机牌、高速道路通行记录" in answer
         assert "支付记录" in answer
         assert "出差审批邮件、短信、微信等" in answer
         assert "（3）" not in answer
-        assert "## 依据" not in answer
+        assert "## 依据" in answer
 
 
 def test_user_agent_model_prompt_supports_contextual_personalization() -> None:
diff --git a/web/src/assets/styles/components/sidebar-rail.css b/web/src/assets/styles/components/sidebar-rail.css
index 6e6d207..dcbc8d3 100644
--- a/web/src/assets/styles/components/sidebar-rail.css
+++ b/web/src/assets/styles/components/sidebar-rail.css
@@ -506,6 +506,29 @@
   width: 100%;
 }
 
+:global(.rail-tooltip-popper) {
+  max-width: 180px;
+  padding: 7px 10px !important;
+  border: 1px solid rgba(var(--theme-primary-rgb, 58, 124, 165), 0.22) !important;
+  border-radius: 4px !important;
+  background: rgba(255, 255, 255, 0.98) !important;
+  color: #1f2937 !important;
+  box-shadow: 0 12px 28px rgba(15, 23, 42, 0.14) !important;
+  font-size: 12px;
+  font-weight: 700;
+  line-height: 1.4;
+  letter-spacing: 0;
+}
+
+:global(.rail-tooltip-popper.el-popper.is-light) {
+  border-color: rgba(var(--theme-primary-rgb, 58, 124, 165), 0.22) !important;
+}
+
+:global(.rail-tooltip-popper .el-popper__arrow::before) {
+  border-color: rgba(var(--theme-primary-rgb, 58, 124, 165), 0.22) !important;
+  background: rgba(255, 255, 255, 0.98) !important;
+}
+
 @keyframes railUserMenuIn {
   from {
     opacity: 0;
diff --git a/web/src/assets/styles/components/top-bar.css b/web/src/assets/styles/components/top-bar.css
index 94c838d..601bb99 100644
--- a/web/src/assets/styles/components/top-bar.css
+++ b/web/src/assets/styles/components/top-bar.css
@@ -270,6 +270,12 @@
   font-size: 14px;
 }
 
+.detail-alert-pill.neutral {
+  border-color: #d7e0ea;
+  background: #f8fafc;
+  color: #334155;
+}
+
 .detail-alert-pill.success {
   border-color: var(--success-line);
   background: var(--success-soft);
diff --git a/web/src/assets/styles/components/travel-reimbursement-insight-panel.css b/web/src/assets/styles/components/travel-reimbursement-insight-panel.css
index 5680ec9..339affb 100644
--- a/web/src/assets/styles/components/travel-reimbursement-insight-panel.css
+++ b/web/src/assets/styles/components/travel-reimbursement-insight-panel.css
@@ -508,3 +508,17 @@
     max-height: min(34dvh, 360px);
   }
 }
+
+.review-insight-title-copy {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.review-insight-title-copy .title-icon {
+  font-size: 20px;
+  color: var(--theme-primary, #3a7ca5);
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+}
diff --git a/web/src/assets/styles/detail-page-corners.css b/web/src/assets/styles/detail-page-corners.css
new file mode 100644
index 0000000..352898a
--- /dev/null
+++ b/web/src/assets/styles/detail-page-corners.css
@@ -0,0 +1,210 @@
+:root {
+  --enterprise-detail-radius: 4px;
+}
+
+.approval-page.approval-page .approval-detail :is(
+  .detail-hero,
+  .progress-card,
+  .detail-card,
+  .detail-side-card,
+  .detail-note,
+  .detail-note-editor textarea,
+  .detail-expense-table,
+  .detail-attachment,
+  .risk-list,
+  .application-detail-fact,
+  .application-budget-analysis__state,
+  .application-budget-analysis__metrics article,
+  .application-budget-analysis__summary,
+  .application-budget-analysis__lists > div,
+  .application-leader-opinion,
+  .application-leader-opinion-event,
+  .draft-blocking-note,
+  .draft-blocking-issue,
+  .expense-file-chip,
+  .expense-editor-panel,
+  .expense-editor-grid input,
+  .expense-editor-grid select,
+  .expense-total-under-table,
+  .attachment-insight-pane,
+  .attachment-source-pane,
+  .attachment-preview-card,
+  .attachment-preview-nav,
+  .attachment-preview-close,
+  .attachment-preview-alert,
+  .attachment-preview-action,
+  .attachment-preview-empty,
+  .attachment-risk-card,
+  .attachment-insight-section,
+  .risk-summary-card,
+  .risk-detail-card,
+  .risk-advice-card,
+  .risk-advice-meta > div,
+  .risk-override-card,
+  .risk-override-nav-btn,
+  .system-row-lock,
+  .system-attachment-note,
+  .submit-confirm-summary,
+  .smart-entry-btn,
+  .icon-action,
+  .inline-action
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.approval-page.approval-page .ai-entry-modal :is(
+  .detail-modal,
+  .modal-card,
+  .close-btn,
+  .ai-chat-card,
+  .ai-preview-card,
+  .ai-chat-content,
+  .ai-composer,
+  .ai-composer-surface,
+  .ai-tool-btn,
+  .ai-upload-btn,
+  .ai-send-btn,
+  .preview-field,
+  .preview-empty,
+  .ai-preview-secondary,
+  .ai-preview-primary,
+  .modal-action
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.approval-page.approval-page :is(
+  .approval-opinion-field textarea,
+  .return-reason-option,
+  .return-reason-section textarea
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.log-detail-page.log-detail-page :is(
+  .detail-state,
+  .detail-state button,
+  .detail-hero,
+  .refresh-btn,
+  .detail-card,
+  .info-grid > div,
+  .feedback-grid > div,
+  .trace-step,
+  .code-block,
+  .inline-empty,
+  .detail-actions button,
+  .knowledge-ingest-panel,
+  .ingest-run-info,
+  .info-item,
+  .run-info-card,
+  .run-stat-card,
+  .graph-search,
+  .graph-toolbar button,
+  .graph-theater,
+  .graph-stage,
+  .graph-toolbar,
+  .graph-inspector,
+  .node-facts > div,
+  .node-meta,
+  .node-detail-panel,
+  .detail-section,
+  .evidence-document,
+  .evidence-chunk,
+  .evidence-empty,
+  .node-evidence-card,
+  .relation-detail-list button,
+  .detail-empty
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.skill-detail.skill-detail :is(
+  .detail-hero,
+  .detail-inline-state,
+  .detail-loading-state,
+  .review-note-block,
+  .hero-stat,
+  .detail-card,
+  .side-card,
+  .field input,
+  .field textarea,
+  .prompt-block textarea,
+  .json-editor,
+  .markdown-editor,
+  .spreadsheet-editor-shell,
+  .spreadsheet-workbench,
+  .spreadsheet-change-center,
+  .version-pair-card,
+  .change-center-item,
+  .change-record-preview,
+  .spreadsheet-meta-strip span,
+  .json-risk-editor-shell,
+  .json-risk-generation-failure,
+  .json-risk-meta-item,
+  .json-risk-description-text,
+  .json-risk-description-source,
+  .json-risk-flow-card,
+  .diagram-zoom-controls,
+  .rule-spreadsheet-stage,
+  .compare-panel,
+  .compare-summary-grid article,
+  .compare-sheet-list article,
+  .change-detail-meta article,
+  .compare-sheet-list span,
+  .compare-table-wrap,
+  .subtle-banner,
+  .preview-mode-note,
+  .prompt-block,
+  .contract-panel,
+  .version-row,
+  .version-modal-summary div,
+  .version-modal-note,
+  .review-submit-test-state,
+  .risk-rule-action-confirm,
+  .risk-rule-action-note,
+  .risk-rule-action-note textarea,
+  .review-submit-form input:not([type='checkbox']),
+  .review-submit-form select,
+  .review-submit-form textarea,
+  .review-submit-hint,
+  .publish-summary,
+  .empty-side-note,
+  .back-action,
+  .minor-action,
+  .major-action,
+  .mini-btn,
+  .risk-level-menu,
+  .risk-level-option
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.skill-detail.skill-detail .rule-drawer-backdrop :is(
+  .rule-drawer,
+  .rule-drawer-state,
+  .change-detail-meta article,
+  .compare-panel,
+  .compare-sheet-list span,
+  .compare-table-wrap
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
+
+.employee-center.employee-center .employee-detail :is(
+  .detail-hero,
+  .hero-profile,
+  .hero-tag,
+  .hero-stat,
+  .detail-card,
+  .side-card,
+  .history-row,
+  .field input,
+  .field textarea,
+  .role-option,
+  .sync-card,
+  .permission-pill,
+  .detail-actions button,
+  .detail-action-group
+) {
+  border-radius: var(--enterprise-detail-radius);
+}
diff --git a/web/src/assets/styles/views/documents-center-view.css b/web/src/assets/styles/views/documents-center-view.css
index abf7691..1354182 100644
--- a/web/src/assets/styles/views/documents-center-view.css
+++ b/web/src/assets/styles/views/documents-center-view.css
@@ -420,7 +420,7 @@
 
 table {
   width: 100%;
-  min-width: 1320px;
+  min-width: 1420px;
   border-collapse: collapse;
   table-layout: fixed;
 }
@@ -429,8 +429,9 @@ table {
 .col-created { width: 10%; }
 .col-stay { width: 9%; }
 .col-doc-type { width: 9%; }
-.col-scene { width: 10%; }
-.col-title { width: 18%; }
+.col-scene { width: 9%; }
+.col-initiator { width: 8%; }
+.col-title { width: 16%; }
 .col-amount { width: 9%; }
 .col-node { width: 12%; }
 .col-status { width: 8%; }
diff --git a/web/src/assets/styles/views/logs-view.css b/web/src/assets/styles/views/logs-view.css
index b238bac..7984053 100644
--- a/web/src/assets/styles/views/logs-view.css
+++ b/web/src/assets/styles/views/logs-view.css
@@ -85,6 +85,7 @@
 }
 
 .console-toolbar {
+  --logs-filter-control-height: 38px;
   display: grid;
   grid-template-columns: minmax(220px, 1.35fr) repeat(2, minmax(138px, 0.7fr)) auto;
   gap: 10px;
@@ -103,9 +104,9 @@
 }
 
 .field-input {
-  min-height: 38px;
+  min-height: var(--logs-filter-control-height);
   border: 1px solid #d8e1eb;
-  border-radius: 8px;
+  border-radius: 4px;
   background: #fff;
 }
 
@@ -130,10 +131,38 @@
   outline: none;
 }
 
+.console-toolbar :deep(.enterprise-select) {
+  width: 100%;
+}
+
+.console-toolbar :deep(.el-select__wrapper) {
+  min-height: var(--logs-filter-control-height);
+  height: var(--logs-filter-control-height);
+  padding: 0 12px;
+  border-radius: 4px;
+  box-shadow: 0 0 0 1px #d8e1eb inset;
+}
+
+.console-toolbar :deep(.el-select__wrapper:hover) {
+  box-shadow: 0 0 0 1px #b8c2d2 inset;
+}
+
+.console-toolbar :deep(.el-select__wrapper.is-focused) {
+  box-shadow:
+    0 0 0 1px var(--theme-primary) inset,
+    0 0 0 3px var(--theme-focus-ring);
+}
+
+.console-toolbar :deep(.el-select__placeholder),
+.console-toolbar :deep(.el-select__selected-item) {
+  font-size: 13px;
+  line-height: var(--logs-filter-control-height);
+}
+
 .toolbar-btn {
-  min-height: 38px;
+  min-height: var(--logs-filter-control-height);
   padding: 0 14px;
-  border-radius: 8px;
+  border-radius: 4px;
   border: 1px solid #d8e1eb;
   font-size: 13px;
   font-weight: 700;
diff --git a/web/src/assets/styles/views/policies-view.css b/web/src/assets/styles/views/policies-view.css
index 3c14120..f0fd85e 100644
--- a/web/src/assets/styles/views/policies-view.css
+++ b/web/src/assets/styles/views/policies-view.css
@@ -36,12 +36,12 @@
   overflow: hidden;
 }
 
-.panel-title {
-  display: flex;
-  align-items: flex-start;
-  justify-content: space-between;
-  gap: 16px;
-}
+.panel-title {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+}
 
 .panel-title h2,
 .preview-head h2 {
@@ -86,9 +86,17 @@
   background: transparent;
 }
 
-.file-search input:focus {
-  outline: none;
-}
+.file-search input:focus {
+  outline: none;
+}
+
+.panel-tools {
+  min-width: min(470px, 100%);
+  display: flex;
+  align-items: center;
+  justify-content: flex-end;
+  gap: 10px;
+}
 
 .library-body {
   min-height: 0;
@@ -101,11 +109,11 @@
 .folder-rail {
   min-height: 0;
   display: grid;
-  grid-template-rows: minmax(0, 1fr) auto;
+  grid-template-rows: minmax(0, 1fr);
   gap: 12px;
-  border-right: 1px solid #edf2f7;
-  padding-right: 12px;
-}
+  border-right: 1px solid #edf2f7;
+  padding-right: 12px;
+}
 
 .folder-tree {
   min-height: 0;
@@ -148,45 +156,38 @@
   font-size: 11px;
 }
 
-.folder-sync-block {
-  display: grid;
-  gap: 8px;
-}
-
-.new-folder-btn {
+.knowledge-sync-btn {
   min-height: 36px;
   display: inline-flex;
-  align-items: center;
-  justify-content: center;
-  gap: 8px;
-  border: 1px solid rgba(var(--theme-primary-rgb), .28);
-  border-radius: 8px;
-  background: var(--theme-primary-light-9);
-  color: var(--theme-primary-active);
-  font-size: 13px;
-  font-weight: 850;
-}
-
-.new-folder-btn.fixed {
-  border-color: rgba(148, 163, 184, 0.3);
-  background: #f8fafc;
-  color: #64748b;
-}
-
-.knowledge-sync-btn:not(:disabled) {
+  align-items: center;
+  justify-content: center;
+  gap: 7px;
+  padding: 0 14px;
+  border: 1px solid #1d4ed8;
+  border-radius: 8px;
+  background: #2563eb;
+  color: #fff;
+  font-size: 13px;
+  font-weight: 850;
+  white-space: nowrap;
+  box-shadow: 0 8px 18px rgba(37, 99, 235, 0.18);
   cursor: pointer;
+  transition: background 160ms ease, border-color 160ms ease, box-shadow 160ms ease, transform 160ms ease;
 }
 
-.knowledge-sync-btn:not(:disabled):hover {
-  border-color: rgba(var(--theme-primary-rgb), 0.38);
-  background: var(--theme-primary-light-9);
-  color: var(--theme-primary-active);
+.knowledge-sync-btn:hover:not(:disabled) {
+  border-color: #1e40af;
+  background: #1d4ed8;
+  box-shadow: 0 10px 22px rgba(37, 99, 235, 0.24);
+  transform: translateY(-1px);
 }
 
-.folder-sync-meta {
+.knowledge-sync-btn:disabled {
+  cursor: not-allowed;
+  border-color: #cbd5e1;
+  background: #e2e8f0;
   color: #64748b;
-  font-size: 12px;
-  line-height: 1.6;
+  box-shadow: none;
 }
 
 .document-area {
@@ -1179,6 +1180,12 @@ th {
     align-items: stretch;
   }
 
+  .panel-tools,
+  .file-search,
+  .knowledge-sync-btn {
+    width: 100%;
+  }
+
   .summary-grid,
   .list-foot {
     grid-template-columns: 1fr;
diff --git a/web/src/assets/styles/views/settings-view-form.css b/web/src/assets/styles/views/settings-view-form.css
index 6cdddf9..2568541 100644
--- a/web/src/assets/styles/views/settings-view-form.css
+++ b/web/src/assets/styles/views/settings-view-form.css
@@ -99,7 +99,8 @@
   font-style: normal;
 }
 
-.field input {
+.field input,
+.field :deep(.el-select__wrapper) {
   width: 100%;
   min-height: 44px;
   padding: 0 14px;
@@ -116,7 +117,8 @@
   color: #94a3b8;
 }
 
-.field input:focus {
+.field input:focus,
+.field :deep(.el-select__wrapper.is-focused) {
   outline: none;
   border-color: var(--theme-primary);
   box-shadow: 0 0 0 3px var(--theme-focus-ring);
diff --git a/web/src/components/layout/SidebarRail.vue b/web/src/components/layout/SidebarRail.vue
index 4446327..c3c38a1 100644
--- a/web/src/components/layout/SidebarRail.vue
+++ b/web/src/components/layout/SidebarRail.vue
@@ -9,33 +9,52 @@
         </svg>
       </div>
       <strong class="brand-name">{{ displayCompanyName }}</strong>
-      <button
-        class="rail-collapse-btn"
-        type="button"
-        :aria-label="collapsed ? '展开侧边栏' : '折叠侧边栏'"
-        :title="collapsed ? '展开侧边栏' : '折叠侧边栏'"
-        :aria-expanded="!collapsed"
-        @click="emit('toggle-collapse')"
+      <ElTooltip
+        :content="collapseTooltipContent"
+        placement="right"
+        effect="light"
+        :show-after="180"
+        :hide-after="0"
+        :offset="12"
+        popper-class="rail-tooltip-popper"
       >
-        <i :class="collapsed ? 'mdi mdi-chevron-right' : 'mdi mdi-chevron-left'"></i>
-      </button>
+        <button
+          class="rail-collapse-btn"
+          type="button"
+          :aria-label="collapseTooltipContent"
+          :aria-expanded="!collapsed"
+          @click="emit('toggle-collapse')"
+        >
+          <i :class="collapsed ? 'mdi mdi-chevron-right' : 'mdi mdi-chevron-left'"></i>
+        </button>
+      </ElTooltip>
     </div>
 
     <nav class="rail-nav" aria-label="功能导航">
-      <button
+      <ElTooltip
         v-for="item in decoratedNavItems"
         :key="item.id"
-        class="nav-btn"
-        :class="{ active: activeView === item.id }"
-        type="button"
-        :title="collapsed ? item.displayLabel : undefined"
-        @click="emit('navigate', item.id)"
+        :content="item.displayLabel"
+        placement="right"
+        effect="light"
+        :disabled="!collapsed"
+        :show-after="180"
+        :hide-after="0"
+        :offset="12"
+        popper-class="rail-tooltip-popper"
       >
-        <span class="nav-icon" v-html="item.icon"></span>
-        <span class="nav-label">{{ item.displayLabel }}</span>
-        <span v-if="item.hasNewMessage" class="nav-unread-dot" aria-hidden="true"></span>
-        <span v-if="item.badge" class="nav-badge">{{ item.badge }}</span>
-      </button>
+        <button
+          class="nav-btn"
+          :class="{ active: activeView === item.id }"
+          type="button"
+          @click="emit('navigate', item.id)"
+        >
+          <span class="nav-icon" v-html="item.icon"></span>
+          <span class="nav-label">{{ item.displayLabel }}</span>
+          <span v-if="item.hasNewMessage" class="nav-unread-dot" aria-hidden="true"></span>
+          <span v-if="item.badge" class="nav-badge">{{ item.badge }}</span>
+        </button>
+      </ElTooltip>
     </nav>
 
     <div
@@ -69,19 +88,31 @@
         </div>
       </Teleport>
 
-      <div class="user-summary" tabindex="0" aria-label="用户信息" :title="collapsed ? displayUser.name : undefined">
-        <span class="user-avatar">{{ displayUser.avatar }}</span>
-        <span class="user-copy">
-          <strong>{{ displayUser.name }}</strong>
-          <span>{{ displayUser.role }}</span>
-        </span>
-        <i class="mdi mdi-chevron-up"></i>
-      </div>
+      <ElTooltip
+        :content="userTooltipContent"
+        placement="top"
+        effect="light"
+        :disabled="!collapsed"
+        :show-after="180"
+        :hide-after="0"
+        :offset="10"
+        popper-class="rail-tooltip-popper"
+      >
+        <div class="user-summary" tabindex="0" :aria-label="userTooltipContent">
+          <span class="user-avatar">{{ displayUser.avatar }}</span>
+          <span class="user-copy">
+            <strong>{{ displayUser.name }}</strong>
+            <span>{{ displayUser.role }}</span>
+          </span>
+          <i class="mdi mdi-chevron-up"></i>
+        </div>
+      </ElTooltip>
     </div>
   </aside>
 </template>
 
 <script setup>
+import { ElTooltip } from 'element-plus'
 import { computed, onBeforeUnmount, onMounted, reactive, ref, watch } from 'vue'
 
 import { useDocumentCenterInbox } from '../../composables/useDocumentCenterInbox.js'
@@ -154,6 +185,8 @@ const displayUser = computed(() => ({
 }))
 
 const displayCompanyName = computed(() => props.companyName || 'X-Financial')
+const collapseTooltipContent = computed(() => (props.collapsed ? '展开侧边栏' : '折叠侧边栏'))
+const userTooltipContent = computed(() => [displayUser.value.name, displayUser.value.role].filter(Boolean).join(' · '))
 
 const userMenuOpen = ref(false)
 let userMenuCloseTimer = null
diff --git a/web/src/components/layout/TopBar.vue b/web/src/components/layout/TopBar.vue
index 68edb5e..a45ab82 100644
--- a/web/src/components/layout/TopBar.vue
+++ b/web/src/components/layout/TopBar.vue
@@ -91,7 +91,7 @@
             class="detail-alert-pill"
             :class="alert.tone"
           >
-            <i class="mdi mdi-alert-circle-outline"></i>
+            <i :class="alert.icon || 'mdi mdi-alert-circle-outline'"></i>
             <span>{{ alert.label }}</span>
           </span>
         </div>
diff --git a/web/src/components/shared/ReturnReasonDialog.vue b/web/src/components/shared/ReturnReasonDialog.vue
index 01c3c5d..c8f28c8 100644
--- a/web/src/components/shared/ReturnReasonDialog.vue
+++ b/web/src/components/shared/ReturnReasonDialog.vue
@@ -22,14 +22,15 @@
           <label
             v-for="option in options"
             :key="option.code"
-            :class="['return-reason-option', { active: selectedCodes.includes(option.code) }]"
+            :class="['return-reason-option', { active: isOptionActive(option.code) }]"
           >
             <input
-              v-model="selectedCodes"
-              type="checkbox"
+              :type="application ? 'radio' : 'checkbox'"
+              :name="application ? 'application-return-reason' : undefined"
+              :checked="isOptionActive(option.code)"
               :value="option.code"
               :disabled="busy"
-              @change="handleOptionChange"
+              @change="handleOptionChange(option)"
             />
             <i :class="option.icon"></i>
             <strong>{{ option.label }}</strong>
@@ -99,6 +100,12 @@ const APPLICATION_RETURN_REASON_OPTIONS = [
     label: '前置材料需补充',
     icon: 'mdi mdi-file-document-plus-outline',
     defaultReason: '请补充会议通知、客户邀约、项目安排或其他能支撑申请必要性的材料。'
+  },
+  {
+    code: 'application_other',
+    label: '其他',
+    icon: 'mdi mdi-pencil-box-outline',
+    defaultReason: ''
   }
 ]
 
@@ -117,12 +124,18 @@ const props = defineProps({
 const emit = defineEmits(['close', 'confirm'])
 
 const selectedCodes = ref([])
+const selectedApplicationCode = ref('')
 const reasonText = ref('')
 const touched = ref(false)
 const selectionTouched = ref(false)
 const lastAutoReason = ref('')
 
 const options = computed(() => (props.application ? APPLICATION_RETURN_REASON_OPTIONS : CLAIM_RETURN_REASON_OPTIONS))
+const selectedReasonCodes = computed(() => (
+  props.application
+    ? (selectedApplicationCode.value ? [selectedApplicationCode.value] : [])
+    : selectedCodes.value
+))
 const dialogBadge = computed(() => (props.application ? '退回申请' : '退回单据'))
 const optionsTitle = computed(() => (props.application ? '退单选项' : '默认风险点'))
 const optionsAriaLabel = computed(() => (props.application ? '申请退单选项' : '默认退回风险点'))
@@ -133,10 +146,10 @@ const reasonPlaceholder = computed(() => (
 ))
 const trimmedReason = computed(() => reasonText.value.trim())
 const selectionError = computed(() => {
-  if (!props.application || !selectionTouched.value || selectedCodes.value.length > 0) {
+  if (!props.application || !selectionTouched.value || selectedReasonCodes.value.length > 0) {
     return ''
   }
-  return '请选择至少一个退单选项，便于后续看板统计。'
+  return '请选择一个退单选项，便于后续看板统计。'
 })
 const reasonError = computed(() => {
   if (!touched.value || trimmedReason.value.length >= 6) {
@@ -159,6 +172,7 @@ watch(
   (open) => {
     if (open) {
       selectedCodes.value = []
+      selectedApplicationCode.value = ''
       reasonText.value = ''
       touched.value = false
       selectionTouched.value = false
@@ -167,25 +181,35 @@ watch(
   }
 )
 
-watch(selectedCodes, () => {
-  if (!props.application) {
-    return
-  }
-
-  const defaultReason = selectedCodes.value
-    .map((code) => options.value.find((option) => option.code === code)?.defaultReason || '')
-    .filter(Boolean)
-    .join('\n')
-
+function syncApplicationDefaultReason(option) {
+  const defaultReason = String(option?.defaultReason || '').trim()
   const canAutoFill = !touched.value || !reasonText.value.trim() || reasonText.value === lastAutoReason.value
   if (canAutoFill) {
     reasonText.value = defaultReason
   }
   lastAutoReason.value = defaultReason
-})
+}
 
-function handleOptionChange() {
+function isOptionActive(code) {
+  return props.application ? selectedApplicationCode.value === code : selectedCodes.value.includes(code)
+}
+
+function handleOptionChange(option) {
   selectionTouched.value = true
+
+  if (props.application) {
+    selectedApplicationCode.value = option.code
+    syncApplicationDefaultReason(option)
+    return
+  }
+
+  const selected = new Set(selectedCodes.value)
+  if (selected.has(option.code)) {
+    selected.delete(option.code)
+  } else {
+    selected.add(option.code)
+  }
+  selectedCodes.value = Array.from(selected)
 }
 
 function handleClose() {
@@ -197,13 +221,13 @@ function handleClose() {
 function handleConfirm() {
   touched.value = true
   selectionTouched.value = true
-  if ((props.application && selectedCodes.value.length === 0) || trimmedReason.value.length < 6 || props.busy) {
+  if ((props.application && selectedReasonCodes.value.length === 0) || trimmedReason.value.length < 6 || props.busy) {
     return
   }
 
   emit('confirm', {
     reason: trimmedReason.value,
-    reason_codes: [...selectedCodes.value]
+    reason_codes: [...selectedReasonCodes.value]
   })
 }
 </script>
diff --git a/web/src/components/travel/TravelReimbursementInsightPanel.vue b/web/src/components/travel/TravelReimbursementInsightPanel.vue
index 4935326..9dfac94 100644
--- a/web/src/components/travel/TravelReimbursementInsightPanel.vue
+++ b/web/src/components/travel/TravelReimbursementInsightPanel.vue
@@ -16,6 +16,7 @@
         </div>
         <div v-else class="review-insight-title-row">
           <div class="review-insight-title-copy">
+            <i v-if="!ui.activeReviewPayload && ui.isReviewFlowDrawer" :class="ui.reviewFlowDrawerIcon" class="title-icon"></i>
             <h3>{{ ui.reviewDrawerTitle }}</h3>
           </div>
         </div>
@@ -23,7 +24,7 @@
         <p v-if="!ui.activeReviewPayload && !ui.isReviewFlowDrawer">{{ ui.currentInsight.summary }}</p>
       </div>
 
-      <div v-if="ui.activeReviewPayload || ui.isReviewFlowDrawer" class="review-insight-tools">
+      <div v-if="ui.activeReviewPayload" class="review-insight-tools">
         <button
           v-if="ui.activeReviewPayload && ui.reviewOverviewDrawerAvailable"
           type="button"
diff --git a/web/src/components/travel/TravelRequestApprovalDialog.vue b/web/src/components/travel/TravelRequestApprovalDialog.vue
index d701a23..ebb00e5 100644
--- a/web/src/components/travel/TravelRequestApprovalDialog.vue
+++ b/web/src/components/travel/TravelRequestApprovalDialog.vue
@@ -14,21 +14,6 @@
     @close="emit('close')"
     @confirm="emit('confirm')"
   >
-    <div class="submit-confirm-summary" aria-label="领导审批通过摘要">
-      <div class="submit-confirm-row">
-        <span>单据编号</span>
-        <strong>{{ documentNo }}</strong>
-      </div>
-      <div class="submit-confirm-row">
-        <span>当前节点</span>
-        <strong>{{ node }}</strong>
-      </div>
-      <div class="submit-confirm-row">
-        <span>{{ summaryLabel }}</span>
-        <strong>{{ nextStage }}</strong>
-      </div>
-    </div>
-
     <label class="approval-opinion-field">
       <span>
         {{ opinionTitle }}
@@ -64,10 +49,6 @@ const props = defineProps({
   confirmText: { type: String, required: true },
   busyText: { type: String, required: true },
   busy: { type: Boolean, required: true },
-  documentNo: { type: [String, Number], required: true },
-  node: { type: String, default: '' },
-  summaryLabel: { type: String, required: true },
-  nextStage: { type: String, required: true },
   opinionTitle: { type: String, required: true },
   opinion: { type: String, default: '' },
   opinionPlaceholder: { type: String, default: '' },
diff --git a/web/src/components/travel/TravelRequestBudgetAnalysis.vue b/web/src/components/travel/TravelRequestBudgetAnalysis.vue
new file mode 100644
index 0000000..8d77ded
--- /dev/null
+++ b/web/src/components/travel/TravelRequestBudgetAnalysis.vue
@@ -0,0 +1,293 @@
+<template>
+  <section class="application-budget-analysis" aria-label="预算分析">
+    <div class="application-budget-analysis__head">
+      <span><i class="mdi mdi-chart-donut"></i>预算分析</span>
+      <strong v-if="analysis && !loading">{{ scoreLabel }}</strong>
+    </div>
+
+    <div v-if="loading" class="application-budget-analysis__state">正在读取预算管控模型...</div>
+    <div v-else-if="error" class="application-budget-analysis__state danger">{{ error }}</div>
+    <div v-else-if="analysis" class="application-budget-analysis__body">
+      <div class="application-budget-analysis__metrics">
+        <article v-for="metric in metrics" :key="metric.key">
+          <span>{{ metric.label }}</span>
+          <strong>{{ metric.value }}</strong>
+        </article>
+      </div>
+
+      <div class="application-budget-analysis__summary">
+        <div :class="['application-budget-score', analysis.risk_level || 'low']">
+          <span>{{ analysis.score }}</span>
+          <em>综合评分</em>
+        </div>
+        <p>{{ analysis.summary }}</p>
+      </div>
+
+      <div class="application-budget-analysis__lists">
+        <div>
+          <span>规则依据</span>
+          <ul>
+            <li v-for="item in analysis.basis" :key="item">{{ item }}</li>
+          </ul>
+        </div>
+        <div>
+          <span>模型建议</span>
+          <ul>
+            <li v-for="item in analysis.suggestions" :key="item">{{ item }}</li>
+          </ul>
+        </div>
+      </div>
+    </div>
+  </section>
+</template>
+
+<script setup>
+import { computed, ref, watch } from 'vue'
+
+import { fetchExpenseClaimBudgetAnalysis } from '../../services/reimbursements.js'
+
+const props = defineProps({
+  claimId: { type: String, default: '' }
+})
+
+const analysis = ref(null)
+const loading = ref(false)
+const error = ref('')
+
+const scoreLabel = computed(() => {
+  const labels = {
+    recommended: '建议通过',
+    caution: '谨慎通过',
+    review: '需要复核',
+    block: '不建议直接通过',
+    reference: '参考口径'
+  }
+  return labels[String(analysis.value?.rating || '').trim()] || '模型建议'
+})
+
+const metrics = computed(() => {
+  const metric = analysis.value?.metrics || {}
+  return [
+    { key: 'total', label: '当前预算额度', value: formatMoney(metric.total_amount) },
+    { key: 'ratio', label: '此次费用占预算', value: formatPercent(metric.claim_amount_ratio) },
+    { key: 'after', label: '审批后使用率', value: formatPercent(metric.after_usage_rate) },
+    { key: 'available', label: '当前可用预算', value: formatMoney(metric.available_amount) }
+  ]
+})
+
+watch(
+  () => props.claimId,
+  (claimId) => {
+    loadAnalysis(claimId)
+  },
+  { immediate: true }
+)
+
+async function loadAnalysis(claimId) {
+  const normalizedClaimId = String(claimId || '').trim()
+  analysis.value = null
+  error.value = ''
+  if (!normalizedClaimId) {
+    return
+  }
+
+  loading.value = true
+  try {
+    analysis.value = await fetchExpenseClaimBudgetAnalysis(normalizedClaimId)
+  } catch (err) {
+    error.value = err?.message || '预算分析加载失败，请稍后重试。'
+  } finally {
+    loading.value = false
+  }
+}
+
+function formatMoney(value) {
+  const amount = Number(value)
+  if (!Number.isFinite(amount)) {
+    return '待匹配'
+  }
+  return `${amount.toLocaleString('zh-CN', { minimumFractionDigits: 2, maximumFractionDigits: 2 })} 元`
+}
+
+function formatPercent(value) {
+  const amount = Number(value)
+  if (!Number.isFinite(amount)) {
+    return '0.00%'
+  }
+  return `${amount.toFixed(2)}%`
+}
+</script>
+
+<style scoped>
+.application-budget-analysis {
+  display: grid;
+  gap: 14px;
+  margin-top: 18px;
+  padding-top: 18px;
+  border-top: 1px solid #e2e8f0;
+}
+
+.application-budget-analysis__head {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 12px;
+}
+
+.application-budget-analysis__head span {
+  display: inline-flex;
+  align-items: center;
+  gap: 8px;
+  color: #0f172a;
+  font-size: 15px;
+  font-weight: 850;
+}
+
+.application-budget-analysis__head i {
+  color: var(--theme-primary-active);
+  font-size: 18px;
+}
+
+.application-budget-analysis__head strong {
+  border-radius: 999px;
+  padding: 4px 10px;
+  background: rgba(var(--theme-primary-rgb), .1);
+  color: var(--theme-primary-active);
+  font-size: 12px;
+}
+
+.application-budget-analysis__state {
+  border-radius: 8px;
+  padding: 10px 12px;
+  background: #f8fafc;
+  color: #64748b;
+  font-size: 13px;
+}
+
+.application-budget-analysis__state.danger {
+  background: #fff1f2;
+  color: #be123c;
+}
+
+.application-budget-analysis__body {
+  display: grid;
+  gap: 14px;
+}
+
+.application-budget-analysis__metrics {
+  display: grid;
+  grid-template-columns: repeat(4, minmax(0, 1fr));
+  gap: 10px;
+}
+
+.application-budget-analysis__metrics article {
+  min-width: 0;
+  border-radius: 8px;
+  padding: 10px 12px;
+  background: #f8fafc;
+  border: 1px solid #e2e8f0;
+}
+
+.application-budget-analysis__metrics span,
+.application-budget-analysis__lists span {
+  display: block;
+  color: #64748b;
+  font-size: 12px;
+  font-weight: 750;
+}
+
+.application-budget-analysis__metrics strong {
+  display: block;
+  margin-top: 5px;
+  color: #0f172a;
+  font-size: 15px;
+  font-weight: 850;
+}
+
+.application-budget-analysis__summary {
+  display: grid;
+  grid-template-columns: auto minmax(0, 1fr);
+  gap: 12px;
+  align-items: center;
+  border-radius: 8px;
+  padding: 12px;
+  background: #fff;
+  border: 1px solid #e2e8f0;
+}
+
+.application-budget-analysis__summary p {
+  margin: 0;
+  color: #334155;
+  font-size: 13px;
+  line-height: 1.7;
+}
+
+.application-budget-score {
+  display: grid;
+  place-items: center;
+  width: 72px;
+  height: 72px;
+  border-radius: 50%;
+  background: #eef6ff;
+  color: #2563eb;
+}
+
+.application-budget-score.medium {
+  background: #fff7ed;
+  color: #c2410c;
+}
+
+.application-budget-score.high {
+  background: #fff1f2;
+  color: #be123c;
+}
+
+.application-budget-score span {
+  font-size: 22px;
+  font-weight: 900;
+  line-height: 1;
+}
+
+.application-budget-score em {
+  font-style: normal;
+  font-size: 11px;
+  font-weight: 800;
+}
+
+.application-budget-analysis__lists {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 12px;
+}
+
+.application-budget-analysis__lists > div {
+  border-radius: 8px;
+  padding: 12px;
+  background: #f8fafc;
+  border: 1px solid #e2e8f0;
+}
+
+.application-budget-analysis__lists ul {
+  margin: 8px 0 0;
+  padding-left: 18px;
+  color: #334155;
+  font-size: 12px;
+  line-height: 1.7;
+}
+
+@media (max-width: 900px) {
+  .application-budget-analysis__metrics,
+  .application-budget-analysis__lists {
+    grid-template-columns: 1fr 1fr;
+  }
+}
+
+@media (max-width: 640px) {
+  .application-budget-analysis__metrics,
+  .application-budget-analysis__lists,
+  .application-budget-analysis__summary {
+    grid-template-columns: 1fr;
+  }
+}
+</style>
+
diff --git a/web/src/composables/useRequests.js b/web/src/composables/useRequests.js
index 3ff08cd..152edca 100644
--- a/web/src/composables/useRequests.js
+++ b/web/src/composables/useRequests.js
@@ -57,6 +57,7 @@ const REIMBURSEMENT_PROGRESS_LABELS = [
 const APPLICATION_PROGRESS_LABELS = [
   '创建申请',
   '直属领导审批',
+  '预算管理者审批',
   '审批完成'
 ]
 
@@ -386,10 +387,13 @@ function resolveApplicationProgressCurrentIndex(approvalMeta, workflowNode) {
   const normalizedNode = String(workflowNode || '').trim()
 
   if (approvalMeta.key === 'completed') {
-    return 2
+    return 3
   }
 
   if (normalizedNode.includes('审批完成') || normalizedNode.includes('申请完成')) {
+    return 3
+  }
+  if (normalizedNode.includes('预算')) {
     return 2
   }
   if (
@@ -437,15 +441,42 @@ function resolveApplicationApproverName(claim) {
   ) || '直属领导'
 }
 
+function resolveApplicationBudgetApproverName(claim) {
+  const routeEvent = findApprovalEventForStep(claim, '直属领导审批')
+  return resolveDisplayName(
+    routeEvent?.next_approver_name,
+    routeEvent?.nextApproverName,
+    routeEvent?.budget_approver_name,
+    routeEvent?.budgetApproverName
+  ) || 'P8预算监控者'
+}
+
 function resolveProgressDisplayLabel(label, documentTypeCode, claim, approvalMeta) {
+  const normalizedLabel = normalizeText(label)
+  const workflowNode = normalizeText(claim?.approval_stage || claim?.workflowNode)
   if (
     documentTypeCode === DOCUMENT_TYPE_APPLICATION
     && approvalMeta.key !== 'completed'
-    && normalizeText(label) === '直属领导审批'
+    && normalizedLabel === '直属领导审批'
+    && (
+      workflowNode.includes('直属领导')
+      || workflowNode.includes('领导审批')
+      || workflowNode.includes('部门负责人')
+      || workflowNode.includes('负责人审批')
+    )
   ) {
     return `等待 ${resolveApplicationApproverName(claim)} 批复`
   }
 
+  if (
+    documentTypeCode === DOCUMENT_TYPE_APPLICATION
+    && approvalMeta.key !== 'completed'
+    && normalizedLabel === '预算管理者审批'
+    && workflowNode.includes('预算')
+  ) {
+    return `等待 ${resolveApplicationBudgetApproverName(claim)} 批复`
+  }
+
   return label
 }
 
@@ -471,7 +502,7 @@ function findApprovalEventForStep(claim, label) {
     }
 
     const source = normalizeText(flag.source)
-    if (!['manual_approval', 'finance_approval'].includes(source)) {
+    if (!['manual_approval', 'budget_approval', 'finance_approval'].includes(source)) {
       return false
     }
 
@@ -482,10 +513,19 @@ function findApprovalEventForStep(claim, label) {
       return (
         previousStage.includes('直属领导')
         || previousStage.includes('领导审批')
+        || nextStage.includes('预算')
         || nextStage.includes('财务')
       )
     }
 
+    if (stepLabel === '预算管理者审批') {
+      return (
+        source === 'budget_approval'
+        || previousStage.includes('预算')
+        || nextStage.includes('审批完成')
+      )
+    }
+
     if (stepLabel === '财务审批') {
       return (
         previousStage.includes('财务')
@@ -557,15 +597,16 @@ function buildCompletedStepMeta(claim, label) {
     return buildProgressStepMeta('AI预审通过', reviewedAt)
   }
 
-  if (stepLabel === '直属领导审批' || stepLabel === '财务审批') {
+  if (stepLabel === '直属领导审批' || stepLabel === '预算管理者审批' || stepLabel === '财务审批') {
     const approvalEvent = findApprovalEventForStep(claim, stepLabel)
     if (approvalEvent) {
       const operator = resolveDisplayName(
         approvalEvent.operator,
         approvalEvent.operator_name,
         approvalEvent.operatorName,
-        stepLabel === '直属领导审批' ? claim?.manager_name : ''
-      ) || (stepLabel === '财务审批' ? '财务' : '直属领导')
+        stepLabel === '直属领导审批' ? claim?.manager_name : '',
+        stepLabel === '预算管理者审批' ? approvalEvent.next_approver_name : ''
+      ) || (stepLabel === '财务审批' ? '财务' : stepLabel === '预算管理者审批' ? '预算监控者' : '直属领导')
       const approvedAt = formatDateTime(approvalEvent.created_at || approvalEvent.createdAt)
       return buildProgressStepMeta(`${operator}通过`, approvedAt, `${operator}审批通过 ${approvedAt}`.trim())
     }
@@ -626,6 +667,10 @@ function resolveCurrentStepStartedAt(claim, label) {
   if (stepLabel === '直属领导审批') {
     return claim?.submitted_at || claim?.updated_at || claim?.created_at
   }
+  if (stepLabel === '预算管理者审批') {
+    const leaderApprovalEvent = findApprovalEventForStep(claim, '直属领导审批')
+    return leaderApprovalEvent?.created_at || leaderApprovalEvent?.createdAt || claim?.updated_at || claim?.submitted_at
+  }
   if (stepLabel === '财务审批') {
     const leaderApprovalEvent = findApprovalEventForStep(claim, '直属领导审批')
     return leaderApprovalEvent?.created_at || leaderApprovalEvent?.createdAt || claim?.updated_at || claim?.submitted_at
diff --git a/web/src/main.js b/web/src/main.js
index bb1373f..88e1231 100644
--- a/web/src/main.js
+++ b/web/src/main.js
@@ -9,6 +9,7 @@ import router from './router/index.js'
 import { installThemeSkin } from './composables/useThemeSkin.js'
 import { installSessionNavigation } from './composables/useSystemState.js'
 import './assets/styles/element-plus-theme.css'
+import './assets/styles/detail-page-corners.css'
 
 const app = createApp(App)
 
diff --git a/web/src/services/reimbursements.js b/web/src/services/reimbursements.js
index 0c9837f..c11c94c 100644
--- a/web/src/services/reimbursements.js
+++ b/web/src/services/reimbursements.js
@@ -16,6 +16,10 @@ export function fetchExpenseClaimDetail(claimId) {
   return apiRequest(`/reimbursements/claims/${encodeURIComponent(String(claimId || '').trim())}`)
 }
 
+export function fetchExpenseClaimBudgetAnalysis(claimId) {
+  return apiRequest(`/reimbursements/claims/${encodeURIComponent(String(claimId || '').trim())}/budget-analysis`)
+}
+
 export function updateExpenseClaim(claimId, payload = {}) {
   return apiRequest(`/reimbursements/claims/${encodeURIComponent(String(claimId || '').trim())}`, {
     method: 'PATCH',
diff --git a/web/src/utils/accessControl.js b/web/src/utils/accessControl.js
index a1bfa9d..ec1df7b 100644
--- a/web/src/utils/accessControl.js
+++ b/web/src/utils/accessControl.js
@@ -19,9 +19,10 @@ const VIEW_ROLE_RULES = {
   employees: ['manager'],
   settings: ['manager']
 }
-const CLAIM_MANAGER_ROLE_CODES = new Set(['executive'])
-const CLAIM_RETURN_ROLE_CODES = new Set(['finance', 'executive', 'manager', 'approver'])
-const CLAIM_LEADER_APPROVAL_ROLE_CODES = new Set(['manager', 'approver'])
+const CLAIM_MANAGER_ROLE_CODES = new Set(['executive'])
+const CLAIM_RETURN_ROLE_CODES = new Set(['finance', 'executive', 'manager', 'approver', 'budget_monitor'])
+const CLAIM_LEADER_APPROVAL_ROLE_CODES = new Set(['manager', 'approver'])
+const CLAIM_BUDGET_APPROVAL_GRADE = 'P8'
 
 function normalizedRoleCodes(user) {
   if (!user) {
@@ -55,6 +56,25 @@ function identityIntersects(leftValues, rightValues) {
   return leftValues.some((item) => rightSet.has(item))
 }
 
+function normalizedGrade(user) {
+  return String(user?.grade || user?.employeeGrade || '').trim().toUpperCase()
+}
+
+function departmentIntersects(request, user) {
+  const requestDepartments = collectIdentityNames(
+    request?.dept,
+    request?.departmentName,
+    request?.department_name
+  )
+  const currentDepartments = collectIdentityNames(
+    user?.department,
+    user?.departmentName,
+    user?.department_name
+  )
+
+  return requestDepartments.length > 0 && identityIntersects(requestDepartments, currentDepartments)
+}
+
 function hasPlatformAdminIdentity(user) {
   if (!user) {
     return false
@@ -130,6 +150,25 @@ export function canApproveLeaderExpenseClaims(user) {
   return normalizedRoleCodes(user).some((roleCode) => CLAIM_LEADER_APPROVAL_ROLE_CODES.has(roleCode))
 }
 
+export function canApproveBudgetExpenseApplications(user, request = null) {
+  if (isPlatformAdminUser(user)) {
+    return true
+  }
+
+  const roleCodes = normalizedRoleCodes(user)
+  if (roleCodes.includes('executive')) {
+    return true
+  }
+  if (!roleCodes.includes('budget_monitor')) {
+    return false
+  }
+  if (normalizedGrade(user) !== CLAIM_BUDGET_APPROVAL_GRADE) {
+    return false
+  }
+
+  return request ? departmentIntersects(request, user) : true
+}
+
 export function isCurrentRequestApplicant(request, user) {
   const applicantNames = collectIdentityNames(
     request?.person,
diff --git a/web/src/utils/approvalInbox.js b/web/src/utils/approvalInbox.js
index 290d127..c3a310e 100644
--- a/web/src/utils/approvalInbox.js
+++ b/web/src/utils/approvalInbox.js
@@ -1,5 +1,6 @@
 import { mapExpenseClaimToRequest } from '../composables/useRequests.js'
 import {
+  canApproveBudgetExpenseApplications,
   canApproveLeaderExpenseClaims,
   isCurrentDirectManagerForRequest,
   isCurrentRequestApplicant,
@@ -17,6 +18,10 @@ export function canProcessApprovalRequest(request, currentUser) {
     return true
   }
 
+  if (node.includes('预算')) {
+    return canApproveBudgetExpenseApplications(currentUser, request)
+  }
+
   const isLeaderApprovalNode = (
     node.includes('直属领导')
     || node.includes('领导审批')
diff --git a/web/src/utils/detailAlerts.js b/web/src/utils/detailAlerts.js
index eacf026..ffae4cb 100644
--- a/web/src/utils/detailAlerts.js
+++ b/web/src/utils/detailAlerts.js
@@ -115,10 +115,69 @@ export function hasPendingInfo(request) {
   return false
 }
 
-function resolveDetailAlertTone(request) {
-  if (request?.approvalKey === 'completed') return 'success'
-  if (request?.approvalKey === 'rejected') return 'danger'
-  return 'warning'
+function getRiskFlags(request) {
+  if (Array.isArray(request?.riskFlags)) {
+    return request.riskFlags
+  }
+  if (Array.isArray(request?.risk_flags_json)) {
+    return request.risk_flags_json
+  }
+  return []
+}
+
+function parseNonNegativeInteger(value) {
+  const nextValue = Number(value)
+  return Number.isFinite(nextValue) && nextValue > 0 ? Math.floor(nextValue) : 0
+}
+
+function resolveSlaReminderCount(request) {
+  const directCount = [
+    request?.slaReminderCount,
+    request?.sla_reminder_count,
+    request?.slaUrgeCount,
+    request?.sla_urge_count,
+    request?.urgeCount,
+    request?.urge_count,
+    request?.reminderCount,
+    request?.reminder_count
+  ].reduce((max, value) => Math.max(max, parseNonNegativeInteger(value)), 0)
+
+  if (directCount > 0) {
+    return directCount
+  }
+
+  return getRiskFlags(request).reduce((count, flag) => {
+    if (!flag || typeof flag !== 'object') {
+      return count
+    }
+
+    const explicitCount = [
+      flag.slaReminderCount,
+      flag.sla_reminder_count,
+      flag.slaUrgeCount,
+      flag.sla_urge_count,
+      flag.urgeCount,
+      flag.urge_count,
+      flag.reminderCount,
+      flag.reminder_count
+    ].reduce((max, value) => Math.max(max, parseNonNegativeInteger(value)), 0)
+
+    if (explicitCount > 0) {
+      return count + explicitCount
+    }
+
+    const signal = [
+      flag.source,
+      flag.event_type,
+      flag.eventType,
+      flag.action,
+      flag.type,
+      flag.label,
+      flag.message
+    ].join(' ')
+
+    return /sla|remind|reminder|urge|催单/i.test(signal) ? count + 1 : count
+  }, 0)
 }
 
 export function buildDetailAlerts(request) {
@@ -127,11 +186,13 @@ export function buildDetailAlerts(request) {
   }
 
   const alerts = []
-  const nodeLabel = String(request.node || request.approval || '').trim()
+  const slaReminderCount = resolveSlaReminderCount(request)
 
-  if (nodeLabel) {
-    alerts.push({ label: nodeLabel, tone: resolveDetailAlertTone(request) })
-  }
+  alerts.push({
+    label: `SLA 催单次数 ${slaReminderCount}`,
+    tone: slaReminderCount > 0 ? 'warning' : 'neutral',
+    icon: 'mdi mdi-bell-ring-outline'
+  })
 
   if (hasMissingAttachment(request)) {
     alerts.push({ label: '缺少票据', tone: 'warning' })
diff --git a/web/src/views/DocumentsCenterView.vue b/web/src/views/DocumentsCenterView.vue
index 891965d..6444e23 100644
--- a/web/src/views/DocumentsCenterView.vue
+++ b/web/src/views/DocumentsCenterView.vue
@@ -171,6 +171,7 @@
             <col v-if="showStayTimeColumn" class="col-stay">
             <col class="col-doc-type">
             <col class="col-scene">
+            <col class="col-initiator">
             <col class="col-title">
             <col class="col-amount">
             <col class="col-node">
@@ -184,6 +185,7 @@
               <th v-if="showStayTimeColumn">停留时间</th>
               <th>单据类型</th>
               <th>费用场景</th>
+              <th>发起人</th>
               <th>事项</th>
               <th>金额</th>
               <th>当前环节</th>
@@ -201,6 +203,7 @@
               <td v-if="showStayTimeColumn">{{ row.stayTimeDisplay }}</td>
               <td><span class="doc-kind-tag" :class="row.documentTypeCode">{{ row.documentTypeLabel }}</span></td>
               <td><span class="type-tag" :class="row.typeTone">{{ row.typeLabel }}</span></td>
+              <td>{{ row.initiatorName }}</td>
               <td>{{ row.reason }}</td>
               <td>{{ row.amountDisplay }}</td>
               <td>{{ row.node }}</td>
@@ -437,6 +440,7 @@ const filteredRows = computed(() => {
       row.documentNo,
       row.documentTypeLabel,
       row.typeLabel,
+      row.initiatorName,
       row.reason,
       row.node,
       row.statusLabel
@@ -538,6 +542,16 @@ function buildDocumentRow(request, options = {}) {
   const documentTypeLabel =
     normalized.documentTypeLabel
     || (documentTypeCode === DOCUMENT_TYPE_APPLICATION ? '申请单' : '报销单')
+  const initiatorName = String(
+    normalized.person
+    || normalized.employeeName
+    || normalized.profileName
+    || normalized.applicant
+    || request?.employee_name
+    || request?.employeeName
+    || request?.person
+    || ''
+  ).trim() || '待补充'
 
   return {
     ...normalized,
@@ -547,6 +561,7 @@ function buildDocumentRow(request, options = {}) {
     documentTypeLabel,
     claimId,
     documentNo,
+    initiatorName,
     node: archived ? resolveArchivedDocumentNode(normalized, documentTypeCode) : (normalized.node || normalized.workflowNode || '待提交'),
     statusGroup,
     statusLabel,
diff --git a/web/src/views/LogsView.vue b/web/src/views/LogsView.vue
index ff540a0..1189a8a 100644
--- a/web/src/views/LogsView.vue
+++ b/web/src/views/LogsView.vue
@@ -31,7 +31,6 @@
               v-model="systemLevelFilter"
               :options="systemLevelFilterOptions"
               placeholder="全部"
-              size="small"
             />
           </label>
 
@@ -41,7 +40,6 @@
               v-model="systemEventTypeFilter"
               :options="systemEventTypeFilterOptions"
               placeholder="全部"
-              size="small"
             />
           </label>
 
diff --git a/web/src/views/PoliciesView.vue b/web/src/views/PoliciesView.vue
index d57d665..b231641 100644
--- a/web/src/views/PoliciesView.vue
+++ b/web/src/views/PoliciesView.vue
@@ -8,10 +8,21 @@
               <h2>文档库 / 文件夹</h2>
               <p>默认展示文件列表，点击具体文件后以弹窗方式展开预览。</p>
             </div>
-            <label class="file-search">
-              <i class="mdi mdi-magnify"></i>
-              <input v-model="documentSearch" type="search" placeholder="搜索当前文件夹内文件" />
-            </label>
+            <div class="panel-tools">
+              <label class="file-search">
+                <i class="mdi mdi-magnify"></i>
+                <input v-model="documentSearch" type="search" placeholder="搜索当前文件夹内文件" />
+              </label>
+              <button
+                class="knowledge-sync-btn"
+                type="button"
+                :disabled="!canTriggerKnowledgeSync"
+                @click="handleKnowledgeSync"
+              >
+                <i :class="syncingFolder ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-book-sync-outline'"></i>
+                <span>{{ knowledgeSyncButtonLabel }}</span>
+              </button>
+            </div>
           </header>
 
           <div class="library-body">
@@ -30,19 +41,7 @@
                 </button>
               </nav>
 
-              <div class="folder-sync-block">
-                <button
-                  class="new-folder-btn fixed knowledge-sync-btn"
-                  type="button"
-                  :disabled="!canTriggerKnowledgeSync"
-                  @click="handleKnowledgeSync"
-                >
-                  <i :class="syncingFolder ? 'mdi mdi-loading mdi-spin' : 'mdi mdi-book-sync-outline'"></i>
-                  <span>{{ knowledgeSyncButtonLabel }}</span>
-                </button>
-                <p class="folder-sync-meta">{{ knowledgeSyncHint }}</p>
-              </div>
-            </aside>
+            </aside>
 
             <section class="document-area" :class="{ 'read-only': !isAdmin }">
               <div
diff --git a/web/src/views/TravelRequestDetailView.vue b/web/src/views/TravelRequestDetailView.vue
index 67b8dfb..066eaeb 100644
--- a/web/src/views/TravelRequestDetailView.vue
+++ b/web/src/views/TravelRequestDetailView.vue
@@ -177,6 +177,11 @@
                 </div>
               </div>
 
+              <TravelRequestBudgetAnalysis
+                v-if="showBudgetAnalysis"
+                :claim-id="request.claimId"
+              />
+
               <div v-if="showApplicationLeaderOpinion" class="application-leader-opinion">
                 <div class="application-leader-opinion-head">
                   <span><i class="mdi mdi-account-tie-outline"></i>领导意见</span>
@@ -760,10 +765,6 @@
       :confirm-text="approveConfirmText"
       :busy-text="approveBusyText"
       :busy="approveBusy"
-      :document-no="request.documentNo || request.id"
-      :node="request.node"
-      :summary-label="approvalConfirmSummaryLabel"
-      :next-stage="approvalNextStage"
       :opinion-title="approvalOpinionTitle"
       v-model:opinion="leaderOpinion"
       :opinion-placeholder="approvalOpinionPlaceholder"
diff --git a/web/src/views/scripts/EmployeeManagementView.js b/web/src/views/scripts/EmployeeManagementView.js
index 98bb6d6..d119995 100644
--- a/web/src/views/scripts/EmployeeManagementView.js
+++ b/web/src/views/scripts/EmployeeManagementView.js
@@ -372,7 +372,7 @@ function matchKeyword(employee, keyword) {
     return true
   }
 
-  const haystack = [
+  const fields = [
     employee.name,
     employee.employeeNo,
     employee.department,
@@ -380,9 +380,13 @@ function matchKeyword(employee, keyword) {
     employee.email,
     employee.manager,
     employee.financeOwner,
-    employee.syncState,
-    ...(employee.roles || [])
+    employee.syncState
   ]
+
+  const roles = Array.isArray(employee.roles) ? employee.roles : []
+
+  const haystack = [...fields, ...roles]
+    .map((val) => String(val || '').trim())
     .filter(Boolean)
     .join(' ')
     .toLowerCase()
diff --git a/web/src/views/scripts/PoliciesView.js b/web/src/views/scripts/PoliciesView.js
index 874c787..09540e0 100644
--- a/web/src/views/scripts/PoliciesView.js
+++ b/web/src/views/scripts/PoliciesView.js
@@ -167,31 +167,15 @@ export default {
       }
       return stats
     })
-    const knowledgeSyncButtonLabel = computed(() => {
-      if (syncingFolder.value || activeFolderIngestStats.value.syncing > 0) {
-        return '归纳中...'
-      }
-      return '知识归纳'
-    })
-    const knowledgeSyncHint = computed(() => {
-      const stats = activeFolderIngestStats.value
-      if (!activeFolder.value) {
-        return '请选择一个固定知识目录后再触发归纳。'
-      }
-      if (!stats.total) {
-        return '当前目录暂无文档，上传后即可进行知识归纳。'
-      }
-      if (stats.syncing > 0) {
-        return `当前目录有 ${stats.syncing} 份文档正在归纳，完成后会自动刷新状态。`
-      }
-      if (stats.pending > 0 || stats.failed > 0) {
-        return `当前目录待归纳 ${stats.pending} 份，需重试 ${stats.failed} 份。`
-      }
-      return `当前目录 ${stats.ingested} 份文档已归纳，可手动触发一次增量检查。`
-    })
-    const canTriggerKnowledgeSync = computed(
-      () =>
-        isAdmin.value
+    const knowledgeSyncButtonLabel = computed(() => {
+      if (syncingFolder.value || activeFolderIngestStats.value.syncing > 0) {
+        return '归纳中...'
+      }
+      return '知识归纳'
+    })
+    const canTriggerKnowledgeSync = computed(
+      () =>
+        isAdmin.value
         && Boolean(activeFolder.value)
         && activeFolderIngestStats.value.total > 0
         && !syncingFolder.value
@@ -445,11 +429,11 @@ export default {
 
       syncingFolder.value = true
       try {
-        const payload = await syncKnowledgeLibrary({
-          folder: activeFolder.value,
-          documentIds: [],
-          force: false
-        })
+        const payload = await syncKnowledgeLibrary({
+          folder: activeFolder.value,
+          documentIds: [],
+          force: true
+        })
 
         const queuedIds = Array.isArray(payload?.document_ids) ? payload.document_ids : []
         for (const documentId of queuedIds) {
@@ -461,8 +445,9 @@ export default {
           })
         }
 
-        await loadLibrary({ preserveSelection: true })
-        toast(payload?.summary || '\u77e5\u8bc6\u5f52\u7eb3\u4efb\u52a1\u5df2\u63d0\u4ea4\u3002')
+        await loadLibrary({ preserveSelection: true })
+        const runHint = payload?.agent_run_id ? `日志编号：${payload.agent_run_id}` : ''
+        toast([payload?.summary || '知识归纳任务已提交。', runHint].filter(Boolean).join(' '))
       } catch (error) {
         await loadLibrary({ preserveSelection: true })
         toast(error.message || '\u77e5\u8bc6\u5f52\u7eb3\u89e6\u53d1\u5931\u8d25\u3002')
@@ -647,9 +632,8 @@ export default {
       handleFileInput,
       handleKnowledgeSync,
       isAdmin,
-      knowledgeSyncButtonLabel,
-      knowledgeSyncHint,
-      loading,
+      knowledgeSyncButtonLabel,
+      loading,
       pageSize,
       pageSizeOptions,
       pageSizes,
diff --git a/web/src/views/scripts/TravelRequestDetailView.js b/web/src/views/scripts/TravelRequestDetailView.js
index d5f0a71..417e997 100644
--- a/web/src/views/scripts/TravelRequestDetailView.js
+++ b/web/src/views/scripts/TravelRequestDetailView.js
@@ -5,6 +5,7 @@ import { useToast } from '../../composables/useToast.js'
 import EnterpriseSelect from '../../components/shared/EnterpriseSelect.vue'
 import ConfirmDialog from '../../components/shared/ConfirmDialog.vue'
 import TravelRequestApprovalDialog from '../../components/travel/TravelRequestApprovalDialog.vue'
+import TravelRequestBudgetAnalysis from '../../components/travel/TravelRequestBudgetAnalysis.vue'
 import TravelRequestDeleteDialog from '../../components/travel/TravelRequestDeleteDialog.vue'
 import TravelRequestReturnDialog from '../../components/travel/TravelRequestReturnDialog.vue'
 import {
@@ -22,6 +23,7 @@ import {
   updateExpenseClaimItem
 } from '../../services/reimbursements.js'
 import {
+  canApproveBudgetExpenseApplications,
   canApproveLeaderExpenseClaims,
   canDeleteArchivedExpenseClaims,
   canManageExpenseClaims,
@@ -369,6 +371,7 @@ export default {
     ConfirmDialog,
     EnterpriseSelect,
     TravelRequestApprovalDialog,
+    TravelRequestBudgetAnalysis,
     TravelRequestDeleteDialog,
     TravelRequestReturnDialog
   },
@@ -490,6 +493,10 @@ export default {
       const node = String(request.value.node || request.value.approvalStage || '').trim()
       return node === '财务审批'
     })
+    const isBudgetApprovalStage = computed(() => {
+      const node = String(request.value.node || request.value.approvalStage || '').trim()
+      return node === '预算管理者审批'
+    })
     const isCurrentApplicant = computed(() => isCurrentRequestApplicant(request.value, currentUser.value))
     const isCurrentDirectManagerApprover = computed(() => (
       canApproveLeaderExpenseClaims(currentUser.value)
@@ -501,6 +508,18 @@ export default {
       && isFinanceUser(currentUser.value)
       && !isCurrentApplicant.value
     ))
+    const canProcessBudgetApprovalStage = computed(() => (
+      isApplicationDocument.value
+      && isBudgetApprovalStage.value
+      && canApproveBudgetExpenseApplications(currentUser.value, request.value)
+      && !isCurrentApplicant.value
+    ))
+    const showBudgetAnalysis = computed(() => (
+      isApplicationDocument.value
+      && isBudgetApprovalStage.value
+      && canApproveBudgetExpenseApplications(currentUser.value, request.value)
+      && !isCurrentApplicant.value
+    ))
     const canReturnRequest = computed(() => {
       if (request.value.approvalKey !== 'in_progress' || !request.value.claimId || !canReturnExpenseClaims(currentUser.value)) {
         return false
@@ -508,6 +527,9 @@ export default {
       if (isDirectManagerApprovalStage.value) {
         return isCurrentDirectManagerApprover.value
       }
+      if (isBudgetApprovalStage.value) {
+        return canProcessBudgetApprovalStage.value
+      }
       return canProcessFinanceApprovalStage.value
     })
     const canApproveRequest = computed(() =>
@@ -520,6 +542,7 @@ export default {
           && isCurrentDirectManagerApprover.value
         )
         || canProcessFinanceApprovalStage.value
+        || canProcessBudgetApprovalStage.value
       )
     )
     const leaderApprovalInfo = computed(() => buildLeaderApprovalInfo(request.value))
@@ -536,39 +559,43 @@ export default {
       isApplicationDocument.value
       && hasLeaderApprovalEvents.value
     ))
-    const requiresApprovalOpinion = computed(() => isDirectManagerApprovalStage.value)
-    const approvalOpinionTitle = computed(() => (isFinanceApprovalStage.value ? '财务意见' : '领导意见'))
+    const requiresApprovalOpinion = computed(() => false)
+    const approvalOpinionTitle = computed(() => (isFinanceApprovalStage.value ? '财务意见' : '附加意见'))
     const approvalOpinionPlaceholder = computed(() => {
       if (isFinanceApprovalStage.value) {
         return '请输入财务终审意见，可补充票据核验、金额一致性或入账关注点。'
       }
       if (isApplicationDocument.value) {
-        return '请输入审批意见，可补充业务必要性、预算合理性或执行要求。'
+        return '可选填审批补充说明，例如业务必要性、预算合理性或执行要求；不填写默认为同意。'
       }
-      return '请输入审批意见，可补充核实情况、费用合理性或后续财务关注点。'
+      return '可选填审批补充说明，例如核实情况、费用合理性或后续财务关注点；不填写默认为同意。'
     })
     const approvalOpinionHint = computed(() => {
       if (isFinanceApprovalStage.value) {
         return '审核通过后将进入归档入账。'
       }
-      return isApplicationDocument.value ? '领导意见为必填，确认后会生成报销草稿。' : '领导意见为必填，审批通过后将流转至财务审批。'
+      if (isBudgetApprovalStage.value) {
+        return '不填写附加意见则默认同意，确认后会归档申请单并生成报销草稿。'
+      }
+      return isApplicationDocument.value ? '不填写附加意见则默认同意，确认后会流转至预算管理者审批。' : '不填写附加意见则默认同意，审批通过后将流转至财务审批。'
+    })
+    const approvalConfirmBadge = computed(() => {
+      if (isFinanceApprovalStage.value) {
+        return '财务终审'
+      }
+      return isBudgetApprovalStage.value ? '预算审核' : '领导审批'
     })
-    const approvalConfirmBadge = computed(() => (isFinanceApprovalStage.value ? '财务终审' : '领导审批'))
     const approvalConfirmDescription = computed(() => {
       if (isFinanceApprovalStage.value) {
         return '确认后该报销单会完成财务终审并进入归档入账，请确认票据、金额与财务意见无误。'
       }
       if (isApplicationDocument.value) {
-        return '确认后该申请单会完成直属领导审批，并自动进入申请人的报销草稿中。'
+        return isBudgetApprovalStage.value
+          ? '确认后该申请单会完成预算审核，归档申请单，并自动进入申请人的报销草稿中。'
+          : '确认后该申请单会完成直属领导审批，并流转给预算管理者进一步审核。'
       }
       return '确认后该报销单会从直属领导审批流转到财务审批，请确认申请信息与领导意见无误。'
     })
-    const approvalNextStage = computed(() => {
-      if (isFinanceApprovalStage.value) {
-        return '归档入账'
-      }
-      return isApplicationDocument.value ? '报销草稿' : '财务审批'
-    })
     const approveActionLabel = computed(() => (isApplicationDocument.value ? '确认审核' : '审批通过'))
     const approveBusyLabel = computed(() => (isApplicationDocument.value ? '确认中' : '通过中'))
     const approveConfirmTitle = computed(() => (
@@ -581,15 +608,14 @@ export default {
         ? '退回后该申请单会回到待提交状态，申请人需要调整后重新提交。'
         : '退回后该单据会回到待提交状态，申请人需要调整后重新提交并再次经过 AI 预审。'
     ))
-    const approvalConfirmSummaryLabel = computed(() => (
-      isApplicationDocument.value ? '生成结果' : '下一节点'
-    ))
     const approvalSuccessToast = computed(() => {
       if (isFinanceApprovalStage.value) {
         return `${request.value.id} 已完成财务终审，进入归档入账。`
       }
       return isApplicationDocument.value
-        ? `${request.value.id} 已确认审核，正在生成报销草稿。`
+        ? isBudgetApprovalStage.value
+          ? `${request.value.id} 已完成预算审核，正在生成报销草稿。`
+          : `${request.value.id} 已确认审核，已流转至预算管理者审批。`
         : `${request.value.id} 已审批通过，流转至财务审批。`
     })
     const deleteActionLabel = computed(() => (isDraftRequest.value ? '删除草稿' : '删除单据'))
@@ -1751,15 +1777,10 @@ export default {
         return
       }
 
-      if (requiresApprovalOpinion.value && !leaderOpinion.value.trim()) {
-        toast('请先填写领导意见，填写后才能确认审核。')
-        return
-      }
-
       approveBusy.value = true
       try {
         const responsePayload = await approveExpenseClaim(request.value.claimId, {
-          opinion: leaderOpinion.value.trim()
+          opinion: leaderOpinion.value.trim() || '同意'
         })
         const generatedDraftClaimNo = resolveGeneratedDraftClaimNo(responsePayload)
         approveConfirmDialogOpen.value = false
@@ -1805,7 +1826,7 @@ export default {
       emit, actionBusy, aiAdvice, aiAdviceHint, aiAdviceTitle, attachmentPreviewError, attachmentPreviewIndexLabel,
       attachmentPreviewLoading, attachmentPreviewMediaType, attachmentPreviewName, attachmentPreviewOpen,
       attachmentPreviewUrl, approveBusy, approveConfirmDialogOpen, approvalConfirmBadge,
-      approvalConfirmDescription, approvalConfirmSummaryLabel, approvalNextStage, approvalOpinionHint,
+      approvalConfirmDescription, approvalOpinionHint,
       approvalOpinionPlaceholder, approvalOpinionTitle, approveActionLabel, approveBusyLabel,
       applicationDetailFactItems,
       approveBusyText, approveConfirmText, approveConfirmTitle, canDeleteRequest, canManageCurrentClaim,
@@ -1836,6 +1857,7 @@ export default {
       requiresApprovalOpinion,
       riskOverrideReasons, saveDetailNote, savingDetailNote, savingExpenseId,
       showAiAdvicePanel, showApplicationLeaderOpinion,
+      showBudgetAnalysis,
       showExpenseRisk, startExpenseEdit, submitBusy, submitConfirmDialogOpen,
       submitRiskWarnings,
       triggerExpenseUpload, uploadedExpenseCount, uploadingExpenseId, saveExpenseEdit
diff --git a/web/src/views/scripts/auditViewDigitalEmployeeModel.js b/web/src/views/scripts/auditViewDigitalEmployeeModel.js
new file mode 100644
index 0000000..3249fa5
--- /dev/null
+++ b/web/src/views/scripts/auditViewDigitalEmployeeModel.js
@@ -0,0 +1,263 @@
+const DIGITAL_EMPLOYEE_AGENT = 'hermes'
+
+const TASK_TYPE_LABELS = {
+  daily_risk_scan: '每日风险巡检',
+  global_risk_scan: '全局风险巡检',
+  weekly_ar_summary: '周度应收账龄汇总',
+  weekly_expense_report: '周度费用洞察',
+  rule_review_digest: '规则待审摘要',
+  knowledge_index_sync: '知识库归集',
+  x_financial_callback: '任务回调上报'
+}
+
+const CONTENT_LABELS = {
+  task_type: '技能类型',
+  schedule: '执行计划',
+  cron: '调度表达式',
+  folder: '归集范围',
+  changed_only: '仅处理变更',
+  force: '强制重建',
+  index_engine: '索引引擎',
+  callback_type: '回调类型',
+  status: '回写状态',
+  summary: '结果摘要'
+}
+
+const HIDDEN_CONTENT_KEYS = new Set([
+  'agent',
+  'target_agent',
+  'callback_token',
+  'token',
+  'api_key',
+  'authorization'
+])
+
+export function normalizeDigitalEmployeeText(value) {
+  return String(value ?? '').trim()
+}
+
+export function sanitizeDigitalEmployeeText(value, fallback = '') {
+  const text = normalizeDigitalEmployeeText(value)
+    .replace(/hermes/gi, '数字员工')
+    .replace(/赫尔墨斯/g, '数字员工')
+    .replace(/\s+/g, ' ')
+    .trim()
+  return text || fallback
+}
+
+export function sanitizeDigitalEmployeeName(value, fallback = '数字员工技能') {
+  const text = sanitizeDigitalEmployeeText(value, fallback)
+    .replace(/^数字员工[\s·:：-]*/i, '')
+    .trim()
+  return text || fallback
+}
+
+export function parseDigitalEmployeeContent(value) {
+  if (!value) {
+    return {}
+  }
+  if (typeof value === 'object' && !Array.isArray(value)) {
+    return value
+  }
+  if (typeof value !== 'string') {
+    return {}
+  }
+  try {
+    const parsed = JSON.parse(value)
+    return parsed && typeof parsed === 'object' && !Array.isArray(parsed) ? parsed : {}
+  } catch {
+    return {}
+  }
+}
+
+export function resolveDigitalEmployeeTaskType(source = {}, content = {}) {
+  const config = source.config_json || source.configJson || {}
+  const raw =
+    normalizeDigitalEmployeeText(content.task_type) ||
+    normalizeDigitalEmployeeText(config.task_type) ||
+    normalizeDigitalEmployeeText(source.task_type) ||
+    normalizeDigitalEmployeeText(source.code).replace(/^task\.hermes\./i, '')
+  return raw.replace(/[-.]/g, '_')
+}
+
+export function isDigitalEmployeeAsset(source = {}) {
+  const config = source.config_json || source.configJson || {}
+  const haystack = [
+    source.asset_type,
+    source.code,
+    source.name,
+    source.description,
+    config.agent,
+    config.target_agent,
+    config.worker,
+    config.runtime_agent
+  ]
+    .map((item) => normalizeDigitalEmployeeText(item).toLowerCase())
+    .filter(Boolean)
+    .join(' ')
+
+  return (
+    normalizeDigitalEmployeeText(source.asset_type) === 'task' &&
+    (haystack.includes(DIGITAL_EMPLOYEE_AGENT) || haystack.includes('task.hermes.'))
+  )
+}
+
+export function formatDigitalEmployeeCron(value) {
+  const raw = normalizeDigitalEmployeeText(value)
+  if (!raw) {
+    return '手动触发'
+  }
+
+  const parts = raw.split(/\s+/)
+  if (parts.length < 5) {
+    return sanitizeDigitalEmployeeText(raw)
+  }
+
+  const [minute, hour, dayOfMonth, month, dayOfWeek] = parts
+  const hourNumber = Number(hour)
+  const minuteNumber = Number(minute)
+  const timeLabel =
+    Number.isFinite(hourNumber) && Number.isFinite(minuteNumber)
+      ? `${String(hourNumber).padStart(2, '0')}:${String(minuteNumber).padStart(2, '0')}`
+      : `${hour}:${minute}`
+
+  if (dayOfMonth === '*' && month === '*' && dayOfWeek === '*') {
+    return `每天 ${timeLabel}`
+  }
+
+  if (dayOfMonth === '*' && month === '*' && dayOfWeek !== '*') {
+    const weekdayLabels = {
+      '0': '周日',
+      '1': '周一',
+      '2': '周二',
+      '3': '周三',
+      '4': '周四',
+      '5': '周五',
+      '6': '周六',
+      '7': '周日'
+    }
+    return `每${weekdayLabels[dayOfWeek] || `周${dayOfWeek}`} ${timeLabel}`
+  }
+
+  return sanitizeDigitalEmployeeText(raw)
+}
+
+export function resolveDigitalEmployeeSchedule(source = {}, content = {}) {
+  const config = source.config_json || source.configJson || {}
+  const raw =
+    normalizeDigitalEmployeeText(content.schedule) ||
+    normalizeDigitalEmployeeText(config.cron) ||
+    normalizeDigitalEmployeeText(config.schedule) ||
+    normalizeDigitalEmployeeText(config.cron_expression)
+  return {
+    value: raw,
+    label: formatDigitalEmployeeCron(raw)
+  }
+}
+
+export function resolveDigitalEmployeeEnabled(source = {}) {
+  const config = source.config_json || source.configJson || {}
+  if (config.enabled === false || config.is_enabled === false) {
+    return false
+  }
+  if (source.enabled === false || source.is_enabled === false) {
+    return false
+  }
+  return normalizeDigitalEmployeeText(source.status || 'active') === 'active'
+}
+
+export function resolveDigitalEmployeeDisplayCode(source = {}, content = {}) {
+  const taskType = resolveDigitalEmployeeTaskType(source, content)
+  return taskType ? `digital.${taskType}` : 'digital.skill'
+}
+
+function formatDigitalEmployeeValue(value) {
+  if (typeof value === 'boolean') {
+    return value ? '是' : '否'
+  }
+  if (Array.isArray(value)) {
+    return value.map((item) => sanitizeDigitalEmployeeText(item)).filter(Boolean).join('、') || '-'
+  }
+  if (value && typeof value === 'object') {
+    return sanitizeDigitalEmployeeText(JSON.stringify(value, null, 2))
+  }
+  return sanitizeDigitalEmployeeText(value, '-')
+}
+
+export function buildDigitalEmployeeContentRows(content = {}) {
+  return Object.entries(content)
+    .filter(([key]) => !HIDDEN_CONTENT_KEYS.has(normalizeDigitalEmployeeText(key).toLowerCase()))
+    .map(([key, value]) => ({
+      key,
+      label: CONTENT_LABELS[key] || key,
+      value: formatDigitalEmployeeValue(value)
+    }))
+}
+
+export function buildDigitalEmployeeContentPreview(content = {}) {
+  const visiblePayload = {}
+  for (const [key, value] of Object.entries(content)) {
+    if (HIDDEN_CONTENT_KEYS.has(normalizeDigitalEmployeeText(key).toLowerCase())) {
+      continue
+    }
+    visiblePayload[key] = value
+  }
+  return sanitizeDigitalEmployeeText(JSON.stringify(visiblePayload, null, 2))
+}
+
+export function buildDigitalEmployeeListMeta(source = {}) {
+  const content = parseDigitalEmployeeContent(source.current_version_content)
+  const taskType = resolveDigitalEmployeeTaskType(source, content)
+  const schedule = resolveDigitalEmployeeSchedule(source, content)
+  const enabled = resolveDigitalEmployeeEnabled(source)
+  const fallbackName = TASK_TYPE_LABELS[taskType] || '数字员工技能'
+
+  return {
+    name: sanitizeDigitalEmployeeName(source.name, fallbackName),
+    code: resolveDigitalEmployeeDisplayCode(source, content),
+    summary: sanitizeDigitalEmployeeText(source.description, '面向后台自动执行的数字员工技能。'),
+    category: '数字员工',
+    owner: sanitizeDigitalEmployeeText(source.owner, '平台运营'),
+    reviewer: sanitizeDigitalEmployeeText(source.reviewer, '系统'),
+    scope: schedule.label,
+    scheduleLabel: schedule.label,
+    executionMode: schedule.value ? '定时执行' : '手动触发',
+    enabled,
+    enabledLabel: enabled ? '已启动' : '未启动',
+    enabledTone: enabled ? 'success' : 'disabled',
+    taskType
+  }
+}
+
+export function buildDigitalEmployeeDetailMeta(source = {}) {
+  const content = parseDigitalEmployeeContent(source.current_version_content)
+  const listMeta = buildDigitalEmployeeListMeta({
+    ...source,
+    current_version_content: content
+  })
+  const schedule = resolveDigitalEmployeeSchedule(source, content)
+  const contentRows = buildDigitalEmployeeContentRows(content)
+
+  return {
+    ...listMeta,
+    rawCode: normalizeDigitalEmployeeText(source.code),
+    description: sanitizeDigitalEmployeeText(
+      source.description,
+      '该技能由后台数字员工按计划执行，并把结果沉淀到对应业务资产或运行日志中。'
+    ),
+    contentRows,
+    contentPreview: buildDigitalEmployeeContentPreview(content),
+    scheduleRows: [
+      { label: '执行计划', value: schedule.label },
+      { label: '调度表达式', value: schedule.value || '手动触发' },
+      { label: '启动状态', value: listMeta.enabledLabel, tone: listMeta.enabledTone },
+      { label: '执行方式', value: listMeta.executionMode }
+    ],
+    overviewRows: [
+      { label: '能力编号', value: listMeta.code },
+      { label: '业务归口', value: listMeta.owner },
+      { label: '当前版本', value: source.working_version || source.current_version || '-' },
+      { label: '最近更新', value: source.updated_at || '-' }
+    ]
+  }
+}
diff --git a/web/src/views/scripts/auditViewMetadata.js b/web/src/views/scripts/auditViewMetadata.js
index ea08d5f..d611114 100644
--- a/web/src/views/scripts/auditViewMetadata.js
+++ b/web/src/views/scripts/auditViewMetadata.js
@@ -96,6 +96,32 @@ export const TAB_META = {
     ...TYPE_META.mcp,
     typeKey: 'mcp',
     badgeTone: 'amber'
+  },
+  digitalWorkers: {
+    assetType: 'task',
+    typeKey: 'digitalWorkers',
+    label: '数字员工',
+    typeLabel: '数字员工',
+    createButtonLabel: '数字员工已接入',
+    hintText: '归集后台自动执行的数字员工技能，可查看技能内容、执行计划、启动状态和最近版本。',
+    searchPlaceholder: '搜索数字员工技能、编号、执行计划或维护人',
+    showMetricColumn: true,
+    showRuntimeColumn: true,
+    showVersionColumn: true,
+    showStatusColumn: true,
+    showEnabledColumn: true,
+    tableColumns: {
+      name: '技能名称',
+      category: '归集标签',
+      owner: '维护归口',
+      scope: '执行计划',
+      runtime: '触发方式',
+      version: '当前版本',
+      status: '资产状态',
+      metric: '运行方式',
+      updatedAt: '最近更新'
+    },
+    badgeTone: 'violet'
   }
 }
 
@@ -208,6 +234,24 @@ export const DETAIL_TITLES = {
     historyDesc: '最近版本记录',
     publishTitle: '服务状态',
     publishDesc: 'MCP 资产已接入规则中心，但真实外部调用仍以后续链路集成为准。'
+  },
+  digitalWorkers: {
+    configTitle: '技能档案',
+    configDesc: '展示数字员工技能的编号、归口、执行计划和启停状态。',
+    detailTitle: '技能内容',
+    detailDesc: '展示当前版本记录的任务类型、调度范围和执行参数。',
+    outputTitle: '执行安排',
+    outputDesc: '展示什么时候执行、是否启动，以及当前运行方式。',
+    ruleListTitle: '技能参数',
+    checkListTitle: '启动状态',
+    triggerTitle: '执行计划',
+    triggerDesc: '当前技能的计划执行时间或触发方式。',
+    toolTitle: '运行归口',
+    toolDesc: '数字员工技能由后台调度执行，运行结果进入对应日志或业务资产。',
+    historyTitle: '版本记录',
+    historyDesc: '最近的技能配置快照。',
+    publishTitle: '启动状态',
+    publishDesc: '数字员工技能由资产状态和调度配置共同决定是否启动。'
   }
 }
 
diff --git a/web/src/views/scripts/auditViewModel.js b/web/src/views/scripts/auditViewModel.js
index 1e1f3fa..17463c8 100644
--- a/web/src/views/scripts/auditViewModel.js
+++ b/web/src/views/scripts/auditViewModel.js
@@ -34,6 +34,13 @@ import {
   resolveRiskRuleSeverity,
   resolveRiskRuleSeverityLabel
 } from './auditViewRiskRuleModel.js'
+import {
+  buildDigitalEmployeeContentRows,
+  buildDigitalEmployeeDetailMeta,
+  buildDigitalEmployeeListMeta,
+  isDigitalEmployeeAsset,
+  sanitizeDigitalEmployeeText
+} from './auditViewDigitalEmployeeModel.js'
 
 const EXPENSE_TYPE_SCENARIO_LABELS = {
   travel: '差旅费',
@@ -335,6 +342,9 @@ export function resolveTabId(source, typeKey) {
   if (typeKey === 'rules') {
     return resolveRuleTabId(source)
   }
+  if (typeKey === 'digitalWorkers') {
+    return isDigitalEmployeeAsset(source) ? 'digitalWorkers' : ''
+  }
   return typeKey
 }
 
@@ -895,6 +905,9 @@ export function resolveTypeKey(assetType) {
   if (assetType === 'mcp') {
     return 'mcp'
   }
+  if (assetType === 'task') {
+    return 'digitalWorkers'
+  }
   return ''
 }
 
@@ -958,6 +971,9 @@ export function buildRowRuntime(asset, typeKey) {
   if (typeKey === 'mcp') {
     return normalizeText(asset.config_json?.endpoint) || '未配置地址'
   }
+  if (typeKey === 'digitalWorkers') {
+    return buildDigitalEmployeeListMeta(asset).executionMode
+  }
   return ''
 }
 
@@ -971,6 +987,9 @@ export function buildRowMetric(asset, typeKey) {
   if (typeKey === 'mcp') {
     return asset.config_json?.timeout_ms ? `${asset.config_json.timeout_ms} ms` : '未配置超时'
   }
+  if (typeKey === 'digitalWorkers') {
+    return buildDigitalEmployeeListMeta(asset).executionMode
+  }
   return ''
 }
 
@@ -1042,6 +1061,19 @@ export function buildListItem(asset) {
       ? resolveRiskRuleScoreLabel(asset.config_json, asset.config_json) || resolveRiskRuleSeverityLabel(asset.config_json)
       : resolveRiskRuleSeverityLabel(asset.config_json)
     : ''
+  const digitalMeta = typeKey === 'digitalWorkers' ? buildDigitalEmployeeListMeta(asset) : null
+  const displayName = digitalMeta?.name || asset.name
+  const displayCode = digitalMeta?.code || asset.code
+  const displaySummary = digitalMeta?.summary || listSubtitle
+  const displayOwner = digitalMeta?.owner || (isRiskRule ? creator : asset.owner)
+  const displayReviewer = digitalMeta?.reviewer || reviewer
+  const displayCategory = digitalMeta?.category || resolveDomainLabel(asset.domain)
+  const displayScope =
+    digitalMeta?.scope ||
+    (typeKey === 'rules' ? ruleScenarioCategory || '閫氱敤' : formatScenarioList(asset.scenario_json))
+  const displayEnabledValue = digitalMeta ? digitalMeta.enabled : isEnabledValue
+  const displayEnabledLabel = digitalMeta?.enabledLabel || (isEnabledValue ? '鏄? : '鍚?)
+  const displayEnabledTone = digitalMeta?.enabledTone || (isEnabledValue ? 'success' : 'disabled')
 
   return {
     id: asset.id,
@@ -1052,15 +1084,17 @@ export function buildListItem(asset) {
     usesJsonRiskRule,
     ruleDocument,
     typeLabel: tabMeta.typeLabel,
-    short: makeShort(asset.name),
-    name: asset.name,
-    code: asset.code,
-    summary: listSubtitle,
-    listSubtitle,
-    category: resolveDomainLabel(asset.domain),
-    owner: isRiskRule ? creator : asset.owner,
-    reviewer,
+    short: makeShort(displayName),
+    name: displayName,
+    code: displayCode,
+    rawCode: asset.code,
+    summary: displaySummary,
+    listSubtitle: displaySummary,
+    category: displayCategory,
+    owner: displayOwner,
+    reviewer: displayReviewer,
     scope: typeKey === 'rules' ? ruleScenarioCategory || '通用' : formatScenarioList(asset.scenario_json),
+    scope: displayScope,
     riskCategory: ruleScenarioCategory,
     scenarioList: ruleScenarioList,
     businessStageValue: businessStage.value,
@@ -1086,6 +1120,9 @@ export function buildListItem(asset) {
     isEnabledValue,
     isEnabledLabel: isEnabledValue ? '是' : '否',
     isEnabledTone: isEnabledValue ? 'success' : 'disabled',
+    isEnabledValue: displayEnabledValue,
+    isEnabledLabel: displayEnabledLabel,
+    isEnabledTone: displayEnabledTone,
     modifiedBy,
     changeCount,
     updatedAt: isRiskRule ? riskRuleCreatedAt : formatDateTime(asset.updated_at),
@@ -1417,6 +1454,25 @@ export function buildDetailViewModel(detail, runs) {
   const initialRiskRuleScore = resolveRiskRuleScore(configJson, configJson)
   const initialRiskRuleScoreLevel = resolveRiskRuleScoreLevel(configJson, configJson)
   const initialRiskRuleSeverity = initialRiskRuleScoreLevel || resolveRiskRuleSeverity(configJson)
+  const digitalMeta = typeKey === 'digitalWorkers'
+    ? buildDigitalEmployeeDetailMeta({
+        ...detail,
+        updated_at: formatDateTime(detail.updated_at)
+      })
+    : null
+  const detailName = digitalMeta?.name || detail.name
+  const detailCode = digitalMeta?.code || detail.code
+  const detailSummary = digitalMeta?.description ||
+    (usesJsonRiskRule ? buildRiskListSubtitle(detail.description) : detail.description)
+  const detailOwner = digitalMeta?.owner || detail.owner
+  const detailReviewer = digitalMeta?.reviewer || detail.reviewer || detail.latest_review?.reviewer || '寰呭垎閰?
+  const detailCategory = digitalMeta?.category || resolveDomainLabel(detail.domain)
+  const detailScope =
+    digitalMeta?.scope ||
+    (typeKey === 'rules' ? ruleScenarioCategory || '閫氱敤' : formatScenarioList(detail.scenario_json))
+  const detailEnabledValue = digitalMeta ? digitalMeta.enabled : isEnabledValue
+  const detailEnabledLabel = digitalMeta?.enabledLabel || (isEnabledValue ? '鏄? : '鍚?)
+  const detailEnabledTone = digitalMeta?.enabledTone || (isEnabledValue ? 'success' : 'disabled')
 
   return {
     id: detail.id,
diff --git a/web/tests/accessControl.test.mjs b/web/tests/accessControl.test.mjs
index 7769b0a..9479727 100644
--- a/web/tests/accessControl.test.mjs
+++ b/web/tests/accessControl.test.mjs
@@ -2,6 +2,7 @@ import assert from 'node:assert/strict'
 import test from 'node:test'
 
 import {
+  canApproveBudgetExpenseApplications,
   canApproveLeaderExpenseClaims,
   canAccessAppView,
   canDeleteArchivedExpenseClaims,
@@ -22,6 +23,24 @@ test('direct approvers can return claims without receiving delete permissions',
   assert.equal(canReturnExpenseClaims(approverUser), true)
   assert.equal(canApproveLeaderExpenseClaims(managerUser), true)
   assert.equal(canApproveLeaderExpenseClaims(approverUser), true)
+  assert.equal(canApproveBudgetExpenseApplications({ roleCodes: ['budget_monitor'], grade: 'P6' }), false)
+  assert.equal(canApproveBudgetExpenseApplications({ roleCodes: ['budget_monitor'], grade: 'P8' }), true)
+  assert.equal(
+    canApproveBudgetExpenseApplications(
+      { roleCodes: ['budget_monitor'], grade: 'P8', departmentName: '交付部' },
+      { departmentName: '交付部' }
+    ),
+    true
+  )
+  assert.equal(
+    canApproveBudgetExpenseApplications(
+      { roleCodes: ['budget_monitor'], grade: 'P8', departmentName: '财务部' },
+      { departmentName: '交付部' }
+    ),
+    false
+  )
+  assert.equal(canApproveBudgetExpenseApplications({ roleCodes: [], grade: 'P8' }), false)
+  assert.equal(canApproveBudgetExpenseApplications({ roleCodes: ['executive'] }), true)
   assert.equal(canManageExpenseClaims(managerUser), false)
   assert.equal(canManageExpenseClaims(approverUser), false)
 })
@@ -81,6 +100,37 @@ test('finance approval inbox only processes finance-stage requests', () => {
   )
 })
 
+test('budget approval inbox only processes budget-stage requests for budget monitor or senior finance roles', () => {
+  const budgetUser = { roleCodes: ['budget_monitor'], grade: 'P8', name: '赵预算', departmentName: '交付部' }
+  const otherDepartmentBudgetUser = { roleCodes: ['budget_monitor'], grade: 'P8', name: '王预算', departmentName: '财务部' }
+  const seniorFinanceUser = { roleCodes: ['executive'], grade: 'P7', name: '高级财务' }
+  const p8WithoutBudgetRole = { roleCodes: ['manager'], grade: 'P8', name: '高职级经理' }
+
+  assert.equal(
+    canProcessApprovalRequest({ workflowNode: '预算管理者审批', person: '张三', departmentName: '交付部' }, budgetUser),
+    true
+  )
+  assert.equal(
+    canProcessApprovalRequest({ workflowNode: '预算管理者审批', person: '张三', departmentName: '交付部' }, seniorFinanceUser),
+    true
+  )
+  assert.equal(
+    canProcessApprovalRequest(
+      { workflowNode: '预算管理者审批', person: '张三', departmentName: '交付部' },
+      otherDepartmentBudgetUser
+    ),
+    false
+  )
+  assert.equal(
+    canProcessApprovalRequest({ workflowNode: '预算管理者审批', person: '张三' }, p8WithoutBudgetRole),
+    false
+  )
+  assert.equal(
+    canProcessApprovalRequest({ workflowNode: '财务审批', person: '张三' }, budgetUser),
+    false
+  )
+})
+
 test('users with both finance and manager roles can process both relevant stages', () => {
   const financeManagerUser = { roleCodes: ['finance', 'manager'], name: '李经理' }
 
diff --git a/web/tests/app-shell-detail-alerts.test.mjs b/web/tests/app-shell-detail-alerts.test.mjs
index 6b862a3..3153ee2 100644
--- a/web/tests/app-shell-detail-alerts.test.mjs
+++ b/web/tests/app-shell-detail-alerts.test.mjs
@@ -57,7 +57,7 @@ test('detail topbar ignores system allowance rows when checking missing tickets'
 
   assert.equal(hasMissingAttachment(request), false)
   assert.equal(hasPendingInfo(request), false)
-  assert.deepEqual(alerts, ['直属领导审批'])
+  assert.deepEqual(alerts, ['SLA 催单次数 0'])
 })
 
 test('detail topbar still flags real manual rows without required ticket info', () => {
@@ -96,7 +96,7 @@ test('detail topbar still flags real manual rows without required ticket info',
 
   assert.equal(hasMissingAttachment(request), true)
   assert.equal(hasPendingInfo(request), true)
-  assert.deepEqual(alerts, ['待提交', '缺少票据', '待补信息'])
+  assert.deepEqual(alerts, ['SLA 催单次数 0', '缺少票据', '待补信息'])
 })
 
 test('application detail topbar does not ask for receipt attachments', () => {
@@ -122,5 +122,29 @@ test('application detail topbar does not ask for receipt attachments', () => {
 
   assert.equal(hasMissingAttachment(request), false)
   assert.equal(alerts.includes('缺少票据'), false)
-  assert.deepEqual(alerts, ['直属领导审批'])
+  assert.deepEqual(alerts, ['SLA 催单次数 0'])
+})
+
+test('detail topbar shows SLA reminder count from direct fields and reminder events', () => {
+  const directAlerts = buildDetailAlerts({
+    node: '直属领导审批',
+    approvalKey: 'in_progress',
+    slaReminderCount: 2,
+    expenseItems: []
+  })
+
+  const eventAlerts = buildDetailAlerts({
+    node: '直属领导审批',
+    approvalKey: 'in_progress',
+    riskFlags: [
+      { source: 'sla_reminder', message: '下属已催单' },
+      { event_type: 'urge', message: '再次催单' }
+    ],
+    expenseItems: []
+  })
+
+  assert.equal(directAlerts[0].label, 'SLA 催单次数 2')
+  assert.equal(directAlerts[0].tone, 'warning')
+  assert.equal(directAlerts[0].icon, 'mdi mdi-bell-ring-outline')
+  assert.equal(eventAlerts[0].label, 'SLA 催单次数 2')
 })
diff --git a/web/tests/documents-center-status-filter.test.mjs b/web/tests/documents-center-status-filter.test.mjs
index 68619b5..4ff0ee7 100644
--- a/web/tests/documents-center-status-filter.test.mjs
+++ b/web/tests/documents-center-status-filter.test.mjs
@@ -85,15 +85,20 @@ test('documents center list shows created time and conditional stay time columns
   assert.match(documentsCenterView, /import \{[\s\S]*formatDocumentListTime[\s\S]*resolveDocumentStayTimeDisplay[\s\S]*\} from '..\/utils\/documentCenterTime\.js'/)
   assert.match(documentsCenterView, /<col class="col-created">/)
   assert.match(documentsCenterView, /<col v-if="showStayTimeColumn" class="col-stay">/)
+  assert.match(documentsCenterView, /<col class="col-initiator">/)
   assert.match(documentsCenterView, /<th>单号<\/th>[\s\S]*<th>创建时间<\/th>[\s\S]*<th v-if="showStayTimeColumn">停留时间<\/th>/)
+  assert.match(documentsCenterView, /<th>费用场景<\/th>[\s\S]*<th>发起人<\/th>[\s\S]*<th>事项<\/th>/)
   assert.match(documentsCenterView, /<td>\{\{ row\.createdAtDisplay \}\}<\/td>/)
   assert.match(documentsCenterView, /<td v-if="showStayTimeColumn">\{\{ row\.stayTimeDisplay \}\}<\/td>/)
+  assert.match(documentsCenterView, /<td>\{\{ row\.initiatorName \}\}<\/td>/)
   assert.match(
     documentsCenterView,
     /const showStayTimeColumn = computed\(\(\) =>[\s\S]*DOCUMENT_SCOPE_APPLICATION[\s\S]*DOCUMENT_SCOPE_REVIEW/
   )
   assert.match(documentsCenterView, /createdAtDisplay: formatDocumentListTime\(createdAtSource\)/)
   assert.match(documentsCenterView, /stayTimeDisplay: resolveDocumentStayTimeDisplay\(normalized\)/)
+  assert.match(documentsCenterView, /initiatorName,/)
+  assert.match(documentsCenterView, /row\.initiatorName/)
 })
 
 test('documents center action buttons are scoped to application and reimbursement tabs', () => {
@@ -225,9 +230,10 @@ test('documents center status dropdown uses compact filter styling', () => {
   assert.match(documentsCenterStyles, /\.documents-list\s*\{[\s\S]*grid-template-rows:\s*auto auto minmax\(0,\s*1fr\) auto;/)
   assert.match(documentsCenterStyles, /\.status-tabs button\s*\{[\s\S]*display:\s*inline-flex;/)
   assert.match(documentsCenterStyles, /\.scope-tab-badge\s*\{[\s\S]*border-radius:\s*999px;/)
-  assert.match(documentsCenterStyles, /min-width:\s*1320px;/)
+  assert.match(documentsCenterStyles, /min-width:\s*1420px;/)
   assert.match(documentsCenterStyles, /\.col-created\s*\{\s*width:\s*10%;\s*\}/)
   assert.match(documentsCenterStyles, /\.col-stay\s*\{\s*width:\s*9%;\s*\}/)
+  assert.match(documentsCenterStyles, /\.col-initiator\s*\{\s*width:\s*8%;\s*\}/)
   assert.match(documentsCenterStyles, /\.document-status-filter\s*\{[\s\S]*display:\s*inline-flex;/)
   assert.match(documentsCenterStyles, /\.document-status-filter\s*\{[\s\S]*min-height:\s*38px;/)
   assert.match(documentsCenterStyles, /\.status-filter-trigger\s*\{[\s\S]*min-width:\s*154px;/)
diff --git a/web/tests/requestProgressSteps.test.mjs b/web/tests/requestProgressSteps.test.mjs
index 62fe3f6..d2c91ab 100644
--- a/web/tests/requestProgressSteps.test.mjs
+++ b/web/tests/requestProgressSteps.test.mjs
@@ -5,10 +5,12 @@ import { mapExpenseClaimToRequest } from '../src/composables/useRequests.js'
 
 const CREATE_APPLICATION = '\u521b\u5efa\u7533\u8bf7'
 const DIRECT_MANAGER_APPROVAL = '\u76f4\u5c5e\u9886\u5bfc\u5ba1\u6279'
+const BUDGET_MANAGER_APPROVAL = '\u9884\u7b97\u7ba1\u7406\u8005\u5ba1\u6279'
 const APPROVAL_COMPLETED = '\u5ba1\u6279\u5b8c\u6210'
 const RETURNED = '\u9000\u56de'
 const WAIT_SUBMIT = '\u5f85\u63d0\u4ea4'
 const WAIT_LEADER_LI_APPROVAL = '\u7b49\u5f85 Leader Li \u6279\u590d'
+const WAIT_BUDGET_ZHAO_APPROVAL = '\u7b49\u5f85 \u8d75\u9884\u7b97 \u6279\u590d'
 const LEADER_RETURNED_STATUS = '\u9886\u5bfc\u5df2\u9000\u56de\uff0c\u5f85\u91cd\u65b0\u63d0\u4ea4'
 
 test('application claims are mapped as application documents', () => {
@@ -41,7 +43,7 @@ test('application claims are mapped as application documents', () => {
   assert.equal(request.expenseTableSummary, '预计金额已随申请提交')
   assert.deepEqual(
     request.progressSteps.map((step) => step.label),
-    [CREATE_APPLICATION, WAIT_LEADER_LI_APPROVAL, APPROVAL_COMPLETED]
+    [CREATE_APPLICATION, WAIT_LEADER_LI_APPROVAL, BUDGET_MANAGER_APPROVAL, APPROVAL_COMPLETED]
   )
   assert.equal(request.progressSteps.some((step) => step.label === 'AI预审'), false)
   assert.equal(request.progressSteps.some((step) => step.label === '财务审批'), false)
@@ -50,6 +52,47 @@ test('application claims are mapped as application documents', () => {
   assert.equal(request.progressSteps.find((step) => step.label === WAIT_LEADER_LI_APPROVAL)?.current, true)
 })
 
+test('application claims wait for department P8 budget monitor after leader approval', () => {
+  const request = mapExpenseClaimToRequest({
+    id: 'claim-application-budget',
+    claim_no: 'AP-20260525103145-BUDGET',
+    employee_name: '张三',
+    department_name: '交付部',
+    manager_name: 'Leader Li',
+    expense_type: 'travel_application',
+    reason: '支撑国网服务器上线部署',
+    location: '上海',
+    amount: 12000,
+    invoice_count: 0,
+    occurred_at: '2026-05-25T00:00:00.000Z',
+    submitted_at: '2026-05-25T02:00:00.000Z',
+    created_at: '2026-05-25T01:30:00.000Z',
+    updated_at: '2026-05-25T03:00:00.000Z',
+    status: 'submitted',
+    approval_stage: BUDGET_MANAGER_APPROVAL,
+    risk_flags_json: [
+      {
+        source: 'manual_approval',
+        event_type: 'expense_application_approval',
+        operator: 'Leader Li',
+        previous_approval_stage: DIRECT_MANAGER_APPROVAL,
+        next_approval_stage: BUDGET_MANAGER_APPROVAL,
+        next_approver_name: '赵预算',
+        next_approver_grade: 'P8',
+        created_at: '2026-05-25T03:00:00.000Z'
+      }
+    ],
+    items: []
+  })
+
+  assert.deepEqual(
+    request.progressSteps.map((step) => step.label),
+    [CREATE_APPLICATION, DIRECT_MANAGER_APPROVAL, WAIT_BUDGET_ZHAO_APPROVAL, APPROVAL_COMPLETED]
+  )
+  assert.equal(request.progressSteps.find((step) => step.label === WAIT_BUDGET_ZHAO_APPROVAL)?.current, true)
+  assert.equal(request.progressSteps.find((step) => step.label === DIRECT_MANAGER_APPROVAL)?.time, 'Leader Li通过')
+})
+
 test('returned application claims include leader return node and supplement status', () => {
   const request = mapExpenseClaimToRequest({
     id: 'claim-application-returned',
@@ -86,7 +129,7 @@ test('returned application claims include leader return node and supplement stat
 
   assert.deepEqual(
     request.progressSteps.map((step) => step.label),
-    [CREATE_APPLICATION, WAIT_LEADER_LI_APPROVAL, RETURNED, WAIT_SUBMIT]
+    [CREATE_APPLICATION, DIRECT_MANAGER_APPROVAL, RETURNED, WAIT_SUBMIT]
   )
   assert.equal(request.progressSteps.find((step) => step.label === RETURNED)?.time, 'Leader Li\u9000\u56de')
   assert.match(request.progressSteps.find((step) => step.label === RETURNED)?.detail, /2026-05-25/)
@@ -96,7 +139,7 @@ test('returned application claims include leader return node and supplement stat
   assert.equal(request.progressSteps.some((step) => step.label === APPROVAL_COMPLETED), false)
 })
 
-test('approved application claims complete after direct manager approval only', () => {
+test('approved application claims complete after budget approval', () => {
   const request = mapExpenseClaimToRequest({
     id: 'claim-application-approved',
     claim_no: 'AP-20260525113045-HGFEDCBA',
@@ -120,6 +163,16 @@ test('approved application claims complete after direct manager approval only',
         event_type: 'expense_application_approval',
         operator: '李经理',
         previous_approval_stage: '直属领导审批',
+        next_approval_stage: '预算管理者审批',
+        next_approver_name: '赵预算',
+        next_approver_grade: 'P8',
+        created_at: '2026-05-25T03:00:00.000Z'
+      },
+      {
+        source: 'budget_approval',
+        event_type: 'expense_application_budget_approval',
+        operator: '赵预算',
+        previous_approval_stage: '预算管理者审批',
         next_approval_stage: '审批完成',
         created_at: '2026-05-25T03:00:00.000Z'
       }
@@ -131,10 +184,11 @@ test('approved application claims complete after direct manager approval only',
   assert.equal(request.workflowNode, '审批完成')
   assert.deepEqual(
     request.progressSteps.map((step) => step.label),
-    ['创建申请', '直属领导审批', '审批完成']
+    ['创建申请', '直属领导审批', '预算管理者审批', '审批完成']
   )
   assert.equal(request.progressSteps.every((step) => step.done), true)
   assert.equal(request.progressSteps.find((step) => step.label === '直属领导审批')?.time, '李经理通过')
+  assert.equal(request.progressSteps.find((step) => step.label === '预算管理者审批')?.time, '赵预算通过')
 })
 
 test('progress steps show approval operator time and current stay duration', () => {
diff --git a/web/tests/travel-request-detail-leader-approval.test.mjs b/web/tests/travel-request-detail-leader-approval.test.mjs
index aa0e502..10197e9 100644
--- a/web/tests/travel-request-detail-leader-approval.test.mjs
+++ b/web/tests/travel-request-detail-leader-approval.test.mjs
@@ -19,6 +19,10 @@ const approvalDialog = readFileSync(
   fileURLToPath(new URL('../src/components/travel/TravelRequestApprovalDialog.vue', import.meta.url)),
   'utf8'
 )
+const budgetAnalysisComponent = readFileSync(
+  fileURLToPath(new URL('../src/components/travel/TravelRequestBudgetAnalysis.vue', import.meta.url)),
+  'utf8'
+)
 const reimbursementService = readFileSync(
   fileURLToPath(new URL('../src/services/reimbursements.js', import.meta.url)),
   'utf8'
@@ -53,18 +57,23 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
   assert.match(detailScript, /const approveConfirmDialogOpen = ref\(false\)/)
   assert.match(detailScript, /const canApproveRequest = computed/)
   assert.match(detailScript, /canApproveLeaderExpenseClaims/)
+  assert.match(detailScript, /canApproveBudgetExpenseApplications/)
   assert.match(detailScript, /isCurrentDirectManagerForRequest/)
   assert.match(detailScript, /isCurrentRequestApplicant/)
   assert.match(detailScript, /isFinanceApprovalStage/)
+  assert.match(detailScript, /const isBudgetApprovalStage = computed/)
+  assert.match(detailScript, /const showBudgetAnalysis = computed/)
   assert.match(detailScript, /const isCurrentApplicant = computed/)
   assert.match(detailScript, /const isCurrentDirectManagerApprover = computed/)
   assert.match(detailScript, /const canProcessFinanceApprovalStage = computed/)
+  assert.match(detailScript, /const canProcessBudgetApprovalStage = computed/)
   assert.match(detailScript, /approvalOpinionTitle/)
   assert.match(detailScript, /approvalConfirmDescription/)
-  assert.match(detailScript, /approvalNextStage/)
+  assert.doesNotMatch(detailScript, /approvalNextStage/)
   assert.doesNotMatch(detailScript, /showApplicationLeaderOpinionInput/)
   assert.doesNotMatch(detailScript, /showLeaderApprovalPanel/)
-  assert.match(detailScript, /const requiresApprovalOpinion = computed\(\(\) => isDirectManagerApprovalStage\.value\)/)
+  assert.match(detailScript, /const requiresApprovalOpinion = computed\(\(\) => false\)/)
+  assert.match(detailScript, /approvalOpinionTitle = computed\(\(\) => \(isFinanceApprovalStage\.value \? '财务意见' : '附加意见'\)\)/)
   assert.match(detailScript, /buildLeaderApprovalEvents/)
   assert.match(detailScript, /buildLeaderApprovalInfo/)
   assert.match(detailScript, /const leaderApprovalEvents = computed/)
@@ -76,11 +85,13 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
   assert.match(detailScript, /isDirectManagerApprovalStage\.value\)[\s\S]*return isCurrentDirectManagerApprover\.value/)
   assert.match(detailScript, /isDirectManagerApprovalStage\.value[\s\S]*&& isCurrentDirectManagerApprover\.value/)
   assert.match(detailScript, /canProcessFinanceApprovalStage\.value/)
+  assert.match(detailScript, /canProcessBudgetApprovalStage\.value/)
   assert.doesNotMatch(detailScript, /leaderApprovalReadonlyText/)
   assert.match(detailScript, /resolveGeneratedDraftClaimNo/)
   assert.match(detailScript, /approveActionLabel/)
-  assert.match(detailScript, /approveExpenseClaim\(request\.value\.claimId, \{[\s\S]*opinion: leaderOpinion\.value\.trim\(\)/)
+  assert.match(detailScript, /approveExpenseClaim\(request\.value\.claimId, \{[\s\S]*opinion: leaderOpinion\.value\.trim\(\) \|\| '同意'/)
   assert.match(detailScript, /报销草稿 \$\{generatedDraftClaimNo\} 已生成/)
+  assert.match(detailScript, /流转至预算管理者审批/)
 
   assert.doesNotMatch(detailTemplate, /v-if="showLeaderApprovalPanel"/)
   assert.doesNotMatch(detailTemplate, /showApplicationLeaderOpinionInput/)
@@ -96,6 +107,7 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
   assert.doesNotMatch(detailTemplate, /leaderApprovalReadonlyText/)
   assert.doesNotMatch(detailTemplate, /\u5f85\u76f4\u5c5e\u9886\u5bfc\u586b\u5199\u5ba1\u6279\u610f\u89c1/)
   assert.match(detailTemplate, /领导意见/)
+  assert.match(detailTemplate, /<TravelRequestBudgetAnalysis[\s\S]*v-if="showBudgetAnalysis"[\s\S]*:claim-id="request\.claimId"/)
   assert.match(approvalDialog, /\{\{ opinionTitle \}\}/)
   assert.doesNotMatch(detailTemplate, /v-model="leaderOpinion"/)
   assert.match(detailTemplate, /@click="handleApproveRequest"/)
@@ -105,7 +117,10 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
   assert.match(detailTemplate, /:description="approvalConfirmDescription"/)
   assert.match(detailTemplate, /:confirm-text="approveConfirmText"/)
   assert.match(detailTemplate, /:busy-text="approveBusyText"/)
-  assert.match(detailTemplate, /:next-stage="approvalNextStage"/)
+  assert.doesNotMatch(detailTemplate, /:next-stage="approvalNextStage"/)
+  assert.doesNotMatch(approvalDialog, /submit-confirm-summary/)
+  assert.doesNotMatch(approvalDialog, /单据编号/)
+  assert.doesNotMatch(approvalDialog, /当前节点/)
   assert.match(detailTemplate, /v-model:opinion="leaderOpinion"/)
   assert.match(detailTemplate, /:opinion-placeholder="approvalOpinionPlaceholder"/)
   assert.match(detailTemplate, /:opinion-hint="approvalOpinionHint"/)
@@ -119,8 +134,8 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
   assert.doesNotMatch(handleApproveRequest, /approveExpenseClaim/)
   assert.doesNotMatch(handleApproveRequest, /leaderOpinion\.value\.trim/)
   assert.match(confirmApproveRequest, /approveExpenseClaim/)
-  assert.match(confirmApproveRequest, /requiresApprovalOpinion\.value && !leaderOpinion\.value\.trim\(\)/)
-  assert.match(confirmApproveRequest, /请先填写领导意见，填写后才能确认审核。/)
+  assert.doesNotMatch(confirmApproveRequest, /requiresApprovalOpinion\.value && !leaderOpinion\.value\.trim\(\)/)
+  assert.doesNotMatch(confirmApproveRequest, /请先填写领导意见，填写后才能确认审核。/)
 
   assert.match(approvalDialog, /<textarea/)
   assert.match(approvalDialog, /update:opinion/)
@@ -141,4 +156,11 @@ test('approval-mode detail collects leader opinion inside confirm dialog before
 
   assert.match(reimbursementService, /export function approveExpenseClaim\(claimId, payload = \{\}\)/)
   assert.match(reimbursementService, /\/approve/)
+  assert.match(reimbursementService, /export function fetchExpenseClaimBudgetAnalysis/)
+  assert.match(reimbursementService, /\/budget-analysis/)
+  assert.match(budgetAnalysisComponent, /预算分析/)
+  assert.match(budgetAnalysisComponent, /当前预算额度/)
+  assert.match(budgetAnalysisComponent, /此次费用占预算/)
+  assert.match(budgetAnalysisComponent, /综合评分/)
+  assert.match(budgetAnalysisComponent, /fetchExpenseClaimBudgetAnalysis/)
 })
diff --git a/web/tests/travel-request-detail-risk-advice.test.mjs b/web/tests/travel-request-detail-risk-advice.test.mjs
index cdd5665..9087d47 100644
--- a/web/tests/travel-request-detail-risk-advice.test.mjs
+++ b/web/tests/travel-request-detail-risk-advice.test.mjs
@@ -684,9 +684,12 @@ test('return reason dialog is wired into approval and detail return actions', ()
   assert.match(returnReasonDialog, /application_budget_basis_missing/)
   assert.match(returnReasonDialog, /application_policy_mismatch/)
   assert.match(returnReasonDialog, /application_attachment_needed/)
+  assert.match(returnReasonDialog, /application_other/)
   assert.match(returnReasonDialog, /退单选项/)
   assert.match(returnReasonDialog, /selectionError/)
-  assert.match(returnReasonDialog, /selectedCodes\.value\.length === 0/)
+  assert.match(returnReasonDialog, /selectedApplicationCode/)
+  assert.match(returnReasonDialog, /application \? 'radio' : 'checkbox'/)
+  assert.match(returnReasonDialog, /selectedReasonCodes\.value\.length === 0/)
   assert.match(returnReasonDialog, /lastAutoReason/)
   assert.match(returnReasonDialog, /reason_codes/)
   assert.match(approvalCenterTemplate, /<TravelRequestDetailView/)