feat(governance): AI 治理事件處理鏈四軌交付（C/D/B/A）

【十二人專家團隊全景掃描 + 並行四軌實施】

統帥質疑「有讓 12-agent 一起協作嗎」後，依照團隊規則完成全鏈路交付：
onboarder + critic + db-expert + debugger + frontend-designer 並行掃描，
找到 6 大 Gap，再由 fullstack-engineer × 4、refactor-specialist 協作落地。

【Track C — trust_drift 雙寫整併】

兩條獨立寫 event_type=trust_drift 路徑互不呼叫，下游 consumer 拿到雙份資料
無法判定 source-of-truth。整併保留 governance_agent.check_trust_drift（功能
更全：auto-deprecate + Telegram + PG），TrustDriftDetector 降為純統計 lib，
W-6 watchdog 改呼叫 governance_agent。新增 TestSinglePgWritePerDriftScenario
驗證同一 drift 場景只觸發一次 PG 寫入。

  變更：
    - apps/api/src/services/trust_drift_detector.py（lib only，不再寫 PG）
    - apps/api/tests/test_trust_drift_watchdog.py（W-6 改 mock governance_agent）

【Track D — governance_remediation_dispatch 派遣表】

ai_governance_events 是不可變 Event Sourcing，不能塞執行狀態。新建派遣表
作為投影層：1 event → 0..N dispatches，狀態可變、可重試、可審計。

  - PgEnum 5 種 event_type + 7 階段狀態機（pending → dispatched → executing →
    succeeded/failed/cancelled/skipped）
  - 失敗重試 INSERT 新 row（不改舊 row 的 status，保留審計痕跡）
  - Partial unique index ux_grd_one_active_per_event 強制「同事件唯一活躍」
  - 4 個複合 index 支援 worker poll、去重查詢、觀測面板
  - FK 對應 ai_governance_events / playbooks / incidents / approval_records
    全部 SET NULL（avoid cascade lock，但 governance_event 用 RESTRICT）

  變更：
    - apps/api/src/db/models.py（GovernanceRemediationDispatch ORM class）
    - apps/api/migrations/governance_remediation_dispatch_2026-05-03.sql
    - apps/api/src/repositories/governance_remediation_dispatch_repo.py
      （6 個 async 函式 + 3 個自訂例外：DispatchAlreadyActive /
       InvalidStatusTransition / DispatchNotFound）
    - apps/api/src/models/governance_dispatch.py（DecisionContextV1 等 4 schema）
    - apps/api/tests/test_governance_remediation_dispatch.py（29 tests）

【Track B — /governance 頁面】

後端 PR1 三個 endpoint + 前端 PR2-5 完整三 Tab。

PR1 後端：
  - GET /api/v1/ai/governance/events（events_tab，含 event_type/severity/
    狀態/時間範圍篩選 + 分頁）
  - GET /api/v1/ai/governance/queue（queue_tab，含 graceful fallback：
    dispatch 表不存在時回 table_pending=True 不拋 500）
  - GET /api/v1/ai/governance/summary（slo_tab 30d 違反時序圖）
  - severity 映射規則寫死（critic 建議未來移 settings）

PR2-5 前端：
  - /governance 路由 + AppLayout + Compliance Badge 橫幅 + PageTabs
  - SLO Tab：3 KPI 卡片（Syne 28px + StatusOrb + 7d sparkline）+
    30d 違反 stacked BarChart
  - Events Tab：篩選列 + 表格 + inline 展開行（JSON / 修復建議 / 派遣記錄）
  - Queue Tab：HITL 待辦卡片 + 信任度進度條 + 批准/拒絕按鈕（本 PR console.log）
  - Sidebar 加入「AI 治理」入口（ShieldCheck icon）
  - i18n 雙語完整（governance namespace + nav.governance）
  - 7 個新元件：slo-kpi-card / slo-violation-chart / events-table /
    events-filter-bar / event-detail-drawer / queue-item-card / queue-history-tabs

  變更：
    - apps/api/src/api/v1/ai_governance.py（router）
    - apps/api/src/services/governance_query_service.py
    - apps/api/src/models/governance.py（Pydantic V2 schemas）
    - apps/api/tests/test_ai_governance_endpoints.py（21 tests）
    - apps/web/src/app/[locale]/governance/（page + 3 tabs）
    - apps/web/src/components/governance/（7 元件）
    - apps/web/messages/{zh-TW,en}.json（governance namespace）
    - apps/web/src/components/layout/sidebar.tsx（+1 行）
    - apps/api/src/main.py（router include）

【Track A — GovernanceDispatcher 決策融合】

把治理事件接到 remediation 執行器，走北極星方向決策融合（LLM × Playbook trust
× MCP），符合「禁寫死規則」鐵律。

  - 設計鐵律：DecisionFusionAdapter 是新增 wrapper，**不修改任何 Tier 3 檔**
    （decision_manager / learning_service / trust_engine），只 consume 既有 API
  - 三維融合公式：confidence = 0.4×llm + 0.3×playbook_trust + 0.3×mcp_consistency
    （權重加 TODO 標明未來由 AI 自學調整）
  - 三分支決策路徑：
    confidence ≥ 0.85 → auto_dispatch（status=dispatched）
    0.65 ≤ confidence < 0.85 → pending_approval（HITL）
    confidence < 0.65 → skip + log
  - decision_context JSONB 完整記錄三維輸入快照（給未來 fine-tune 用）
  - poll 30s 掃 unresolved 事件，仿 governance loop 模式
  - 重複事件擋去重（呼叫 get_active_for_event）

  變更：
    - apps/api/src/services/governance_dispatcher.py
    - apps/api/src/services/decision_fusion_adapter.py
    - apps/api/tests/test_governance_dispatcher.py（14 tests）
    - apps/api/src/main.py（lifespan task 接 run_governance_dispatcher_loop）

【驗證】

1836 個 unit test 全過（29 skipped 為既有 PG integration env 問題）

【調度教訓 — 已記入 memory】

- vuln-verifier 應在 fullstack-engineer **之前**跑（避免並行讀到已修代碼誤判）
- critic 雙輪審查不可省（第二輪抓到 NaN sentinel + Prom rule 連鎖）
- 北極星「禁寫死規則」搭配 decision-fusion 確實實施

【未動 Tier 3 — 已驗證】

git diff 確認本 commit 完全沒改 decision_manager.py / learning_service.py /
trust_engine.py，只新增 wrapper service consume 既有 API。

🤖 Generated with [Claude Code](https://claude.com/claude-code)

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

apps/api/src/services/decision_fusion_adapter.py

@@ -0,0 +1,538 @@
+"""
+GovernanceDispatcher 決策融合適配器
+======================================
+將 decision_fusion / playbook_service / Ollama 的既有能力
+組合成「給治理事件用的三維融合介面」。
+
+設計原則：
+- 不修改任何 Tier 3 檔（decision_manager / learning_service / trust_engine）
+- 只 consume 公開 API（read-only）
+- 三維融合：LLM × Playbook trust × MCP 情報
+- Exception 隔離：任一維度失敗 → 中立值 0.5，不阻塞主流程
+
+融合公式（起始權重，TODO 移到 settings 由 AI 自學調整）：
+  confidence = w_llm * llm_score + w_playbook * playbook_trust + w_mcp * mcp_score
+  w_llm=0.4, w_playbook=0.3, w_mcp=0.3
+
+決策分支（閾值 TODO 移到 settings）：
+  confidence >= 0.85  → auto_dispatch
+  0.65 <= conf < 0.85 → pending_approval
+  conf < 0.65         → skip
+
+2026-05-03 ogt + Claude Sonnet 4.6（亞太）: GovernanceDispatcher Wave 2E 實作
+"""
+
+from __future__ import annotations
+
+import asyncio
+import re
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, Any, Literal
+
+import httpx
+import structlog
+
+from src.core.config import get_settings
+
+if TYPE_CHECKING:
+    from src.db.models import AiGovernanceEvent
+
+logger = structlog.get_logger(__name__)
+
+# =============================================================================
+# 常數
+# TODO: 移到 settings（ADR-P2E-FUTURE），屆時可讓 AI 自學調整
+# =============================================================================
+
+# 三維融合權重（0.4 / 0.3 / 0.3）
+_W_LLM: float = 0.4       # TODO: 由 AI 自學調整，初始值 0.4
+_W_PLAYBOOK: float = 0.3   # TODO: 由 AI 自學調整，初始值 0.3
+_W_MCP: float = 0.3        # TODO: 由 AI 自學調整，初始值 0.3
+
+# 決策分支閾值
+# TODO: 移到 settings，未來由 AI 根據 false-positive rate 動態調整
+_AUTO_DISPATCH_THRESHOLD: float = 0.85     # >= 此值 → auto_dispatch
+_PENDING_APPROVAL_THRESHOLD: float = 0.65  # >= 此值 < AUTO → pending_approval
+#                                           # < 此值 → skip
+
+# Ollama 推理超時（秒）
+_LLM_TIMEOUT_SEC: float = 30.0
+
+# Prometheus 查詢超時（秒）
+_PROM_TIMEOUT_SEC: float = 5.0
+
+
+# =============================================================================
+# FusedDecision 資料結構
+# =============================================================================
+
+@dataclass
+class FusedDecision:
+    """三維融合決策輸出。
+
+    所有分數均為 0.0-1.0（0.5 為中立值，任一維度失敗時使用）。
+    decision_path 決定 GovernanceDispatcher 寫入哪種 dispatch。
+
+    Attributes:
+        confidence: 三維加權融合分數（0.0-1.0）
+        recommended_action: LLM 推薦的修復動作摘要（≤200 字）
+        matched_playbook_id: 最高相似度的 Playbook ID（可 None）
+        playbook_trust: matched_playbook 的 trust_score（可 None）
+        llm_reasoning: LLM 原始輸出摘要（dict，供 decision_context JSONB 記錄）
+        mcp_snapshot: MCP 情報快照（dict，供 decision_context JSONB 記錄）
+        decision_path: auto_dispatch / pending_approval / skip
+        llm_score: LLM 分數（0.0-1.0）
+        playbook_score: Playbook 信任分數（0.0-1.0，無 playbook 時 0.3）
+        mcp_score: MCP 感官品質分數（0.0-1.0）
+    """
+    confidence: float
+    recommended_action: str
+    matched_playbook_id: str | None
+    playbook_trust: float | None
+    llm_reasoning: dict[str, Any]
+    mcp_snapshot: dict[str, Any]
+    decision_path: Literal["auto_dispatch", "pending_approval", "skip"]
+    llm_score: float
+    playbook_score: float
+    mcp_score: float
+
+
+# =============================================================================
+# DecisionFusionAdapter
+# =============================================================================
+
+class DecisionFusionAdapter:
+    """治理事件決策融合適配器。
+
+    將 decision_fusion / playbook_service / MCP 的既有能力組合成
+    「給治理事件用的三維融合介面」。本類不修改任何 Tier 3 檔，只 consume。
+
+    不注入 Tier 3 class：
+    - DecisionManager — 有 incident 中心的複雜狀態機，不適合治理事件
+    - TrustEngine — 只管理 incident 信任分數
+    - LearningService — 只管理 KM 寫入路徑
+
+    本 Adapter 直接呼叫：
+    - Ollama（仿 decision_fusion._score_hermes 模式）→ LLM 推理
+    - playbook_service.get_recommendations → Playbook trust
+    - Prometheus provider → MCP 情報
+    """
+
+    def __init__(self) -> None:
+        self._settings = get_settings()
+
+    # =========================================================================
+    # Public API
+    # =========================================================================
+
+    async def fuse_decision(self, event: "AiGovernanceEvent") -> FusedDecision:
+        """三維融合：LLM × Playbook × MCP → FusedDecision。
+
+        三個維度並行評估（asyncio.gather），任一失敗靜默降為 0.5。
+        依 confidence 決定 decision_path。
+
+        Args:
+            event: AiGovernanceEvent ORM 物件（不修改此物件）
+
+        Returns:
+            FusedDecision 含完整三維快照，供 dispatcher 寫入 decision_context
+        """
+        # 並行取三維分數
+        results = await asyncio.gather(
+            self._score_llm(event),
+            self._score_playbook(event),
+            self._score_mcp(event),
+            return_exceptions=True,
+        )
+
+        # 安全解包（Exception → 中立值 0.5）
+        llm_result = results[0]
+        playbook_result = results[1]
+        mcp_result = results[2]
+
+        if isinstance(llm_result, Exception):
+            logger.warning(
+                "fusion_llm_score_failed",
+                event_id=event.id,
+                event_type=event.event_type,
+                error=str(llm_result),
+            )
+            llm_result = (0.5, "（LLM 評估失敗，使用中立值）", {})
+
+        if isinstance(playbook_result, Exception):
+            logger.warning(
+                "fusion_playbook_score_failed",
+                event_id=event.id,
+                error=str(playbook_result),
+            )
+            playbook_result = (0.3, None, None)
+
+        if isinstance(mcp_result, Exception):
+            logger.warning(
+                "fusion_mcp_score_failed",
+                event_id=event.id,
+                error=str(mcp_result),
+            )
+            mcp_result = (0.5, {})
+
+        llm_score, recommended_action, llm_reasoning = llm_result
+        playbook_score, matched_playbook_id, playbook_trust = playbook_result
+        mcp_score, mcp_snapshot = mcp_result
+
+        # 三維加權融合
+        # TODO: 移到 settings，未來由 AI 自學調整 _W_LLM / _W_PLAYBOOK / _W_MCP
+        confidence = (
+            _W_LLM * llm_score
+            + _W_PLAYBOOK * playbook_score
+            + _W_MCP * mcp_score
+        )
+        confidence = max(0.0, min(1.0, confidence))
+
+        # 決策分支
+        # TODO: 閾值移到 settings，未來由 AI 根據 false-positive rate 動態調整
+        if confidence >= _AUTO_DISPATCH_THRESHOLD:
+            decision_path: Literal["auto_dispatch", "pending_approval", "skip"] = "auto_dispatch"
+        elif confidence >= _PENDING_APPROVAL_THRESHOLD:
+            decision_path = "pending_approval"
+        else:
+            decision_path = "skip"
+
+        logger.info(
+            "governance_fusion_complete",
+            event_id=event.id,
+            event_type=event.event_type,
+            llm_score=round(llm_score, 4),
+            playbook_score=round(playbook_score, 4),
+            mcp_score=round(mcp_score, 4),
+            confidence=round(confidence, 4),
+            decision_path=decision_path,
+        )
+
+        return FusedDecision(
+            confidence=confidence,
+            recommended_action=recommended_action,
+            matched_playbook_id=matched_playbook_id,
+            playbook_trust=playbook_trust,
+            llm_reasoning=llm_reasoning,
+            mcp_snapshot=mcp_snapshot,
+            decision_path=decision_path,
+            llm_score=llm_score,
+            playbook_score=playbook_score,
+            mcp_score=mcp_score,
+        )
+
+    # =========================================================================
+    # 維度 1：LLM 推理（Ollama qwen3:8b — 仿 decision_fusion._score_hermes）
+    # =========================================================================
+
+    async def _score_llm(
+        self, event: "AiGovernanceEvent"
+    ) -> tuple[float, str, dict[str, Any]]:
+        """Ollama LLM 推理：治理事件情境 → 建議動作 + 信心度。
+
+        Prompt 設計：
+        - 提供 event_type + details 摘要（sanitize 後）
+        - 要求輸出「信心度（0-1）+ 建議動作」
+
+        Returns:
+            (llm_score, recommended_action, llm_reasoning_dict)
+        """
+        event_type = str(event.event_type or "unknown")
+        details_summary = self._summarize_details(event.details or {})
+
+        prompt = (
+            "你是 AIOps 治理分析員。根據以下治理事件，評估自動修復的可行性與建議動作。\n\n"
+            f"【事件類型】{event_type}\n"
+            f"【事件摘要】{details_summary}\n\n"
+            "請以以下格式回應（不超過 200 字）：\n"
+            "CONFIDENCE: [0.0-1.0 的數字]\n"
+            "ACTION: [具體建議修復動作，≤100字]\n\n"
+            "注意：\n"
+            "- CONFIDENCE 越高表示越適合自動執行\n"
+            "- 若事件模糊或影響範圍不明，給低分（0.3-0.5）\n"
+            "- 若有明確、低風險的修復路徑，可給高分（0.7-0.9）\n"
+            "只輸出 CONFIDENCE 和 ACTION 兩行，不要其他解釋。"
+        )
+
+        ollama_url = getattr(self._settings, "OLLAMA_URL", "http://192.168.0.111:11434")
+
+        try:
+            async with httpx.AsyncClient(
+                timeout=httpx.Timeout(_LLM_TIMEOUT_SEC, connect=5.0)
+            ) as client:
+                resp = await client.post(
+                    f"{ollama_url}/api/generate",
+                    json={
+                        "model": "qwen3:8b",
+                        "prompt": prompt,
+                        "stream": False,
+                        "options": {"num_predict": 128, "temperature": 0.1},
+                    },
+                )
+                if resp.status_code != 200:
+                    logger.warning(
+                        "fusion_llm_http_error",
+                        status=resp.status_code,
+                        event_id=event.id,
+                    )
+                    return 0.5, "（LLM 不可用，使用中立值）", {"error": f"http_{resp.status_code}"}
+
+                raw_text = resp.json().get("response", "").strip()
+        except Exception as exc:
+            logger.warning("fusion_llm_request_failed", event_id=event.id, error=str(exc))
+            return 0.5, "（LLM 連線失敗，使用中立值）", {"error": str(exc)}
+
+        # 移除 <think> 標籤（qwen3 CoT 輸出）
+        clean = re.sub(r"<think>.*?</think>", "", raw_text, flags=re.DOTALL).strip()
+
+        # 解析 CONFIDENCE 行
+        llm_score = 0.5
+        conf_match = re.search(r"CONFIDENCE:\s*([01]?\.\d+|[01])", clean, re.IGNORECASE)
+        if conf_match:
+            try:
+                llm_score = max(0.0, min(1.0, float(conf_match.group(1))))
+            except ValueError:
+                pass
+
+        # 解析 ACTION 行
+        recommended_action = "（LLM 未提供明確建議）"
+        action_match = re.search(r"ACTION:\s*(.+)", clean, re.IGNORECASE)
+        if action_match:
+            recommended_action = action_match.group(1).strip()[:200]
+
+        llm_reasoning = {
+            "raw_text_preview": raw_text[:300],
+            "parsed_confidence": llm_score,
+            "parsed_action": recommended_action,
+            "event_type": event_type,
+        }
+
+        logger.debug(
+            "fusion_llm_scored",
+            event_id=event.id,
+            llm_score=llm_score,
+            action_preview=recommended_action[:60],
+        )
+        return llm_score, recommended_action, llm_reasoning
+
+    # =========================================================================
+    # 維度 2：Playbook 比對 + trust_score
+    # =========================================================================
+
+    async def _score_playbook(
+        self, event: "AiGovernanceEvent"
+    ) -> tuple[float, str | None, float | None]:
+        """Playbook 相似度比對 → 取最高 trust_score。
+
+        治理事件沒有 SymptomPattern，用 event_type 作為 alert_name 搜尋。
+        無命中時返回保守初始值 (0.3, None, None)。
+
+        Returns:
+            (playbook_score, matched_playbook_id, playbook_trust)
+        """
+        from src.models.playbook import SymptomPattern
+        from src.services.playbook_service import get_playbook_service
+
+        symptoms = SymptomPattern(
+            alert_names=[event.event_type or "unknown"],
+            affected_services=[],
+            severity_range=["P2"],
+            keywords=self._extract_keywords(event.details or {}),
+        )
+
+        try:
+            svc = get_playbook_service()
+            recommendations = await svc.get_recommendations(
+                symptoms=symptoms,
+                top_k=1,
+                use_rag=False,  # 治理事件用 Jaccard 精確比對即可
+            )
+        except Exception as exc:
+            logger.warning("fusion_playbook_lookup_failed", event_id=event.id, error=str(exc))
+            return 0.3, None, None
+
+        if not recommendations:
+            logger.debug("fusion_playbook_no_match", event_id=event.id, event_type=event.event_type)
+            return 0.3, None, None
+
+        best = recommendations[0]
+        trust = float(best.playbook.trust_score)
+        playbook_id = best.playbook.playbook_id
+
+        logger.debug(
+            "fusion_playbook_matched",
+            event_id=event.id,
+            playbook_id=playbook_id,
+            trust_score=trust,
+            similarity=round(best.similarity_score, 4),
+        )
+        return trust, playbook_id, trust
+
+    # =========================================================================
+    # 維度 3：MCP 情報（Prometheus）
+    # =========================================================================
+
+    async def _score_mcp(
+        self, event: "AiGovernanceEvent"
+    ) -> tuple[float, dict[str, Any]]:
+        """Prometheus 情報採集 → MCP 感官品質分數。
+
+        查詢與事件相關的核心指標（autonomy_rate / hallucination_rate）。
+        MCP 不可用時返回中立值 (0.5, {})。
+
+        Returns:
+            (mcp_score, mcp_snapshot_dict)
+        """
+        prom_url = getattr(
+            self._settings, "PROMETHEUS_URL", "http://prometheus.observability.svc:9090"
+        )
+
+        # 依 event_type 選擇查詢指標（治理事件相關）
+        queries: dict[str, str] = self._get_mcp_queries(event.event_type or "unknown")
+
+        snapshot: dict[str, Any] = {}
+        success_count = 0
+        total_count = len(queries)
+
+        if total_count == 0:
+            return 0.5, {"reason": "no_queries_for_event_type"}
+
+        try:
+            async with httpx.AsyncClient(timeout=_PROM_TIMEOUT_SEC) as client:
+                for metric_name, query in queries.items():
+                    try:
+                        resp = await client.get(
+                            f"{prom_url}/api/v1/query",
+                            params={"query": query},
+                        )
+                        data = resp.json()
+                        if data.get("status") == "success":
+                            result_list = data.get("data", {}).get("result", [])
+                            if result_list:
+                                value = float(result_list[0]["value"][1])
+                                snapshot[metric_name] = round(value, 4)
+                                success_count += 1
+                            else:
+                                snapshot[metric_name] = None  # 有回應但無資料
+                    except Exception as exc:
+                        snapshot[metric_name] = f"error:{exc!s:.60}"
+        except Exception as exc:
+            logger.warning("fusion_mcp_prometheus_failed", event_id=event.id, error=str(exc))
+            return 0.5, {"error": str(exc)}
+
+        # 品質分數：成功取得資料的指標比例（映射到 [0.2, 0.9]）
+        if total_count > 0:
+            ratio = success_count / total_count
+            mcp_score = 0.2 + 0.7 * ratio
+        else:
+            mcp_score = 0.5
+
+        snapshot["_meta"] = {
+            "success_count": success_count,
+            "total_queries": total_count,
+            "quality_score": round(mcp_score, 4),
+        }
+
+        logger.debug(
+            "fusion_mcp_scored",
+            event_id=event.id,
+            mcp_score=round(mcp_score, 4),
+            success=success_count,
+            total=total_count,
+        )
+        return mcp_score, snapshot
+
+    # =========================================================================
+    # Helpers
+    # =========================================================================
+
+    @staticmethod
+    def _summarize_details(details: dict[str, Any]) -> str:
+        """從 details dict 提取可讀摘要（≤300 字）。"""
+        if not details:
+            return "（無詳細資訊）"
+
+        parts: list[str] = []
+
+        # 常見欄位優先展示
+        for key in ("status", "impact", "remediation", "reason"):
+            val = details.get(key)
+            if val is None:
+                continue
+            if isinstance(val, dict):
+                inner = "; ".join(f"{k}={v}" for k, v in list(val.items())[:4])
+                parts.append(f"{key}: {inner}")
+            elif isinstance(val, (str, int, float)):
+                parts.append(f"{key}: {val!s:.80}")
+
+        if not parts:
+            # fallback: 取前幾個 top-level k=v
+            parts = [f"{k}={v!s:.40}" for k, v in list(details.items())[:5]]
+
+        return "; ".join(parts)[:300]
+
+    @staticmethod
+    def _extract_keywords(details: dict[str, Any]) -> list[str]:
+        """從 details 提取關鍵字供 Playbook 搜尋（最多 5 個）。"""
+        keywords: list[str] = []
+
+        for key in ("remediation", "actionable", "impact"):
+            val = details.get(key)
+            if isinstance(val, dict):
+                for sub_key in ("next_action", "items"):
+                    sub = val.get(sub_key)
+                    if isinstance(sub, str):
+                        keywords.append(sub[:50])
+                    elif isinstance(sub, list):
+                        keywords.extend(str(x)[:40] for x in sub[:2])
+
+        return keywords[:5]
+
+    @staticmethod
+    def _get_mcp_queries(event_type: str) -> dict[str, str]:
+        """依 event_type 返回相關 Prometheus 查詢指標。
+
+        不硬寫 event_type → action 對應規則，僅決定「看哪些指標」。
+        """
+        # 通用指標（所有 event_type 都查）
+        base_queries: dict[str, str] = {
+            "autonomy_rate": "sli:autonomy_rate:5m",
+            "decision_accuracy": "sli:decision_accuracy:5m",
+        }
+
+        # 依 event_type 補充針對性指標
+        extra: dict[str, str] = {}
+
+        if event_type in ("trust_drift", "execution_blast_radius"):
+            extra["km_growth_rate"] = "sli:km_growth_rate:24h"
+        elif event_type in ("knowledge_degradation", "kb_stale"):
+            extra["km_growth_rate"] = "sli:km_growth_rate:24h"
+            extra["confidence_calibration"] = "sli:confidence_calibration:1h"
+        elif event_type == "llm_hallucination":
+            extra["confidence_calibration"] = "sli:confidence_calibration:1h"
+        elif event_type == "governance_slo_data_gap":
+            extra["confidence_calibration"] = "sli:confidence_calibration:1h"
+            extra["km_growth_rate"] = "sli:km_growth_rate:24h"
+
+        return {**base_queries, **extra}
+
+
+# =============================================================================
+# Singleton
+# =============================================================================
+
+_adapter_instance: DecisionFusionAdapter | None = None
+
+
+def get_decision_fusion_adapter() -> DecisionFusionAdapter:
+    """取得 DecisionFusionAdapter 單例（lazy init）。"""
+    global _adapter_instance
+    if _adapter_instance is None:
+        _adapter_instance = DecisionFusionAdapter()
+    return _adapter_instance
+
+
+def reset_decision_fusion_adapter() -> None:
+    """重置 singleton（測試用）。"""
+    global _adapter_instance
+    _adapter_instance = None

apps/api/src/services/governance_dispatcher.py

@@ -0,0 +1,304 @@
+"""
+GovernanceDispatcher — 治理事件 → 修復派遣
+============================================
+Poll 模式：每 30s 掃 ai_governance_events 中 resolved=False 且
+無活躍 dispatch 的事件，呼叫 DecisionFusionAdapter 三維融合後
+寫入 governance_remediation_dispatch 表。
+
+職責：
+1. Poll unresolved 治理事件（不直接修改 ai_governance_events 表）
+2. 呼叫 DecisionFusionAdapter.fuse_decision → FusedDecision
+3. 依 decision_path 決定是否寫入 dispatch
+4. 不執行 remediation（實際執行由 approval_execution / auto_repair 消費 dispatch 表）
+
+Tier 3 鐵線（絕不觸碰）：
+- decision_manager.py / learning_service.py / trust_engine.py
+- 本模組透過 DecisionFusionAdapter（wrapper）間接使用這些能力
+
+2026-05-03 ogt + Claude Sonnet 4.6（亞太）: GovernanceDispatcher Wave 2E 實作
+"""
+
+from __future__ import annotations
+
+import asyncio
+from typing import Any
+
+import structlog
+from sqlalchemy import select
+
+from src.db.base import get_db_context
+from src.db.models import AiGovernanceEvent
+from src.repositories.governance_remediation_dispatch_repo import (
+    DispatchAlreadyActive,
+    create_dispatch,
+    get_active_for_event,
+)
+from src.services.decision_fusion_adapter import FusedDecision, get_decision_fusion_adapter
+
+logger = structlog.get_logger(__name__)
+
+# =============================================================================
+# 常數
+# TODO: 移到 settings（ADR-P2E-FUTURE），目前暫時 hardcode
+# =============================================================================
+
+# Poll 間隔（秒）
+# TODO: 移到 settings，允許運維不重啟調整 poll 間隔
+_DISPATCHER_INTERVAL_SEC: int = 30
+
+# 每輪最多處理幾個事件（避免單輪阻塞過長）
+_MAX_EVENTS_PER_CYCLE: int = 10
+
+# 允許建立 dispatch 的 event_type（對齊 governance_event_type enum）
+_DISPATCHABLE_EVENT_TYPES: frozenset[str] = frozenset({
+    "trust_drift",
+    "knowledge_degradation",
+    "llm_hallucination",
+    "execution_blast_radius",
+    "governance_slo_data_gap",
+})
+
+
+# =============================================================================
+# 核心函數
+# =============================================================================
+
+async def dispatch_governance_event(event: AiGovernanceEvent) -> str | None:
+    """處理單一治理事件：決策融合 → 寫 dispatch 記錄。
+
+    Args:
+        event: AiGovernanceEvent ORM 物件（唯讀，不修改）
+
+    Returns:
+        建立的 dispatch_id（str），或 None（skip / 已有活躍 dispatch）
+    """
+    event_id = event.id
+    event_type = event.event_type
+
+    # Step 1: 檢查是否已有活躍 dispatch（冪等保護）
+    existing = await get_active_for_event(event_id)
+    if existing is not None:
+        logger.debug(
+            "governance_dispatch_skipped_already_active",
+            event_id=event_id,
+            event_type=event_type,
+            existing_dispatch_id=existing.id,
+            existing_status=existing.dispatch_status,
+        )
+        return None
+
+    # Step 2: 決策融合（三維：LLM × Playbook × MCP）
+    adapter = get_decision_fusion_adapter()
+    try:
+        decision: FusedDecision = await adapter.fuse_decision(event)
+    except Exception as exc:
+        logger.warning(
+            "governance_fusion_failed",
+            event_id=event_id,
+            event_type=event_type,
+            error=str(exc),
+        )
+        # LLM 失敗 fallback：skip + log（不寫 dispatch）
+        logger.info(
+            "governance_dispatch_fallback_skip",
+            event_id=event_id,
+            reason="fusion_exception",
+        )
+        return None
+
+    # Step 3: 依 decision_path 決定要不要寫 dispatch
+    if decision.decision_path == "skip":
+        logger.info(
+            "governance_dispatch_path_skip",
+            event_id=event_id,
+            event_type=event_type,
+            confidence=round(decision.confidence, 4),
+        )
+        return None
+
+    # Step 4: 決定 executor_type 與 dispatch_status
+    # auto_dispatch → dispatched（下游 auto_repair 消費）
+    # pending_approval → pending（等人工審核）
+    if decision.decision_path == "auto_dispatch":
+        executor_type = "playbook_executor"
+        initial_status_note = "auto_dispatch"
+    else:  # pending_approval
+        executor_type = "manual"
+        initial_status_note = "pending_approval"
+
+    # Step 5: 建構 decision_context JSONB（完整三維快照）
+    decision_context = _build_decision_context(event, decision)
+
+    # Step 6: 寫入 governance_remediation_dispatch（用 repo 函數）
+    try:
+        dispatch_row = await create_dispatch(
+            event_id=event_id,
+            event_type=event_type,
+            executor_type=executor_type,
+            playbook_id=decision.matched_playbook_id,
+            decision_context=decision_context,
+            created_by="governance_dispatcher",
+        )
+    except DispatchAlreadyActive:
+        # 並行 race condition：另一個 worker 先建立了 dispatch
+        logger.info(
+            "governance_dispatch_race_condition",
+            event_id=event_id,
+            event_type=event_type,
+        )
+        return None
+    except Exception as exc:
+        logger.warning(
+            "governance_dispatch_create_failed",
+            event_id=event_id,
+            event_type=event_type,
+            error=str(exc),
+        )
+        return None
+
+    logger.info(
+        "governance_dispatched",
+        dispatch_id=dispatch_row.id,
+        event_id=event_id,
+        event_type=event_type,
+        decision_path=decision.decision_path,
+        confidence=round(decision.confidence, 4),
+        executor_type=executor_type,
+        playbook_id=decision.matched_playbook_id,
+    )
+
+    return dispatch_row.id
+
+
+async def _poll_unresolved_events() -> list[AiGovernanceEvent]:
+    """查詢 unresolved 且 event_type 在 dispatchable 範圍內的治理事件。
+
+    Returns:
+        最多 _MAX_EVENTS_PER_CYCLE 筆 AiGovernanceEvent ORM 物件列表
+    """
+    async with get_db_context() as db:
+        result = await db.execute(
+            select(AiGovernanceEvent)
+            .where(AiGovernanceEvent.resolved.is_(False))
+            .where(AiGovernanceEvent.event_type.in_(list(_DISPATCHABLE_EVENT_TYPES)))
+            .order_by(AiGovernanceEvent.triggered_at.asc())
+            .limit(_MAX_EVENTS_PER_CYCLE)
+        )
+        rows = result.scalars().all()
+
+    return list(rows)
+
+
+def _build_decision_context(
+    event: AiGovernanceEvent,
+    decision: FusedDecision,
+) -> dict[str, Any]:
+    """建構 decision_context JSONB（完整三維輸入快照）。
+
+    規格對齊 DecisionContextV1（models/governance_dispatch.py），
+    但直接建 dict 不依賴 Pydantic model（避免引入額外依賴）。
+
+    Fields:
+        version: schema 版本（v1）
+        trigger_source: 觸發來源
+        suggested_action: AI 建議的修復動作摘要
+        fusion_scores: 三維分數詳情
+        llm_reasoning: LLM 原始輸出摘要
+        mcp_snapshot: MCP 情報快照
+        decision_path: 決策分支
+        confidence: 最終融合信心度
+    """
+    return {
+        "version": "v1",
+        "trigger_source": "governance_dispatcher",
+        "triggered_metric": event.event_type,
+        "metric_value": decision.confidence,
+        "threshold": 0.85,  # TODO: 移到 settings
+        "suggested_action": decision.recommended_action,
+        "fusion_scores": {
+            "llm_score": round(decision.llm_score, 4),
+            "playbook_score": round(decision.playbook_score, 4),
+            "mcp_score": round(decision.mcp_score, 4),
+            "confidence": round(decision.confidence, 4),
+            "weights": {"llm": 0.4, "playbook": 0.3, "mcp": 0.3},  # TODO: 移到 settings
+        },
+        "llm_reasoning": decision.llm_reasoning,
+        "mcp_snapshot": decision.mcp_snapshot,
+        "decision_path": decision.decision_path,
+        "matched_playbook_id": decision.matched_playbook_id,
+        "playbook_trust": decision.playbook_trust,
+        "affected_resources": [event.event_type],
+        "extra": {
+            "event_id": event.id,
+            "event_details_keys": list((event.details or {}).keys()),
+        },
+    }
+
+
+# =============================================================================
+# 排程迴圈（仿 run_governance_loop 模式）
+# =============================================================================
+
+async def run_governance_dispatcher_loop(
+    interval_seconds: int = _DISPATCHER_INTERVAL_SEC,
+) -> None:
+    """每 30s 掃 unresolved 事件 → dispatch。
+
+    仿照 governance_agent.run_governance_loop 模式：
+    - while True → try/except → sleep
+    - 任一事件失敗不阻塞其他事件（獨立 try/except）
+    - CancelledError 向上傳播（允許 graceful shutdown）
+
+    2026-05-03 ogt + Claude Sonnet 4.6（亞太）: Wave 2E 實作
+    """
+    logger.info(
+        "governance_dispatcher_loop_started",
+        interval_seconds=interval_seconds,
+        max_events_per_cycle=_MAX_EVENTS_PER_CYCLE,
+    )
+
+    while True:
+        try:
+            events = await _poll_unresolved_events()
+
+            if events:
+                logger.info(
+                    "governance_dispatcher_cycle_start",
+                    event_count=len(events),
+                )
+                dispatched = 0
+                skipped = 0
+                for event in events:
+                    try:
+                        result = await dispatch_governance_event(event)
+                        if result is not None:
+                            dispatched += 1
+                        else:
+                            skipped += 1
+                    except asyncio.CancelledError:
+                        raise
+                    except Exception as exc:
+                        logger.warning(
+                            "governance_dispatcher_event_error",
+                            event_id=event.id,
+                            event_type=event.event_type,
+                            error=str(exc),
+                        )
+                        skipped += 1
+
+                logger.info(
+                    "governance_dispatcher_cycle_done",
+                    total=len(events),
+                    dispatched=dispatched,
+                    skipped=skipped,
+                )
+            else:
+                logger.debug("governance_dispatcher_no_events")
+
+        except asyncio.CancelledError:
+            logger.info("governance_dispatcher_loop_cancelled")
+            raise
+        except Exception as exc:
+            logger.warning("governance_dispatcher_loop_error", error=str(exc))
+
+        await asyncio.sleep(interval_seconds)

apps/api/src/services/governance_query_service.py

@@ -0,0 +1,384 @@
+"""
+Governance Query Service — /governance 頁面 DB 查詢邏輯
+======================================================
+封裝 3 個 governance endpoint 的資料庫查詢。
+Router 層禁直接存取 DB（leWOOOgo 積木化鐵律）。
+
+函式清單:
+  query_governance_events(...)  → GovernanceEventsResponse
+  query_governance_queue(...)   → GovernanceQueueResponse
+  query_governance_summary(...) → GovernanceSummaryResponse
+
+Graceful fallback 規則:
+  queue endpoint — governance_remediation_dispatch 表可能尚未建立（Track D 進行中）。
+  捕捉 sqlalchemy.exc.ProgrammingError（表不存在）後回傳 table_pending=True 的空列表，
+  確保 API 在表建立前不拋 500。
+
+2026-05-02 ogt + Claude Sonnet 4.6 Asia/Taipei
+"""
+
+from __future__ import annotations
+
+from datetime import datetime, timedelta, timezone
+
+import structlog
+from sqlalchemy import func, select, text
+from sqlalchemy.exc import ProgrammingError
+
+from src.db.base import get_db_context
+from src.db.models import AiGovernanceEvent
+from src.models.governance import (
+    DailyCount,
+    DispatchItem,
+    GovernanceEvent,
+    GovernanceEventsResponse,
+    GovernanceQueueResponse,
+    GovernanceSummaryResponse,
+    map_severity,
+)
+from src.utils.timezone import now_taipei
+
+logger = structlog.get_logger(__name__)
+
+# =============================================================================
+# 常數
+# =============================================================================
+
+_TAIPEI = timezone(timedelta(hours=8))
+
+
+# =============================================================================
+# helpers
+# =============================================================================
+
+def _extract_impact(details: dict) -> str:
+    """
+    從 details 抽摘要字串，≤80 字。
+
+    優先讀 details["impact"]（dict），取 status + 主要 metric 欄位。
+    fallback 到 details 頂層常見欄位。
+    """
+    impact_block = details.get("impact")
+    if isinstance(impact_block, dict):
+        parts: list[str] = []
+        if "status" in impact_block:
+            parts.append(str(impact_block["status"]))
+        # 主要 metric 欄位優先順序
+        for key in ("metric", "value", "rate", "ratio", "score", "count"):
+            if key in impact_block:
+                parts.append(f"{key}={impact_block[key]}")
+                break
+        summary = " ".join(parts)
+        return summary[:80] if summary else ""
+
+    # fallback: 頂層常見欄位
+    for key in ("message", "reason", "summary", "description"):
+        val = details.get(key)
+        if isinstance(val, str) and val:
+            return val[:80]
+
+    # 最後 fallback: 把 details 第一個 string value 截取
+    for val in details.values():
+        if isinstance(val, str) and val:
+            return val[:80]
+
+    return ""
+
+
+def _to_governance_event(row: AiGovernanceEvent) -> GovernanceEvent:
+    details = row.details or {}
+    return GovernanceEvent(
+        id=row.id,
+        event_type=row.event_type,
+        severity=map_severity(row.event_type),
+        triggered_at=row.triggered_at,
+        resolved=row.resolved,
+        resolved_at=row.resolved_at,
+        impact=_extract_impact(details),
+        details=details,
+        remediation=details.get("remediation"),
+        dispatch_ids=details.get("dispatch_ids", []),
+    )
+
+
+# =============================================================================
+# Endpoint 1: events
+# =============================================================================
+
+async def query_governance_events(
+    *,
+    event_types: list[str] | None = None,
+    from_dt: datetime | None = None,
+    to_dt: datetime | None = None,
+    status: str | None = None,         # "resolved" | "unresolved"
+    severity: str | None = None,       # "critical" | "warning" | "info"
+    page: int = 1,
+    size: int = 20,
+) -> GovernanceEventsResponse:
+    """
+    查詢 ai_governance_events 表，支援多維度過濾與分頁。
+
+    severity 過濾在 Python 層完成（event_type 映射）；
+    其他過濾在 SQL 層完成（效能優先）。
+    """
+    async with get_db_context() as db:
+        stmt = select(AiGovernanceEvent)
+
+        if event_types:
+            stmt = stmt.where(AiGovernanceEvent.event_type.in_(event_types))
+
+        if from_dt is not None:
+            stmt = stmt.where(AiGovernanceEvent.triggered_at >= from_dt)
+
+        if to_dt is not None:
+            stmt = stmt.where(AiGovernanceEvent.triggered_at <= to_dt)
+
+        if status == "resolved":
+            stmt = stmt.where(AiGovernanceEvent.resolved.is_(True))
+        elif status == "unresolved":
+            stmt = stmt.where(AiGovernanceEvent.resolved.is_(False))
+
+        stmt = stmt.order_by(AiGovernanceEvent.triggered_at.desc())
+
+        # 取全部結果，severity 在 Python 層過濾（避免 DB 不認識 mapping 邏輯）
+        result = await db.execute(stmt)
+        all_rows = result.scalars().all()
+
+    events = [_to_governance_event(r) for r in all_rows]
+
+    # severity 過濾（Python 層）
+    if severity:
+        from src.models.governance import _CRITICAL_TYPES, _WARNING_TYPES
+
+        if severity == "critical":
+            events = [e for e in events if e.event_type in _CRITICAL_TYPES]
+        elif severity == "warning":
+            events = [e for e in events if e.event_type in _WARNING_TYPES]
+        elif severity == "info":
+            events = [
+                e for e in events
+                if e.event_type not in _CRITICAL_TYPES and e.event_type not in _WARNING_TYPES
+            ]
+
+    total = len(events)
+    offset = (page - 1) * size
+    page_items = events[offset: offset + size]
+
+    return GovernanceEventsResponse(
+        items=page_items,
+        total=total,
+        page=page,
+        size=size,
+    )
+
+
+# =============================================================================
+# Endpoint 2: queue
+# =============================================================================
+
+async def query_governance_queue(
+    *,
+    dispatch_status: str = "pending",
+    page: int = 1,
+    size: int = 20,
+) -> GovernanceQueueResponse:
+    """
+    查詢 governance_remediation_dispatch 表。
+
+    Track D 進行中，表可能尚未建立。
+    捕捉 ProgrammingError → 回傳 table_pending=True 的空 response。
+
+    proposed_action 從 decision_context JSONB 抽取（Track D 完成後可改為真實 join）。
+    """
+    try:
+        return await _query_dispatch_table(
+            dispatch_status=dispatch_status,
+            page=page,
+            size=size,
+        )
+    except ProgrammingError as exc:
+        logger.warning(
+            "governance_dispatch_table_not_ready",
+            error=str(exc),
+        )
+        return GovernanceQueueResponse(
+            items=[],
+            total=0,
+            page=page,
+            size=size,
+            table_pending=True,
+        )
+    except ImportError as exc:
+        logger.warning(
+            "governance_dispatch_model_not_ready",
+            error=str(exc),
+        )
+        return GovernanceQueueResponse(
+            items=[],
+            total=0,
+            page=page,
+            size=size,
+            table_pending=True,
+        )
+
+
+async def _query_dispatch_table(
+    *,
+    dispatch_status: str,
+    page: int,
+    size: int,
+) -> GovernanceQueueResponse:
+    """實際查詢 governance_remediation_dispatch 表（不含 graceful fallback）."""
+    # 動態 import：Track D 完成前 ORM class 可能不存在
+    # 使用 raw SQL 降低 ORM 模型缺失的耦合風險
+    sql = text("""
+        SELECT
+            d.id,
+            d.governance_event_id,
+            e.event_type,
+            d.dispatch_status,
+            d.decision_context,
+            d.playbook_id,
+            d.created_at,
+            d.dispatched_at,
+            d.completed_at,
+            d.operator_note
+        FROM governance_remediation_dispatch d
+        JOIN ai_governance_events e ON e.id = d.governance_event_id
+        WHERE d.dispatch_status = :dispatch_status
+        ORDER BY d.created_at DESC
+    """)
+
+    count_sql = text("""
+        SELECT count(*) AS cnt
+        FROM governance_remediation_dispatch
+        WHERE dispatch_status = :dispatch_status
+    """)
+
+    async with get_db_context() as db:
+        count_row = await db.execute(count_sql, {"dispatch_status": dispatch_status})
+        total = int(count_row.scalar_one_or_none() or 0)
+
+        rows = await db.execute(
+            sql.bindparams(dispatch_status=dispatch_status),
+        )
+        all_rows = rows.fetchall()
+
+    offset = (page - 1) * size
+    page_rows = all_rows[offset: offset + size]
+
+    items: list[DispatchItem] = []
+    for row in page_rows:
+        decision_ctx: dict = (row.decision_context or {}) if hasattr(row, "decision_context") else {}
+        proposed_action = _extract_proposed_action(decision_ctx)
+
+        # playbook_trust: Track D 完成後改為 JOIN playbooks 表取 trust_score
+        # 現階段從 decision_context 取 mock 值
+        playbook_trust_raw = decision_ctx.get("playbook_trust")
+        try:
+            playbook_trust = float(playbook_trust_raw) if playbook_trust_raw is not None else None
+        except (TypeError, ValueError):
+            playbook_trust = None
+
+        items.append(DispatchItem(
+            id=str(row.id),
+            governance_event_id=str(row.governance_event_id),
+            event_type=str(row.event_type),
+            dispatch_status=str(row.dispatch_status),
+            proposed_action=proposed_action,
+            playbook_id=str(row.playbook_id) if row.playbook_id else None,
+            playbook_trust=playbook_trust,
+            created_at=row.created_at,
+            dispatched_at=row.dispatched_at,
+            completed_at=row.completed_at,
+            operator_note=row.operator_note,
+        ))
+
+    return GovernanceQueueResponse(
+        items=items,
+        total=total,
+        page=page,
+        size=size,
+        table_pending=False,
+    )
+
+
+def _extract_proposed_action(decision_ctx: dict) -> str:
+    """
+    從 decision_context JSONB 抽取 proposed_action，≤120 字。
+
+    Track D 完成後此函式可改為從真實欄位讀取。
+    """
+    for key in ("proposed_action", "action", "suggestion", "description", "summary"):
+        val = decision_ctx.get(key)
+        if isinstance(val, str) and val:
+            return val[:120]
+    return "（待補充）"
+
+
+# =============================================================================
+# Endpoint 3: summary
+# =============================================================================
+
+async def query_governance_summary(*, days: int = 30) -> GovernanceSummaryResponse:
+    """
+    過去 N 天 SLO 違反時序統計 + compliance_rate。
+
+    compliance_rate = 1 - unresolved / total（total=0 時回 1.0）
+    """
+    since = now_taipei() - timedelta(days=days)
+
+    async with get_db_context() as db:
+        # 總數 & 未解決數
+        count_stmt = select(
+            func.count().label("total"),
+            func.count().filter(AiGovernanceEvent.resolved.is_(False)).label("unresolved"),
+        ).where(AiGovernanceEvent.triggered_at >= since)
+
+        count_row = await db.execute(count_stmt)
+        counts = count_row.one()
+        total_events = int(counts.total)
+        unresolved_count = int(counts.unresolved)
+
+        # 每日計數（DATE_TRUNC 在 Postgres 端執行）
+        daily_sql = text("""
+            SELECT
+                DATE_TRUNC('day', triggered_at AT TIME ZONE 'Asia/Taipei')::date AS day,
+                event_type,
+                count(*) AS cnt
+            FROM ai_governance_events
+            WHERE triggered_at >= :since
+            GROUP BY day, event_type
+            ORDER BY day ASC
+        """)
+        daily_result = await db.execute(daily_sql, {"since": since})
+        daily_rows = daily_result.fetchall()
+
+    # 彙整每日資料
+    daily_map: dict[str, dict[str, int]] = {}
+    for row in daily_rows:
+        day_str = row.day.strftime("%Y-%m-%d") if hasattr(row.day, "strftime") else str(row.day)
+        if day_str not in daily_map:
+            daily_map[day_str] = {}
+        daily_map[day_str][row.event_type] = int(row.cnt)
+
+    daily_counts = [
+        DailyCount(
+            date=day_str,
+            total=sum(by_type.values()),
+            by_type=by_type,
+        )
+        for day_str, by_type in sorted(daily_map.items())
+    ]
+
+    if total_events == 0:
+        compliance_rate = 1.0
+    else:
+        compliance_rate = round(1.0 - unresolved_count / total_events, 4)
+
+    return GovernanceSummaryResponse(
+        compliance_rate=compliance_rate,
+        total_events=total_events,
+        unresolved_count=unresolved_count,
+        daily_counts=daily_counts,
+    )

apps/api/src/services/trust_drift_detector.py

@@ -1,7 +1,16 @@
 """
 AWOOOI AIOps Phase 6 — Trust Drift Detector（信任度漂移偵測器）
 ===============================================================
-職責：偵測 Playbook trust_score 分布的兩種極端偏態：
+【LIB ONLY — NO SIDE EFFECTS】
+
+2026-05-02 ogt + Claude Sonnet 4.6（亞太）: 整併雙寫路徑
+  背景：原本 watchdog W-6 呼叫 detector.run() 會直接寫 event_type=trust_drift 到
+  ai_governance_events；governance_agent.check_trust_drift() 每 1h 也寫同一 event_type。
+  造成雙寫、語義混淆，下游 consumer 無法區分 source-of-truth。
+  整併決策：governance_agent.check_trust_drift() 為唯一 source-of-truth（功能更完整：
+  含 auto-deprecate + Telegram 推送）。本模組降為純統計 lib，不再自行寫 PG。
+
+職責（整併後）：純統計 lib，偵測 Playbook trust_score 分布的兩種極端偏態：
 
   極端 A「盲目樂觀」：> 70% Playbook trust_score > 0.9
     → 可能是 PostExecutionVerifier 失效，或 RAG 資料被污染，讓所有 AI 都以為「我很棒」
@@ -11,13 +20,16 @@ AWOOOI AIOps Phase 6 — Trust Drift Detector（信任度漂移偵測器）
     → 可能是 EWMA 計算出錯，或所有執行都被誤判失敗，讓 AI 對自己完全沒信心
     → 學習機制可能卡死
 
-設計原則：
+設計原則（整併後）：
 1. 只讀 DB，不修改任何數據
-2. 違反 → 寫 trust_drift 事件到 ai_governance_events
-3. 樣本不足（< 10 個 approved Playbook）→ 跳過偵測，不告警
+2. detect() / run() 只回傳 TrustDistribution，不寫 ai_governance_events
+3. save_drift_event() 保留供呼叫方（如需要分布事件）顯式呼叫，不在 run() 內自動觸發
+4. 樣本不足（< 10 個 approved Playbook）→ 跳過偵測，不告警
+5. AI 治理事件的唯一寫入點：governance_agent.check_trust_drift()
 
 ADR-087: AI 自我治理閉環
 2026-04-15 ogt + Claude Sonnet 4.6（亞太）: Phase 6 初始建立
+2026-05-02 ogt + Claude Sonnet 4.6（亞太）: 降為 lib only，移除 run() 自動 PG 寫入
 """
 
 from __future__ import annotations
@@ -222,11 +234,14 @@ class TrustDriftDetector:
             logger.error("trust_drift_event_save_error", error=str(e))
 
     async def run(self) -> TrustDistribution:
-        """完整執行：偵測 → 如有漂移則寫事件。"""
-        dist = await self.detect()
-        if dist.drift_detected:
-            await self.save_drift_event(dist)
-        return dist
+        """統計偵測（LIB ONLY）：只回傳 TrustDistribution，不寫 ai_governance_events。
+
+        2026-05-02 ogt + Claude Sonnet 4.6（亞太）: 整併雙寫路徑
+          原行為：detect() 後若 drift_detected 自動呼叫 save_drift_event() 寫 PG。
+          改為：只回傳結果，由呼叫方決定是否寫入。
+          ai_governance_events 的唯一寫入點：governance_agent.check_trust_drift()。
+        """
+        return await self.detect()
 
 
 # ─────────────────────────────────────────────────────────────────────────────