feat(p47): 6 頁深挖資料庫 — 從 5 表 → 17 表，每頁加 3-5 個 widget

統帥質疑：「6 頁內容太空洞，要更貼近資料庫裡所有數據」
盤點：DB 有 22 表，6 頁原本只用 5 表（22.7% 利用率）。
本 commit 新接 12 張既有但未用的表。

K-1 host_health（接 +5 表 → 8 表）
- 加 incidents 最近 10 筆詳細（task_name/error_type/嚴重度/狀態/重試/錯誤訊息）
- 加 heal_logs 最近 10 筆（action_type/result/耗時/incident 關聯）
- 加 playbooks 庫排行 TOP 12（success_count/fail_count/成功率/啟用狀態）
- 加 backup_log 7 日歷史（type/status/size/duration/error）
- 加 embedding_retry_queue pending/failed 警示

K-2 ai_calls（接 +3 表）
- 加 24h 每小時呼叫趨勢 bucket（含成本+錯誤+流量分布條）
- 加 by model 細分（不只 provider，到 model 版本級別）TOP 15
- 加 agent_context 最近 10 筆（OpenClaw/Hermes 對話 session preview）

K-3 budget（接 +3 表）
- 加當月 Top 5 燒錢呼叫端（caller × cost ranking）
- 加過去 30 日每日成本 by provider 趨勢表
- 加 ai_price_recommendations 7 日統計（strategy 分布 + 平均信心度）

K-4 promotion_review（接 +2 表）
- 加蒸餾池 30 日 status 分布（不只 awaiting，看 8 種完整流動）
- 加 ai_insights 最近 10 筆已晉升內容預覽
- 加 agent_strategy_weights TOP 12（OpenClaw 學習權重 + 成功率）

K-5 quality_trend（接 +3 表）
- 加 RAG 整體 feedback 1-5 分分布（過去 N 日，星等視覺）
- 加 action_plans status 分布（pending/approved/executed/rejected）
- 加 action_outcomes verdict 分布（effective/neutral/backfired
  — ADR-012 閉環學習核心 KPI）

K-6 ppt_audit_history（接 +0 表，但 deeper SQL）
- 加 30 日統計卡（total/passed/failed/error/通過率/總 issue 數/平均信心度）
- 加 Top 10 反覆失敗檔案（30d 失敗次數 + total issues）
- 加 empty state 說明（PPT_VISION_ENABLED=false 時顯示啟用步驟）

DB 利用率對應：
- Phase 38 起點：5 表（22.7%）
- Phase 39-46 累計：12 表（54.5%）
- Phase 47 收官：17 表（77.3%）
  新接：playbooks / backup_log / embedding_retry_queue / agent_context /
        ai_price_recommendations / agent_strategy_weights / action_plans /
        action_outcomes（之前已接：incidents/heal_logs/ai_insights/learning_episodes/
        rag_query_log/mcp_calls/ai_calls/host_health_probes/ppt_audit_results）

每頁 widget 數對應：
- host_health: 5 卡 → 11 卡
- ai_calls: 5 卡 → 8 卡
- budget: 4 卡 → 7 卡
- promotion_review: 2 卡 → 5 卡
- quality_trend: 4 卡 → 7 卡
- ppt_audit: 3 卡 → 6 卡

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

routes/admin_observability_routes.py

@@ -580,6 +580,47 @@ def ai_calls_dashboard():
             {'since': since},
         ).fetchall()
 
+        # 5b. Phase 47 K-2: by model 細分（不只 provider，到實際 model）
+        by_model = session.execute(
+            sa_text("""
+                SELECT model, provider, COUNT(*) AS calls,
+                       COALESCE(SUM(input_tokens + output_tokens), 0) AS tokens,
+                       COALESCE(SUM(cost_usd), 0) AS cost,
+                       COALESCE(AVG(duration_ms), 0) AS avg_ms,
+                       COUNT(*) FILTER (WHERE status NOT IN ('ok','cache_only')) AS errors
+                FROM ai_calls
+                WHERE called_at >= :since
+                  AND model IS NOT NULL AND model != ''
+                GROUP BY model, provider
+                ORDER BY calls DESC
+                LIMIT 15
+            """),
+            {'since': since},
+        ).fetchall()
+
+        # 5c. Phase 47 K-2: hourly 呼叫量趨勢（24 個 bucket）
+        hourly_trend = session.execute(
+            sa_text("""
+                SELECT date_trunc('hour', called_at) AS hr,
+                       COUNT(*) AS calls,
+                       COALESCE(SUM(cost_usd), 0) AS cost,
+                       COUNT(*) FILTER (WHERE status NOT IN ('ok','cache_only')) AS errors
+                FROM ai_calls
+                WHERE called_at >= NOW() - INTERVAL '24 hours'
+                GROUP BY hr ORDER BY hr ASC
+            """),
+        ).fetchall()
+
+        # 5d. Phase 47 K-2: agent_context 最近 10 筆（OpenClaw/Hermes 對話上下文）
+        recent_contexts = session.execute(
+            sa_text("""
+                SELECT created_at, agent_name, context_key, ttl_minutes,
+                       LEFT(context_val, 120) AS preview
+                FROM agent_context
+                ORDER BY created_at DESC LIMIT 10
+            """),
+        ).fetchall()
+
         # 5. Phase 39 D-3: caller × RAG 命中率 × MCP 編排率（跨表 JOIN）
         # 展現「AI 自動化專業」核心：每個 caller 多大比例走了 RAG / MCP
         caller_richness = session.execute(
@@ -638,6 +679,33 @@ def ai_calls_dashboard():
                 for r in recent
             ],
             callers=[r[0] for r in callers],
+            by_model=[
+                {
+                    'model': r[0], 'provider': r[1],
+                    'calls': int(r[2] or 0), 'tokens': int(r[3] or 0),
+                    'cost': float(r[4] or 0), 'avg_ms': int(r[5] or 0),
+                    'errors': int(r[6] or 0),
+                }
+                for r in by_model
+            ],
+            hourly_trend=[
+                {
+                    'hour': r[0].strftime('%H:%M') if r[0] else '',
+                    'calls': int(r[1] or 0),
+                    'cost': float(r[2] or 0),
+                    'errors': int(r[3] or 0),
+                }
+                for r in hourly_trend
+            ],
+            recent_contexts=[
+                {
+                    'created_at': r[0].strftime('%Y-%m-%d %H:%M') if r[0] else '',
+                    'agent_name': r[1], 'context_key': r[2],
+                    'ttl_minutes': int(r[3] or 0),
+                    'preview': r[4] or '',
+                }
+                for r in recent_contexts
+            ],
             caller_richness=[
                 {
                     'caller': r[0],
@@ -660,6 +728,7 @@ def ai_calls_dashboard():
             hours=hours, caller_filter=caller_filter,
             provider_filter=provider_filter,
             summary={}, by_provider=[], recent=[], callers=[], caller_richness=[],
+            by_model=[], hourly_trend=[], recent_contexts=[],
             error=f'查詢失敗: {type(e).__name__}: {str(e)[:200]}',
         )
     finally:
@@ -740,11 +809,63 @@ def promotion_review_list():
         except Exception:
             pass  # rag_service import 失敗（feature flag OFF）→ 略過
 
+        # Phase 47 K-4: 蒸餾池 status 分布（30d）
+        ep_distribution = session.execute(
+            sa_text("""
+                SELECT promotion_status, COUNT(*) AS cnt
+                FROM learning_episodes
+                WHERE created_at >= NOW() - INTERVAL '30 days'
+                GROUP BY promotion_status ORDER BY cnt DESC
+            """),
+        ).fetchall()
+        episode_distribution_30d = {r[0]: int(r[1] or 0) for r in ep_distribution}
+
+        # Phase 47 K-4: ai_insights 最近 10 筆已晉升（type/created_at 視覺）
+        latest_insights = session.execute(
+            sa_text("""
+                SELECT id, insight_type, period, product_sku, created_at,
+                       LEFT(content, 160) AS preview
+                FROM ai_insights
+                ORDER BY created_at DESC LIMIT 10
+            """),
+        ).fetchall()
+
+        # Phase 47 K-4: agent_strategy_weights TOP 12（OpenClaw 學習權重）
+        strategy_weights = session.execute(
+            sa_text("""
+                SELECT strategy_key, weight, success_cnt, fail_cnt, updated_at
+                FROM agent_strategy_weights
+                ORDER BY (success_cnt + fail_cnt) DESC
+                LIMIT 12
+            """),
+        ).fetchall()
+
         return render_template(
             'admin/promotion_review.html',
             active_page='obs_promotion_review',
             episodes=episodes,
             kb_size=kb_size,
+            episode_distribution_30d=episode_distribution_30d,
+            latest_insights=[
+                {
+                    'id': r[0], 'insight_type': r[1], 'period': r[2],
+                    'product_sku': r[3],
+                    'created_at': r[4].strftime('%Y-%m-%d %H:%M') if r[4] else '',
+                    'preview': r[5] or '',
+                }
+                for r in latest_insights
+            ],
+            strategy_weights=[
+                {
+                    'strategy_key': r[0], 'weight': float(r[1] or 0),
+                    'success': int(r[2] or 0), 'fail': int(r[3] or 0),
+                    'updated_at': r[4].strftime('%Y-%m-%d') if r[4] else '',
+                    'success_rate': (
+                        float(r[2] or 0) / float((r[2] or 0) + (r[3] or 0)) * 100
+                    ) if ((r[2] or 0) + (r[3] or 0)) > 0 else 0,
+                }
+                for r in strategy_weights
+            ],
             error=None,
         )
     except Exception as e:
@@ -753,6 +874,9 @@ def promotion_review_list():
             active_page='obs_promotion_review',
             episodes=[],
             kb_size=0,
+            episode_distribution_30d={},
+            latest_insights=[],
+            strategy_weights=[],
             error=f'查詢失敗: {type(e).__name__}: {str(e)[:200]}',
         )
     finally:
@@ -875,6 +999,54 @@ def quality_trend_dashboard():
         except Exception:
             pass
 
+        # Phase 47 K-5: action_outcomes verdict 統計（ADR-012 閉環學習結果）
+        action_outcomes_stats = []
+        action_plans_status = []
+        rag_overall_dist = []
+        try:
+            session = get_session()
+            try:
+                # action_outcomes verdict 分布（30d）
+                ao_rows = session.execute(
+                    sa_text(f"""
+                        SELECT verdict, COUNT(*) AS cnt
+                        FROM action_outcomes
+                        WHERE created_at >= NOW() - INTERVAL '{int(days)} days'
+                        GROUP BY verdict ORDER BY cnt DESC
+                    """),
+                ).fetchall()
+                action_outcomes_stats = [{'verdict': r[0] or 'unknown', 'count': int(r[1] or 0)} for r in ao_rows]
+
+                # action_plans status 分布（30d）
+                ap_rows = session.execute(
+                    sa_text(f"""
+                        SELECT status, plan_type, COUNT(*) AS cnt
+                        FROM action_plans
+                        WHERE created_at >= NOW() - INTERVAL '{int(days)} days'
+                        GROUP BY status, plan_type ORDER BY cnt DESC
+                    """),
+                ).fetchall()
+                action_plans_status = [
+                    {'status': r[0], 'plan_type': r[1] or 'misc', 'count': int(r[2] or 0)}
+                    for r in ap_rows
+                ]
+
+                # rag_query_log 整體 feedback 分布（不只 caller-level，整體）
+                rag_dist_rows = session.execute(
+                    sa_text(f"""
+                        SELECT feedback_score, COUNT(*) AS cnt
+                        FROM rag_query_log
+                        WHERE queried_at >= NOW() - INTERVAL '{int(days)} days'
+                          AND feedback_score IS NOT NULL
+                        GROUP BY feedback_score ORDER BY feedback_score
+                    """),
+                ).fetchall()
+                rag_overall_dist = [{'score': int(r[0] or 0), 'count': int(r[1] or 0)} for r in rag_dist_rows]
+            finally:
+                session.close()
+        except Exception:
+            pass
+
         return render_template(
             'admin/quality_trend.html',
             active_page='obs_quality_trend',
@@ -883,6 +1055,9 @@ def quality_trend_dashboard():
             recommendations=recommendations,
             episode_distribution=episode_distribution,
             rag_root_causes=rag_root_causes,
+            action_outcomes_stats=action_outcomes_stats,
+            action_plans_status=action_plans_status,
+            rag_overall_dist=rag_overall_dist,
             error=None,
         )
     except Exception as e:
@@ -891,6 +1066,7 @@ def quality_trend_dashboard():
             active_page='obs_quality_trend',
             days=days, trends=[], recommendations=[],
             episode_distribution={}, rag_root_causes=[],
+            action_outcomes_stats=[], action_plans_status=[], rag_overall_dist=[],
             error=f'查詢失敗: {type(e).__name__}: {str(e)[:200]}',
         )
 
@@ -950,6 +1126,51 @@ def budget_dashboard():
                 'updated_at': b[5].strftime('%Y-%m-%d %H:%M') if b[5] else '-',
             })
 
+        # Phase 47 K-3: 30d daily cost trend by provider
+        cost_30d = session.execute(
+            sa_text("""
+                SELECT date_trunc('day', called_at)::date AS d,
+                       provider, COALESCE(SUM(cost_usd), 0) AS cost
+                FROM ai_calls
+                WHERE called_at >= NOW() - INTERVAL '30 days'
+                GROUP BY d, provider
+                ORDER BY d DESC, cost DESC
+            """),
+        ).fetchall()
+        cost_trend_30d = []
+        for r in cost_30d:
+            cost_trend_30d.append({
+                'date': r[0].strftime('%m-%d') if r[0] else '',
+                'provider': r[1],
+                'cost': float(r[2] or 0),
+            })
+
+        # Phase 47 K-3: top 5 cost-burning caller (當月)
+        top_cost_callers = session.execute(
+            sa_text("""
+                SELECT caller, COUNT(*) AS calls,
+                       COALESCE(SUM(cost_usd), 0) AS cost,
+                       COALESCE(SUM(input_tokens + output_tokens), 0) AS tokens
+                FROM ai_calls
+                WHERE called_at >= :ms
+                  AND cost_usd > 0
+                GROUP BY caller
+                ORDER BY cost DESC LIMIT 5
+            """),
+            {'ms': month_start},
+        ).fetchall()
+
+        # Phase 47 K-3: ai_price_recommendations 7d 統計
+        price_rec_7d = session.execute(
+            sa_text("""
+                SELECT strategy, COUNT(*) AS cnt,
+                       COALESCE(AVG(confidence), 0) AS avg_conf
+                FROM ai_price_recommendations
+                WHERE created_at >= NOW() - INTERVAL '7 days'
+                GROUP BY strategy ORDER BY cnt DESC
+            """),
+        ).fetchall()
+
         # Phase 39 D-4: RAG 自動建議策略（針對超 80% 的 row）
         budget_strategies = []
         over_threshold_rows = [r for r in rows if r.get('ratio', 0) >= 0.8]
@@ -984,11 +1205,27 @@ def budget_dashboard():
             active_page='obs_budget',
             rows=rows,
             budget_strategies=budget_strategies,
+            cost_trend_30d=cost_trend_30d,
+            top_cost_callers=[
+                {
+                    'caller': r[0], 'calls': int(r[1] or 0),
+                    'cost': float(r[2] or 0), 'tokens': int(r[3] or 0),
+                }
+                for r in top_cost_callers
+            ],
+            price_rec_7d=[
+                {
+                    'strategy': r[0], 'count': int(r[1] or 0),
+                    'avg_confidence': round(float(r[2] or 0), 3),
+                }
+                for r in price_rec_7d
+            ],
             error=None,
         )
     except Exception as e:
         return render_template('admin/budget.html', active_page='obs_budget', rows=[],
-                                budget_strategies=[],
+                                budget_strategies=[], cost_trend_30d=[],
+                                top_cost_callers=[], price_rec_7d=[],
                                 error=f'查詢失敗: {type(e).__name__}: {str(e)[:200]}')
     finally:
         session.close()
@@ -1262,6 +1499,62 @@ def ppt_audit_history():
     except Exception:
         vision_enabled = False
 
+    # Phase 47 K-6: 30d 統計 + top failure files
+    audit_30d_stats = {}
+    top_failure_files = []
+    try:
+        s_ppt = get_session()
+        try:
+            stat_row = s_ppt.execute(
+                sa_text("""
+                    SELECT COUNT(*),
+                           COUNT(*) FILTER (WHERE audit_status = 'passed'),
+                           COUNT(*) FILTER (WHERE audit_status = 'failed'),
+                           COUNT(*) FILTER (WHERE audit_status = 'skipped'),
+                           COUNT(*) FILTER (WHERE audit_status = 'error'),
+                           COALESCE(AVG(confidence) FILTER (WHERE audit_status = 'passed'), 0),
+                           COALESCE(SUM(issues_count), 0)
+                    FROM ppt_audit_results
+                    WHERE audited_at >= NOW() - INTERVAL '30 days'
+                """),
+            ).fetchone()
+            total_30d = int(stat_row[0] or 0)
+            audit_30d_stats = {
+                'total': total_30d,
+                'passed': int(stat_row[1] or 0),
+                'failed': int(stat_row[2] or 0),
+                'skipped': int(stat_row[3] or 0),
+                'error': int(stat_row[4] or 0),
+                'avg_confidence': round(float(stat_row[5] or 0), 3),
+                'total_issues': int(stat_row[6] or 0),
+                'pass_rate': (float(stat_row[1] or 0) / total_30d * 100) if total_30d else 0,
+            }
+
+            top_fail_rows = s_ppt.execute(
+                sa_text("""
+                    SELECT pptx_filename, COUNT(*) AS attempts,
+                           SUM(issues_count) AS total_issues,
+                           MAX(audited_at) AS last_audit
+                    FROM ppt_audit_results
+                    WHERE audit_status IN ('failed', 'error')
+                      AND audited_at >= NOW() - INTERVAL '30 days'
+                    GROUP BY pptx_filename
+                    ORDER BY attempts DESC, total_issues DESC LIMIT 10
+                """),
+            ).fetchall()
+            top_failure_files = [
+                {
+                    'filename': r[0], 'attempts': int(r[1] or 0),
+                    'total_issues': int(r[2] or 0),
+                    'last_audit': r[3].strftime('%Y-%m-%d %H:%M') if r[3] else '',
+                }
+                for r in top_fail_rows
+            ]
+        finally:
+            s_ppt.close()
+    except Exception:
+        pass
+
     # Phase 41 E-2: 對最近 3 筆 failed audit 跑 RAG 找相似修法
     rag_fixes = []
     failed_records = [r for r in audit_records if r.get('audit_status') in ('failed', 'error')][:3]
@@ -1303,6 +1596,8 @@ def ppt_audit_history():
         files=files,
         audit_records=audit_records,
         rag_fixes=rag_fixes,
+        audit_30d_stats=audit_30d_stats,
+        top_failure_files=top_failure_files,
         vision_enabled=vision_enabled,
         error=error,
     )
@@ -1504,6 +1799,115 @@ def host_health_dashboard():
     except Exception:
         pass  # 表可能尚未 migration，失敗安全
 
+    # Phase 47 K-1: incidents + heal_logs 詳細列表 + playbooks 排行 + backup + embed queue
+    recent_incidents = []
+    recent_heals = []
+    playbook_ranking = []
+    backup_history = []
+    embed_queue_pending = 0
+    embed_queue_failed = 0
+    try:
+        s3 = get_session()
+        try:
+            inc_rows = s3.execute(
+                sa_text("""
+                    SELECT id, created_at, task_name, error_type, severity,
+                           status, error_message, retry_count, resolved_at
+                    FROM incidents
+                    ORDER BY created_at DESC LIMIT 10
+                """),
+            ).fetchall()
+            recent_incidents = [
+                {
+                    'id': r[0], 'created_at': r[1].strftime('%Y-%m-%d %H:%M'),
+                    'task_name': r[2], 'error_type': r[3], 'severity': r[4],
+                    'status': r[5], 'error_message': (r[6] or '')[:200],
+                    'retry_count': int(r[7] or 0),
+                    'resolved_at': r[8].strftime('%Y-%m-%d %H:%M') if r[8] else None,
+                }
+                for r in inc_rows
+            ]
+            heal_rows = s3.execute(
+                sa_text("""
+                    SELECT h.id, h.created_at, h.action_type, h.result,
+                           h.duration_ms, h.action_detail, h.incident_id,
+                           i.error_type
+                    FROM heal_logs h
+                    LEFT JOIN incidents i ON i.id = h.incident_id
+                    ORDER BY h.created_at DESC LIMIT 10
+                """),
+            ).fetchall()
+            recent_heals = [
+                {
+                    'id': r[0], 'created_at': r[1].strftime('%Y-%m-%d %H:%M'),
+                    'action_type': r[2], 'result': r[3],
+                    'duration_ms': int(r[4] or 0),
+                    'action_detail': (r[5] or '')[:160],
+                    'incident_id': r[6], 'error_type': r[7],
+                }
+                for r in heal_rows
+            ]
+
+            # playbooks 庫排行（success_count + fail_count + 是否 active）
+            pb_rows = s3.execute(
+                sa_text("""
+                    SELECT name, error_type, action_type, severity_min,
+                           success_count, fail_count, is_active, cooldown_min
+                    FROM playbooks
+                    ORDER BY (success_count + fail_count) DESC, success_count DESC
+                    LIMIT 12
+                """),
+            ).fetchall()
+            playbook_ranking = [
+                {
+                    'name': r[0], 'error_type': r[1], 'action_type': r[2],
+                    'severity': r[3], 'success': int(r[4] or 0),
+                    'fail': int(r[5] or 0), 'is_active': bool(r[6]),
+                    'cooldown_min': int(r[7] or 0),
+                    'success_rate': (
+                        float(r[4] or 0) / float((r[4] or 0) + (r[5] or 0)) * 100
+                    ) if ((r[4] or 0) + (r[5] or 0)) > 0 else 0,
+                }
+                for r in pb_rows
+            ]
+
+            # backup_log 7d 歷史
+            bk_rows = s3.execute(
+                sa_text("""
+                    SELECT created_at, backup_type, status, file_size_bytes,
+                           duration_seconds, error_message
+                    FROM backup_log
+                    WHERE created_at >= NOW() - INTERVAL '7 days'
+                    ORDER BY created_at DESC LIMIT 10
+                """),
+            ).fetchall()
+            backup_history = [
+                {
+                    'created_at': r[0].strftime('%Y-%m-%d %H:%M'),
+                    'backup_type': r[1], 'status': r[2],
+                    'size_mb': round(float(r[3] or 0) / (1024 * 1024), 1),
+                    'duration_s': round(float(r[4] or 0), 1),
+                    'error': (r[5] or '')[:120],
+                }
+                for r in bk_rows
+            ]
+
+            # embedding_retry_queue pending / failed
+            embed_q = s3.execute(
+                sa_text("""
+                    SELECT
+                        COUNT(*) FILTER (WHERE status = 'pending'),
+                        COUNT(*) FILTER (WHERE status = 'failed')
+                    FROM embedding_retry_queue
+                """),
+            ).fetchone()
+            embed_queue_pending = int(embed_q[0] or 0)
+            embed_queue_failed = int(embed_q[1] or 0)
+        finally:
+            s3.close()
+    except Exception:
+        pass
+
     return render_template(
         'admin/host_health.html',
         active_page='obs_host_health',
@@ -1513,4 +1917,10 @@ def host_health_dashboard():
         health_history=health_history,
         mcp_24h=mcp_24h,
         aiops_summary=aiops_summary,
+        recent_incidents=recent_incidents,
+        recent_heals=recent_heals,
+        playbook_ranking=playbook_ranking,
+        backup_history=backup_history,
+        embed_queue_pending=embed_queue_pending,
+        embed_queue_failed=embed_queue_failed,
     )

templates/admin/ai_calls_dashboard.html

@@ -124,6 +124,111 @@
   </div>
   {% endif %}
 
+  <!-- Phase 47 K-2: 24h 每小時呼叫趨勢 -->
+  {% if hourly_trend %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-chart-area me-2"></i>過去 24h 每小時呼叫趨勢</strong>
+      <small class="text-muted">每小時 bucket：呼叫數 · 成本 · 錯誤</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.85em;">
+        <thead class="table-light">
+          <tr>
+            <th>時段</th><th>呼叫數</th><th>成本 USD</th><th>錯誤</th><th style="width: 50%;">流量分布</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% set max_calls = (hourly_trend | map(attribute='calls') | max) or 1 %}
+          {% for h in hourly_trend %}
+          <tr>
+            <td><code>{{ h.hour }}</code></td>
+            <td><strong>{{ "{:,}".format(h.calls) }}</strong></td>
+            <td>${{ "%.3f"|format(h.cost) }}</td>
+            <td>
+              {% if h.errors > 0 %}<span class="text-danger">{{ h.errors }}</span>
+              {% else %}<small class="text-muted">0</small>{% endif %}
+            </td>
+            <td>
+              <div class="progress" style="height: 8px;">
+                <div class="progress-bar {% if h.errors > 0 %}bg-warning{% else %}bg-info{% endif %}"
+                     style="width: {{ (h.calls / max_calls * 100) | round | int }}%"></div>
+              </div>
+            </td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-2: by model 細分 -->
+  {% if by_model %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-microchip me-2"></i>依模型細分</strong>
+      <small class="text-muted">資料來源：ai_calls.model（細到模型版本）— Top 15</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr>
+            <th>模型</th><th>供應商</th>
+            <th class="text-end">呼叫</th><th class="text-end">Token</th>
+            <th class="text-end">成本 USD</th><th class="text-end">平均耗時</th>
+            <th class="text-end">錯誤</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for m in by_model %}
+          <tr>
+            <td><code>{{ m.model[:35] }}</code></td>
+            <td><span class="badge bg-secondary">{{ m.provider }}</span></td>
+            <td class="text-end">{{ "{:,}".format(m.calls) }}</td>
+            <td class="text-end">{{ "{:,}".format(m.tokens) }}</td>
+            <td class="text-end">${{ "%.4f"|format(m.cost) }}</td>
+            <td class="text-end">{{ m.avg_ms }} ms</td>
+            <td class="text-end">
+              {% if m.errors > 0 %}<span class="text-danger">{{ m.errors }}</span>
+              {% else %}<small class="text-muted">0</small>{% endif %}
+            </td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-2: agent_context 最近 10 筆 -->
+  {% if recent_contexts %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-comments-dollar me-2"></i>Agent 對話上下文（最近 10 筆）</strong>
+      <small class="text-muted">資料來源：agent_context（Hermes/OpenClaw/NemoTron 工作 session）</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.85em;">
+        <thead class="table-light">
+          <tr>
+            <th>時間</th><th>Agent</th><th>Key</th>
+            <th class="text-end">TTL min</th><th>預覽</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for c in recent_contexts %}
+          <tr>
+            <td><small>{{ c.created_at }}</small></td>
+            <td><span class="badge bg-info text-dark">{{ c.agent_name }}</span></td>
+            <td><code>{{ c.context_key }}</code></td>
+            <td class="text-end">{{ c.ttl_minutes }}</td>
+            <td><small class="text-muted">{{ c.preview }}{% if c.preview|length >= 120 %}…{% endif %}</small></td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
   <!-- by provider -->
   <div class="card mb-3">
     <div class="card-header"><strong>依供應商分組</strong></div>

templates/admin/budget.html

@@ -99,8 +99,88 @@
     </tbody>
   </table>
 
+  <!-- Phase 47 K-3: Top 5 cost-burning caller -->
+  {% if top_cost_callers %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-fire me-2"></i>當月 Top 5 燒錢呼叫端</strong>
+      <small class="text-muted">資料來源：ai_calls.cost_usd（caller 級彙總）</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr><th>呼叫端</th><th class="text-end">呼叫</th><th class="text-end">Token</th><th class="text-end">成本</th><th>佔比</th></tr>
+        </thead>
+        <tbody>
+          {% set max_cost = (top_cost_callers | map(attribute='cost') | max) or 1 %}
+          {% for c in top_cost_callers %}
+          <tr>
+            <td><code>{{ c.caller }}</code></td>
+            <td class="text-end">{{ "{:,}".format(c.calls) }}</td>
+            <td class="text-end">{{ "{:,}".format(c.tokens) }}</td>
+            <td class="text-end"><strong>${{ "%.2f"|format(c.cost) }}</strong></td>
+            <td>
+              <div class="progress" style="height: 6px;">
+                <div class="progress-bar bg-warning" style="width: {{ (c.cost / max_cost * 100) | round | int }}%"></div>
+              </div>
+            </td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-3: 30d daily cost trend by provider -->
+  {% if cost_trend_30d %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-chart-line me-2"></i>過去 30 日每日成本（依 provider）</strong>
+      <small class="text-muted">資料來源：ai_calls 每日 SUM(cost_usd) GROUP BY provider</small>
+    </div>
+    <div class="card-body p-0" style="max-height: 360px; overflow-y: auto;">
+      <table class="table table-sm mb-0" style="font-size: 0.85em;">
+        <thead class="table-light" style="position: sticky; top: 0;">
+          <tr><th>日期</th><th>供應商</th><th class="text-end">成本 USD</th></tr>
+        </thead>
+        <tbody>
+          {% for r in cost_trend_30d %}
+          <tr>
+            <td><code>{{ r.date }}</code></td>
+            <td><span class="badge bg-secondary">{{ r.provider }}</span></td>
+            <td class="text-end">${{ "%.4f"|format(r.cost) }}</td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-3: AI 價格決策 7d -->
+  {% if price_rec_7d %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-tag me-2"></i>AI 價格決策 7 日（ai_price_recommendations）</strong>
+      <small class="text-muted">展現 AI 編排的商業面產出 — 不只 cost，看實際決策數量</small>
+    </div>
+    <div class="card-body">
+      <div class="row g-2">
+        {% for p in price_rec_7d %}
+        <div class="col-md-3 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">{{ p.strategy }}</small>
+            <strong style="font-size: 1.4em;">{{ p.count }}</strong>
+            <small class="d-block text-muted">信心 {{ "%.2f"|format(p.avg_confidence) }}</small>
+          </div>
+        </div>
+        {% endfor %}
+      </div>
+    </div>
+  </div>
+  {% endif %}
+
   <p class="text-muted mt-3"><small>
-    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 29 — 預算控管
+    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 47 — 預算控管
+    （5 表深挖：ai_call_budgets / ai_calls / ai_price_recommendations / ai_insights / cost_throttle_state）
   </small></p>
 </div>
 

templates/admin/host_health.html

@@ -267,8 +267,204 @@
   </div>
   {% endif %}
 
+  <!-- Phase 47 K-1: 最近 10 筆 Incidents 詳細 -->
+  <div class="card mb-3">
+    <div class="card-header">
+      <strong><i class="fas fa-fire me-2"></i>最近 10 筆 Incidents</strong>
+      <small class="text-muted">資料來源：incidents（ADR-013 AutoHeal 觸發紀錄）</small>
+    </div>
+    <div class="card-body p-0">
+      {% if recent_incidents %}
+        <table class="table table-sm mb-0" style="font-size: 0.9em;">
+          <thead class="table-light">
+            <tr>
+              <th>建立時間</th><th>任務</th><th>錯誤類型</th>
+              <th>嚴重度</th><th>狀態</th><th>重試</th>
+              <th>錯誤訊息</th><th>解決時間</th>
+            </tr>
+          </thead>
+          <tbody>
+            {% for i in recent_incidents %}
+              <tr>
+                <td><small>{{ i.created_at }}</small></td>
+                <td><code>{{ i.task_name }}</code></td>
+                <td><span class="badge bg-secondary">{{ i.error_type }}</span></td>
+                <td>
+                  {% if i.severity in ('P0', 'P1') %}<span class="badge bg-danger">{{ i.severity }}</span>
+                  {% elif i.severity == 'P2' %}<span class="badge bg-warning">{{ i.severity }}</span>
+                  {% else %}<span class="badge bg-info text-dark">{{ i.severity }}</span>{% endif %}
+                </td>
+                <td>
+                  {% if i.status == 'open' %}<span class="badge bg-danger">未解決</span>
+                  {% elif i.status == 'resolved' %}<span class="badge bg-success">已解決</span>
+                  {% else %}<span class="badge bg-light text-dark">{{ i.status }}</span>{% endif %}
+                </td>
+                <td class="text-end">{{ i.retry_count }}</td>
+                <td><small class="text-muted">{{ i.error_message }}</small></td>
+                <td><small>{{ i.resolved_at or '—' }}</small></td>
+              </tr>
+            {% endfor %}
+          </tbody>
+        </table>
+      {% else %}
+        <div class="text-muted text-center p-3 small">
+          <i class="fas fa-shield-check me-1"></i>尚無 incident 紀錄（即系統尚未觸發過 AutoHeal）
+        </div>
+      {% endif %}
+    </div>
+  </div>
+
+  <!-- Phase 47 K-1: 最近 10 筆 Heal Logs 詳細 -->
+  <div class="card mb-3">
+    <div class="card-header">
+      <strong><i class="fas fa-band-aid me-2"></i>最近 10 筆 Heal Logs</strong>
+      <small class="text-muted">資料來源：heal_logs（自癒 playbook 執行歷史）</small>
+    </div>
+    <div class="card-body p-0">
+      {% if recent_heals %}
+        <table class="table table-sm mb-0" style="font-size: 0.9em;">
+          <thead class="table-light">
+            <tr>
+              <th>時間</th><th>動作</th><th>結果</th>
+              <th class="text-end">耗時 ms</th><th>關聯 Incident</th><th>細節</th>
+            </tr>
+          </thead>
+          <tbody>
+            {% for h in recent_heals %}
+              <tr>
+                <td><small>{{ h.created_at }}</small></td>
+                <td><span class="badge bg-info text-dark">{{ h.action_type or '—' }}</span></td>
+                <td>
+                  {% if h.result == 'success' %}<span class="badge bg-success"><i class="fas fa-check me-1"></i>成功</span>
+                  {% elif h.result == 'failed' %}<span class="badge bg-danger"><i class="fas fa-times me-1"></i>失敗</span>
+                  {% elif h.result == 'skipped' %}<span class="badge bg-secondary">跳過</span>
+                  {% else %}<span class="badge bg-light text-dark">{{ h.result }}</span>{% endif %}
+                </td>
+                <td class="text-end">{{ h.duration_ms }}</td>
+                <td><small>#{{ h.incident_id }} · <code>{{ h.error_type or '—' }}</code></small></td>
+                <td><small class="text-muted">{{ h.action_detail }}</small></td>
+              </tr>
+            {% endfor %}
+          </tbody>
+        </table>
+      {% else %}
+        <div class="text-muted text-center p-3 small">
+          <i class="fas fa-info-circle me-1"></i>尚無 heal log（一鍵 AutoHeal 觸發後將累積）
+        </div>
+      {% endif %}
+    </div>
+  </div>
+
+  <!-- Phase 47 K-1: AutoHeal Playbook 庫排行 -->
+  <div class="card mb-3">
+    <div class="card-header">
+      <strong><i class="fas fa-book-medical me-2"></i>AutoHeal Playbook 庫排行</strong>
+      <small class="text-muted">資料來源：playbooks（success/fail 累計，按執行次數排序）</small>
+    </div>
+    <div class="card-body p-0">
+      {% if playbook_ranking %}
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr>
+            <th>Playbook</th><th>錯誤類型</th><th>動作</th>
+            <th>嚴重度</th>
+            <th class="text-end">成功</th><th class="text-end">失敗</th>
+            <th class="text-end">成功率</th>
+            <th>狀態</th><th class="text-end">冷卻 min</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for p in playbook_ranking %}
+          <tr>
+            <td><strong>{{ p.name }}</strong></td>
+            <td><code>{{ p.error_type }}</code></td>
+            <td><span class="badge bg-info text-dark">{{ p.action_type }}</span></td>
+            <td>{{ p.severity }}</td>
+            <td class="text-end text-success">{{ p.success }}</td>
+            <td class="text-end text-danger">{{ p.fail }}</td>
+            <td class="text-end">
+              {% if (p.success + p.fail) > 0 %}
+                <strong class="{% if p.success_rate >= 80 %}text-success{% elif p.success_rate >= 50 %}text-warning{% else %}text-danger{% endif %}">
+                  {{ "%.0f"|format(p.success_rate) }}%
+                </strong>
+              {% else %}<small class="text-muted">—</small>{% endif %}
+            </td>
+            <td>
+              {% if p.is_active %}
+                <span class="badge bg-success">啟用</span>
+              {% else %}
+                <span class="badge bg-secondary">停用</span>
+              {% endif %}
+            </td>
+            <td class="text-end">{{ p.cooldown_min }}</td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+      {% else %}
+        <div class="text-muted text-center p-3 small">
+          <i class="fas fa-info-circle me-1"></i>尚無 playbook 資料（migration 013 + 020 是否已跑？）
+        </div>
+      {% endif %}
+    </div>
+  </div>
+
+  <!-- Phase 47 K-1: 備份歷史 7d -->
+  <div class="card mb-3">
+    <div class="card-header">
+      <strong><i class="fas fa-database me-2"></i>備份歷史（過去 7 日）</strong>
+      <small class="text-muted">資料來源：backup_log</small>
+    </div>
+    <div class="card-body p-0">
+      {% if backup_history %}
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr>
+            <th>時間</th><th>類型</th><th>狀態</th>
+            <th class="text-end">大小 (MB)</th><th class="text-end">耗時 s</th>
+            <th>錯誤</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for b in backup_history %}
+          <tr>
+            <td><small>{{ b.created_at }}</small></td>
+            <td><span class="badge bg-secondary">{{ b.backup_type }}</span></td>
+            <td>
+              {% if b.status == 'success' %}<span class="badge bg-success"><i class="fas fa-check me-1"></i>成功</span>
+              {% elif b.status == 'failed' %}<span class="badge bg-danger"><i class="fas fa-times me-1"></i>失敗</span>
+              {% else %}<span class="badge bg-warning">{{ b.status }}</span>{% endif %}
+            </td>
+            <td class="text-end">{{ "{:,}".format(b.size_mb) }}</td>
+            <td class="text-end">{{ b.duration_s }}</td>
+            <td><small class="text-muted">{{ b.error }}</small></td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+      {% else %}
+        <div class="text-muted text-center p-3 small">
+          <i class="fas fa-info-circle me-1"></i>過去 7 日無備份紀錄（每日 03:00 cron 執行）
+        </div>
+      {% endif %}
+    </div>
+  </div>
+
+  <!-- Phase 47 K-1: Embedding 重試佇列 -->
+  {% if embed_queue_pending > 0 or embed_queue_failed > 0 %}
+  <div class="alert alert-warning mb-3">
+    <strong><i class="fas fa-cog me-1"></i>Embedding 重試佇列：</strong>
+    待處理 <strong>{{ embed_queue_pending }}</strong> 筆 ·
+    失敗 <strong>{{ embed_queue_failed }}</strong> 筆
+    <small class="d-block text-muted mt-1">
+      資料來源：embedding_retry_queue — 卡住的 embedding 工作，需檢查 Ollama 健康
+    </small>
+  </div>
+  {% endif %}
+
   <p class="text-muted mt-3"><small>
-    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 40 — 主機健康監控（含 24h 歷史 / MCP / AIOps / AutoHeal L2）
+    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 47 — 主機健康監控
+    （8 表深挖：host_health_probes / mcp_calls / incidents / heal_logs / playbooks / backup_log / embedding_retry_queue / ai_call_budgets）
   </small></p>
 </div>
 

templates/admin/ppt_audit_history.html

@@ -130,8 +130,107 @@
     <code>python3 -c "from services.ppt_vision_service import ppt_vision_service; print(ppt_vision_service.check_ppt_file('reports/xxx.pptx'))"</code>
   </p>
 
+  <!-- Phase 47 K-6: 30d 統計 -->
+  {% if audit_30d_stats and audit_30d_stats.total > 0 %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-chart-pie me-2"></i>過去 30 日 PPT 審核統計</strong>
+      <small class="text-muted">資料來源：ppt_audit_results 全表聚合</small>
+    </div>
+    <div class="card-body">
+      <div class="row g-2">
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">總筆數</small>
+            <strong style="font-size: 1.4em;">{{ audit_30d_stats.total }}</strong>
+          </div>
+        </div>
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">通過</small>
+            <strong class="text-success" style="font-size: 1.4em;">{{ audit_30d_stats.passed }}</strong>
+          </div>
+        </div>
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">失敗</small>
+            <strong class="{% if audit_30d_stats.failed > 0 %}text-warning{% endif %}" style="font-size: 1.4em;">{{ audit_30d_stats.failed }}</strong>
+          </div>
+        </div>
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">錯誤</small>
+            <strong class="{% if audit_30d_stats.error > 0 %}text-danger{% endif %}" style="font-size: 1.4em;">{{ audit_30d_stats.error }}</strong>
+          </div>
+        </div>
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">通過率</small>
+            <strong class="{% if audit_30d_stats.pass_rate >= 80 %}text-success{% elif audit_30d_stats.pass_rate >= 60 %}text-warning{% else %}text-danger{% endif %}" style="font-size: 1.4em;">{{ "%.0f"|format(audit_30d_stats.pass_rate) }}%</strong>
+          </div>
+        </div>
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">總 issue 數</small>
+            <strong style="font-size: 1.4em;">{{ audit_30d_stats.total_issues }}</strong>
+          </div>
+        </div>
+      </div>
+      <div class="mt-2 small text-muted">
+        <i class="fas fa-info-circle me-1"></i>
+        通過 audit 平均信心度：<strong>{{ "%.2f"|format(audit_30d_stats.avg_confidence) }}</strong>
+      </div>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-6: Top 10 failure files -->
+  {% if top_failure_files %}
+  <div class="card mb-3" style="border-left: 4px solid #ffc107;">
+    <div class="card-header bg-light">
+      <strong><i class="fas fa-exclamation-triangle me-2"></i>Top 10 失敗檔案（30d）</strong>
+      <small class="text-muted">— 反覆失敗的 PPT 檔案，需手動處理或調整 generator</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr>
+            <th>檔名</th>
+            <th class="text-end">失敗次數</th>
+            <th class="text-end">總 issue</th>
+            <th>最近審核</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for f in top_failure_files %}
+          <tr>
+            <td><code>{{ f.filename }}</code></td>
+            <td class="text-end"><span class="badge bg-warning text-dark">{{ f.attempts }}</span></td>
+            <td class="text-end">{{ f.total_issues }}</td>
+            <td><small>{{ f.last_audit }}</small></td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  {% if (not audit_30d_stats or audit_30d_stats.total == 0) and not vision_enabled %}
+  <div class="alert alert-info mt-3">
+    <i class="fas fa-info-circle me-1"></i>
+    <strong>為什麼這頁空？</strong>
+    <ul class="mb-0 small mt-2">
+      <li>PPT_VISION_ENABLED=false（在 .env 設為 true 啟用）</li>
+      <li>188 主機需安裝 LibreOffice：<code>apt install libreoffice</code></li>
+      <li>需 Ollama 拉取 minicpm-v 模型（用於 PPT 視覺檢查）</li>
+      <li>啟用後每日 22:00 cron 自動掃當天新生 .pptx，審核結果寫入 ppt_audit_results 累積歷史</li>
+    </ul>
+  </div>
+  {% endif %}
+
   <p class="text-muted mt-3"><small>
-    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 40 — PPT 視覺審核歷史（含 AiderHeal L2）
+    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 47 — PPT 視覺審核歷史
+    （3 表深挖：ppt_audit_results / reports/ 檔案系統 / ai_insights RAG）
   </small></p>
 </div>
 

templates/admin/promotion_review.html

@@ -76,8 +76,109 @@
     </div>
   {% endif %}
 
+  <!-- Phase 47 K-4: 蒸餾池 30d 分布 -->
+  {% if episode_distribution_30d %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-flask me-2"></i>蒸餾池 30 日狀態分布</strong>
+      <small class="text-muted">資料來源：learning_episodes（不只看 awaiting，看完整流動）</small>
+    </div>
+    <div class="card-body">
+      <div class="row g-2">
+        {% for status, cnt in episode_distribution_30d.items() %}
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">
+              {% if status == 'pending' %}<i class="fas fa-hourglass-start"></i> 待處理
+              {% elif status == 'awaiting_review' %}<i class="fas fa-user-clock"></i> 待審核
+              {% elif status == 'approved' %}<i class="fas fa-check-circle text-success"></i> 已晉升
+              {% elif status == 'rejected_quality' %}<i class="fas fa-times text-danger"></i> 品質拒
+              {% elif status == 'rejected_hallucination' %}<i class="fas fa-times text-danger"></i> 幻覺拒
+              {% elif status == 'rejected_duplicate' %}<i class="fas fa-clone text-warning"></i> 重複拒
+              {% elif status == 'rejected_human' %}<i class="fas fa-user-times text-danger"></i> 人工拒
+              {% elif status == 'expired' %}<i class="fas fa-clock text-muted"></i> 已過期
+              {% else %}{{ status }}{% endif %}
+            </small>
+            <strong style="font-size: 1.4em;">{{ cnt }}</strong>
+          </div>
+        </div>
+        {% endfor %}
+      </div>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-4: 最近 10 筆已晉升 ai_insights -->
+  {% if latest_insights %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-book me-2"></i>知識庫最近 10 筆 ai_insights</strong>
+      <small class="text-muted">資料來源：ai_insights — 已晉升內容預覽</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr><th>#</th><th>類型</th><th>期間</th><th>SKU</th><th>建立時間</th><th>預覽</th></tr>
+        </thead>
+        <tbody>
+          {% for i in latest_insights %}
+          <tr>
+            <td><code>#{{ i.id }}</code></td>
+            <td><span class="badge bg-info text-dark">{{ i.insight_type }}</span></td>
+            <td><small>{{ i.period or '—' }}</small></td>
+            <td><small>{{ i.product_sku or '—' }}</small></td>
+            <td><small>{{ i.created_at }}</small></td>
+            <td><small class="text-muted">{{ i.preview }}{% if i.preview|length >= 160 %}…{% endif %}</small></td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-4: agent_strategy_weights TOP 12 -->
+  {% if strategy_weights %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-balance-scale me-2"></i>OpenClaw 學習策略權重 TOP 12</strong>
+      <small class="text-muted">資料來源：agent_strategy_weights（ADR-012 closed-loop learning）</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr>
+            <th>策略 Key</th>
+            <th class="text-end">權重</th>
+            <th class="text-end">成功</th>
+            <th class="text-end">失敗</th>
+            <th class="text-end">成功率</th>
+            <th>更新時間</th>
+          </tr>
+        </thead>
+        <tbody>
+          {% for s in strategy_weights %}
+          <tr>
+            <td><code>{{ s.strategy_key }}</code></td>
+            <td class="text-end"><strong>{{ "%.2f"|format(s.weight) }}</strong></td>
+            <td class="text-end text-success">{{ s.success }}</td>
+            <td class="text-end text-danger">{{ s.fail }}</td>
+            <td class="text-end">
+              {% if (s.success + s.fail) > 0 %}
+                <span class="{% if s.success_rate >= 70 %}text-success{% elif s.success_rate >= 40 %}text-warning{% else %}text-danger{% endif %}">
+                  {{ "%.0f"|format(s.success_rate) }}%
+                </span>
+              {% else %}<small class="text-muted">—</small>{% endif %}
+            </td>
+            <td><small>{{ s.updated_at }}</small></td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
   <p class="text-muted mt-3"><small>
-    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 29 — RAG 學習晉升審核
+    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 47 — RAG 學習晉升審核
+    （4 表深挖：learning_episodes / ai_insights / agent_strategy_weights / rag_query_log）
   </small></p>
 </div>
 

templates/admin/quality_trend.html

@@ -150,8 +150,98 @@
     </div>
   </div>
 
+  <!-- Phase 47 K-5: RAG 整體 feedback 分布 -->
+  {% if rag_overall_dist %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-poll me-2"></i>RAG 整體反饋分布（過去 {{ days }} 日）</strong>
+      <small class="text-muted">資料來源：rag_query_log.feedback_score（含全 caller，1-5 分）</small>
+    </div>
+    <div class="card-body">
+      <div class="row g-2">
+        {% set total_fb = (rag_overall_dist | sum(attribute='count')) or 1 %}
+        {% for r in rag_overall_dist %}
+        <div class="col-md-2 col-sm-4">
+          <div class="border rounded p-2 text-center">
+            <small class="text-muted d-block">
+              {% for _ in range(r.score) %}<i class="fas fa-star text-warning"></i>{% endfor %}
+              {% for _ in range(5 - r.score) %}<i class="far fa-star text-muted"></i>{% endfor %}
+            </small>
+            <strong style="font-size: 1.4em;">{{ r.count }}</strong>
+            <small class="d-block text-muted">{{ "%.1f"|format(r.count / total_fb * 100) }}%</small>
+          </div>
+        </div>
+        {% endfor %}
+      </div>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-5: Action Plans status 分布 -->
+  {% if action_plans_status %}
+  <div class="card mb-3">
+    <div class="card-header"><strong><i class="fas fa-tasks me-2"></i>Action Plans 狀態分布（過去 {{ days }} 日）</strong>
+      <small class="text-muted">資料來源：action_plans（NemoTron/OpenClaw 的計畫產出 + 審核狀態）</small>
+    </div>
+    <div class="card-body p-0">
+      <table class="table table-sm mb-0" style="font-size: 0.9em;">
+        <thead class="table-light">
+          <tr><th>狀態</th><th>計畫類型</th><th class="text-end">數量</th></tr>
+        </thead>
+        <tbody>
+          {% for a in action_plans_status %}
+          <tr>
+            <td>
+              {% if a.status == 'pending' %}<span class="badge bg-warning text-dark">{{ a.status }}</span>
+              {% elif a.status == 'approved' %}<span class="badge bg-success">{{ a.status }}</span>
+              {% elif a.status == 'executed' %}<span class="badge bg-primary">{{ a.status }}</span>
+              {% elif a.status == 'rejected' %}<span class="badge bg-danger">{{ a.status }}</span>
+              {% else %}<span class="badge bg-secondary">{{ a.status }}</span>{% endif %}
+            </td>
+            <td><code>{{ a.plan_type }}</code></td>
+            <td class="text-end">{{ a.count }}</td>
+          </tr>
+          {% endfor %}
+        </tbody>
+      </table>
+    </div>
+  </div>
+  {% endif %}
+
+  <!-- Phase 47 K-5: Action Outcomes verdict 分布 -->
+  {% if action_outcomes_stats %}
+  <div class="card mb-3" style="border-left: 4px solid #198754;">
+    <div class="card-header bg-light"><strong><i class="fas fa-trophy me-2"></i>Action Outcomes 成效（過去 {{ days }} 日）</strong>
+      <small class="text-muted">資料來源：action_outcomes（ADR-012 閉環學習：實際動作有效嗎？）</small>
+    </div>
+    <div class="card-body">
+      <div class="row g-2">
+        {% set total_ao = (action_outcomes_stats | sum(attribute='count')) or 1 %}
+        {% for r in action_outcomes_stats %}
+        <div class="col-md-3 col-sm-6">
+          <div class="border rounded p-2 text-center"
+               style="border-left-width: 4px !important;
+                      border-left-color: {% if r.verdict == 'effective' %}#198754
+                                          {% elif r.verdict == 'backfired' %}#dc3545
+                                          {% else %}#6c757d{% endif %} !important;">
+            <small class="text-muted d-block">
+              {% if r.verdict == 'effective' %}<i class="fas fa-check-circle text-success"></i> 有效
+              {% elif r.verdict == 'backfired' %}<i class="fas fa-times-circle text-danger"></i> 適得其反
+              {% elif r.verdict == 'neutral' %}<i class="fas fa-minus text-secondary"></i> 無顯著效果
+              {% else %}{{ r.verdict }}{% endif %}
+            </small>
+            <strong style="font-size: 1.4em;">{{ r.count }}</strong>
+            <small class="d-block text-muted">{{ "%.1f"|format(r.count / total_ao * 100) }}%</small>
+          </div>
+        </div>
+        {% endfor %}
+      </div>
+    </div>
+  </div>
+  {% endif %}
+
   <p class="text-muted mt-3"><small>
-    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 29 — Caller 反饋趨勢
+    <i class="fas fa-robot me-1"></i>Operation Ollama-First v5.0 / Phase 47 — Caller 反饋趨勢
+    （6 表深挖：rag_query_log / learning_episodes / ai_insights / action_plans / action_outcomes / agent_strategy_weights）
   </small></p>
 </div>
 {% endblock %}