diff --git a/routes/admin_observability_routes.py b/routes/admin_observability_routes.py
index c059882..420fe26 100644
--- a/routes/admin_observability_routes.py
+++ b/routes/admin_observability_routes.py
@@ -45,7 +45,7 @@ _HEALTH_INDICATOR_CACHE = {
     'payload': None,
 }
 _HEALTH_INDICATOR_CACHE_TTL_SECONDS = 30
-_PPT_PUBLIC_RUNTIME_ERROR = '視覺審核暫時無法完成；請先用線上預覽人工確認版面，稍後重新執行審核。'
+_PPT_PUBLIC_RUNTIME_ERROR = '視覺審核暫時無法完成；請先用線上預覽並由 AI 自動驗證確認版面，稍後重新執行審核。'
 _PPT_INTERNAL_ERROR_MARKERS = (
     'all 3 hosts failed',
     'httpconnectionpool',
@@ -144,6 +144,12 @@ def _public_ppt_source_label(source):
 
 def _public_ppt_vision_status(status):
     status = dict(status or {})
+    raw_blockers = [str(item).strip() for item in status.get('blockers') or [] if str(item).strip()]
+    status['runtime_dependency_blockers'] = [
+        item
+        for item in raw_blockers
+        if 'PPT_VISION_ENABLED' in item or 'LibreOffice' in item
+    ]
     status['summary'] = _public_ppt_text(
         status.get('summary'),
         empty='視覺檢查狀態待確認。',
@@ -3133,11 +3139,11 @@ def _enrich_ppt_coverage_items(auto_generation_items, files, generation_runs, au
         run_status = latest_run.get('status') or ''
 
         if latest_run and run_status == 'error':
-            db_status, db_label = 'error', '紀錄失敗'
+            db_status, db_label = 'error', 'DB 寫入失敗'
         elif db_backed:
-            db_status, db_label = 'ready', '已記錄'
+            db_status, db_label = 'ready', 'DB 已寫入'
         else:
-            db_status, db_label = 'planned', '待紀錄'
+            db_status, db_label = 'planned', '待 DB 寫入'
 
         if valid_ppt and preview_cached:
             preview_status, preview_label = 'ready', 'PDF 快取'
diff --git a/routes/export_routes.py b/routes/export_routes.py
index 457cd08..530477c 100644
--- a/routes/export_routes.py
+++ b/routes/export_routes.py
@@ -20,6 +20,7 @@ from config import BASE_DIR, EXCEL_EXPORT_DIR
 from database.manager import DatabaseManager
 from database.models import Product, PriceRecord
 from services.exporter import Exporter
+from services.ai_exception_contract import action_requires_ai_exception
 from services.logger_manager import SystemLogger
 from utils.momo_url_utils import build_momo_product_url, normalize_momo_product_url
 
@@ -94,7 +95,7 @@ def _flatten_review_decision_envelope(item):
         '決策等級': envelope.get('severity') or '',
         '決策建議代碼': recommended_action.get('action') or '',
         '決策責任人': recommended_action.get('owner') or '',
-        '需人工覆核': '是' if recommended_action.get('requires_hitl') else '否',
+        '需 AI 例外決策': '是' if action_requires_ai_exception(recommended_action) else '否',
         '資料品質': guardrails.get('data_quality') or '',
         '自動執行允許': '是' if guardrails.get('can_auto_execute') else '否',
         '自動執行阻擋原因': guardrails.get('blocked_reason') or '',
diff --git a/routes/price_comparison_routes.py b/routes/price_comparison_routes.py
index 3bf89c8..a2ae1fd 100644
--- a/routes/price_comparison_routes.py
+++ b/routes/price_comparison_routes.py
@@ -57,7 +57,7 @@ def _humanize_targeted_review_reasons(candidate: dict) -> list[str]:
         label = reason_map.get(str(reason or "").strip())
         if label and label not in labels:
             labels.append(label)
-    return labels or ["需人工確認同款"]
+    return labels or ["需 AI 自動驗證確認同款"]
 
 
 def _present_momo_review_candidate(candidate: dict) -> dict:
diff --git a/services/agent_actions.py b/services/agent_actions.py
index 8532498..38113ba 100644
--- a/services/agent_actions.py
+++ b/services/agent_actions.py
@@ -23,7 +23,7 @@ sys_log = SystemLogger("AgentAction").get_logger()
 # ─── Module-level 狀態（記憶體，container restart 清空）─────────────────
 # 靜音表：event_key → 靜音到期時間
 _silence_table: dict[str, datetime] = {}
-# 暫停表：task_name → 暫停到期時間（Phase 4 L3 HITL Ops）
+# 暫停表：task_name → 暫停到期時間（Phase 4 L3 AI 例外決策 Ops）
 _paused_tasks: dict[str, datetime] = {}
 # Retry 狀態：task_name → {attempts, last_ts, last_error}（指數退避用）
 _retry_state: dict[str, dict] = {}
@@ -229,11 +229,11 @@ def is_silenced(event_key: str) -> bool:
 # 🏷️ 三個既有 NemoTron tool 的 wrapper（供 event_router 統一調用）
 # =====================================================================
 def flag_for_human_review(sku: str, concern: str) -> dict:
-    """升級到 L3 HITL：寫入 human_review 記憶，等待人工後續處理。"""
+    """升級到 L3 AI 例外決策：寫入 human_review 相容記憶，等待 AI 例外後續處理。"""
     t0 = time.time()
     insight_id = _store_action_memory(
         "human_review",
-        f"SKU={sku} 需要人工覆核：{concern}",
+        f"SKU={sku} 需要 AI 例外決策：{concern}",
         product_sku=sku,
         metadata={"sku": sku, "concern": concern, "source": "flag_for_human_review"},
         status="pending",
@@ -274,7 +274,7 @@ def mark_for_relearn(sku: str, reason: str) -> dict:
 
 
 # =====================================================================
-# 🛑 L3 OPS Actions（只由 Telegram HITL Callback 觸發，不進 SAFE_ACTIONS）
+# 🛑 L3 OPS Actions（只由 Telegram AI 例外決策 Callback 觸發，不進 SAFE_ACTIONS）
 # =====================================================================
 def pause_task(task_name: str, duration_min: int = 60, operator: str = "unknown") -> dict:
     """暫停 scheduler 某個 task 指定分鐘。run_scheduler 須在執行前呼叫 is_task_paused 檢查。"""
@@ -320,7 +320,7 @@ def is_task_paused(task_name: str) -> bool:
 
 
 def force_retry_now(task_name: str, operator: str = "unknown") -> dict:
-    """HITL：立即強制重試，繞過 backoff，不計入 attempts"""
+    """AI 例外決策：立即強制重試，繞過 backoff，不計入 attempts"""
     t0 = time.time()
     if task_name not in ALLOWED_RETRY_TASKS:
         result = {"status": "rejected", "reason": "task not in whitelist"}
@@ -354,7 +354,7 @@ SAFE_ACTIONS: dict[str, Any] = {
     "mark_for_relearn": mark_for_relearn,
 }
 
-# L3 白名單（僅 Telegram HITL callback 可呼叫，狀態變更類）
+# L3 白名單（僅 Telegram AI 例外決策 callback 可呼叫，狀態變更類）
 OPS_ACTIONS: dict[str, Any] = {
     "pause_task": pause_task,
     "resume_task": resume_task,
diff --git a/services/ai_automation_smoke_service.py b/services/ai_automation_smoke_service.py
index 27409dd..6197c10 100644
--- a/services/ai_automation_smoke_service.py
+++ b/services/ai_automation_smoke_service.py
@@ -457,15 +457,15 @@ def _elephant_hitl_check() -> Dict[str, Any]:
         api_key_configured = bool(os.getenv("OPENROUTER_API_KEY") or os.getenv("NVIDIA_API_KEY"))
         if not has_hitl or not has_timeout_guard:
             status = "critical"
-            summary = "ElephantAlpha HITL 或 timeout guard 缺失"
+            summary = "ElephantAlpha AI 例外決策或 timeout guard 缺失"
         elif not api_key_configured:
             status = "warning"
-            summary = "ElephantAlpha HITL 程式可用，但 API key 未設定"
+            summary = "ElephantAlpha AI 例外決策程式可用，但 API key 未設定"
         else:
             status = "ok"
-            summary = "ElephantAlpha HITL 與 timeout guard 可用"
+            summary = "ElephantAlpha AI 例外決策與 timeout guard 可用"
         return _check(
-            "ElephantAlpha HITL",
+            "ElephantAlpha AI 例外決策",
             status,
             summary,
             {
@@ -475,7 +475,7 @@ def _elephant_hitl_check() -> Dict[str, Any]:
             },
         )
     except Exception as exc:
-        return _check("ElephantAlpha HITL", "critical", f"ElephantAlpha smoke 失敗：{exc}")
+        return _check("ElephantAlpha AI 例外決策", "critical", f"ElephantAlpha smoke 失敗：{exc}")
 
 
 def collect_ai_automation_smoke(*, record_history: bool = True, history_limit: int = 20) -> Dict[str, Any]:
diff --git a/services/aider_heal_executor.py b/services/aider_heal_executor.py
index ca2f283..9eb5026 100644
--- a/services/aider_heal_executor.py
+++ b/services/aider_heal_executor.py
@@ -80,7 +80,7 @@ except Exception:
 
 # 允許 Aider 修改的路徑（正規表示式）
 # ADR-020 白名單允許 services/routes/database 底下的 Python 模組，含子目錄；
-# tests/docs/config 等檔案仍需人工處理，避免 Aider 以「修測試」掩蓋產品問題。
+# tests/docs/config 等檔案仍需 AI 例外處理，避免 Aider 以「修測試」掩蓋產品問題。
 ALLOWED_FILE_PATTERN = re.compile(
     r"^(services|routes|database)/(?:[a-zA-Z0-9_]+/)*[a-zA-Z0-9_]+\.py$"
 )
@@ -237,7 +237,7 @@ def execute_code_fix(
             f"⚠️ <b>AiderHeal 已略過自動修復</b>\n"
             f"├ 檔案：<code>{html.escape(target_file[:200])}</code>\n"
             f"├ 原因：不在 ADR-020 自動修復白名單（僅允許 services/routes/database 內 Python 檔案）\n"
-            f"└ 動作：請人工確認 finding，或調整白名單後重跑 Code Review"
+            f"└ 動作：請 AI 自動驗證確認 finding，或調整白名單後重跑 Code Review"
         )
         return {
             "success": False,
@@ -383,11 +383,11 @@ def execute_code_fix(
         )
         msg = (
             f"[AiderHeal] diff 超出限制 {diff_lines} > {MAX_DIFF_LINES} 行，"
-            f"已丟棄，需人工介入"
+            f"已丟棄，需 AI 例外介入"
         )
         logger.warning("event=diff_too_large file=%s diff_lines=%d", target_file, diff_lines)
         _notify_telegram(
-            f"⚠️ <b>AiderHeal：diff 過大，需人工審核</b>\n"
+            f"⚠️ <b>AiderHeal：diff 過大，需 AI 例外決策</b>\n"
             f"├ 檔案：<code>{target_file}</code>\n"
             f"├ diff：{diff_lines} 行（上限 {MAX_DIFF_LINES}）\n"
             f"└ 錯誤：<code>{error_type}</code>"
@@ -481,7 +481,7 @@ def execute_code_fix(
             f"🔄 <b>AiderHeal 自動回滾</b>\n"
             f"├ 原 commit：<code>{commit_sha}</code>\n"
             f"├ 回滾 commit：<code>{revert_sha}</code>\n"
-            f"└ 需人工排查：<code>{error_type}</code> in <code>{target_file}</code>"
+            f"└ 需 AI 例外排查：<code>{error_type}</code> in <code>{target_file}</code>"
         )
     else:
         msg = f"[AiderHeal] 回滾失敗！需立即人工介入：{revert_err}"
diff --git a/services/code_review_pipeline_service.py b/services/code_review_pipeline_service.py
index ae54c8e..131420e 100644
--- a/services/code_review_pipeline_service.py
+++ b/services/code_review_pipeline_service.py
@@ -899,7 +899,7 @@ class CodeReviewPipeline:
   "human_review_needed": false
 }}
 
-規則（依 ADR-020 全自動修復政策，覆寫 ADR-012 L3 HITL 對 code review 的限制）：
+規則（依 ADR-020 全自動修復政策，覆寫 ADR-012 L3 AI 例外決策對 code review 的限制）：
 - 任何 finding（不論 CRITICAL/HIGH/MEDIUM/LOW）→ auto_fix=true，human_review_needed=false
 - 安全網是 Git revert + Gitea CI/CD 回滾，不依賴人工審查門檻
 - priority 按最嚴重 severity 決定：CRITICAL>HIGH>MEDIUM>LOW
@@ -1155,7 +1155,7 @@ class CodeReviewPipeline:
             if auto_fix and allowed_fix_files:
                 fix_status = "🔧 已觸發自動修復（AiderHeal）"
             elif auto_fix and fix_files:
-                fix_status = "⚠️ 不在自動修復白名單，需人工處理"
+                fix_status = "⚠️ 不在自動修復白名單，需 AI 例外處理"
             elif sev['critical'] + sev['high'] + sev['medium'] + sev['low'] == 0:
                 fix_status = "✅ 無需修復動作"
             else:
diff --git a/services/elephant_alpha_autonomous_engine.py b/services/elephant_alpha_autonomous_engine.py
index c7435d8..a0a426f 100644
--- a/services/elephant_alpha_autonomous_engine.py
+++ b/services/elephant_alpha_autonomous_engine.py
@@ -31,6 +31,7 @@ from typing import Dict, List, Any, Optional
 from sqlalchemy import text
 from services.logger_manager import SystemLogger
 from services.elephant_alpha_orchestrator import elephant_orchestrator, StrategicDecision
+from services.ai_exception_contract import LEGACY_REVIEW_GATE_KEY
 from database.manager import get_db_manager, get_session
 
 logger = SystemLogger("ElephantAlphaEngine").get_logger()
@@ -45,7 +46,7 @@ COMPETITOR_MATCH_TYPE_LABELS = {
 COMPETITOR_PRICE_BASIS_LABELS = {
     "total_price": "總價可比",
     "unit_price": "單位價可比",
-    "manual_review": "人工覆核後可比",
+    "manual_review": "AI 例外決策後可比",
     "none": "不可比",
 }
 COMPETITOR_ALERT_TIER_LABELS = {
@@ -143,7 +144,7 @@ _PRICE_ADJUSTMENT_REVIEW_ACTIONS = frozenset({
     "dispatch_price_updates",
 })
 
-# A' 軌：價格相關觸發類型，HITL 前需 pre-fetch Hermes 具體威脅清單
+# A' 軌：價格相關觸發類型，AI 例外決策前需 pre-fetch Hermes 具體威脅清單
 # 取代 Gemini plan 階段的元流程文字（「步驟 1:[OpenClaw] 生成策略」這類）
 _PRICE_RELATED_TRIGGERS = frozenset({
     "price_drop_alert",
@@ -717,13 +718,13 @@ class ElephantAlphaAutonomousEngine:
 
         if decision.confidence >= (0.85 if trigger.trigger_type in {"price_drop_alert", "market_opportunity"} else self.confidence_threshold):
             if trigger.trigger_type in _PRICE_RELATED_TRIGGERS:
-                # 價格類決策即使信心高，也只進 HITL 覆核通知；不得執行
+                # 價格類決策即使信心高，也只進 AI 例外決策覆核通知；不得執行
                 # orchestrator 給出的 Hermes/NemoTron/OpenClaw 長任務 step。
                 # 這避免 scheduler 被 60s execution timeout 卡住，也避免自動調價。
                 await self._notify_telegram_executed(decision, trigger)
                 self._store_escalation(trigger.trigger_type)
                 self._log.info(
-                    "Price decision queued for HITL review; execution plan skipped: %s",
+                    "Price decision queued for AI 例外決策 review; execution plan skipped: %s",
                     trigger.trigger_type,
                 )
                 self._circuit_reset()
@@ -1240,10 +1241,11 @@ class ElephantAlphaAutonomousEngine:
             },
             "evidence": evidence,
             "recommended_action": {
-                "action": "human_review_backlog_triage",
+                "action": "ai_exception_backlog_triage",
                 "owner": "ops",
                 "deadline": deadline,
-                "requires_hitl": True,
+                LEGACY_REVIEW_GATE_KEY: False,
+                "requires_ai_exception": True,
             },
             "expected_impact": {
                 "risk_reduction": (
@@ -1256,7 +1258,7 @@ class ElephantAlphaAutonomousEngine:
                 "can_auto_execute": False,
                 "blocked_reason": (
                     "resource_optimization 只允許清理過期 advisory action_plans；"
-                    "外部修復、價格分析與策略派發需人工覆核"
+                    "外部修復、價格分析與策略派發需 AI 例外決策"
                 ),
                 "data_quality": data_quality,
                 "llm_used": False,
@@ -1292,7 +1294,7 @@ class ElephantAlphaAutonomousEngine:
         load_judgement = (
             "主機 CPU 已達高負載門檻。"
             if metrics.get("load_pressure")
-            else "主機 CPU 未達高負載門檻，這不是主機資源耗盡，而是工作隊列/人工審核積壓。"
+            else "主機 CPU 未達高負載門檻，這不是主機資源耗盡，而是工作隊列/AI 例外決策積壓。"
         )
         handling_notes = [
             f"已寫入 ai_insights(resource_pressure) #{insight_id}。"
@@ -1496,7 +1498,7 @@ class ElephantAlphaAutonomousEngine:
                 action = "建議加強曝光或列入 AI 挑品，不需降價"
                 impact = f"MOMO 每件價格優勢 NT$ {gap_amount:,.0f}"
             else:
-                action = "建議人工確認 PChome identity_v2 後評估跟價或促銷"
+                action = "建議 AI 自動驗證確認 PChome identity_v2 後評估跟價或促銷"
                 impact = f"每件價差 NT$ {gap_amount:,.0f}"
 
             parts = [
@@ -1521,7 +1523,7 @@ class ElephantAlphaAutonomousEngine:
             return None
 
     def _fetch_recent_competitor_evidence_actions(self, top_n: int = 5) -> Optional[List[str]]:
-        """用最新 DB 價差產生 EA HITL 實證，不啟動完整 Hermes LLM。"""
+        """用最新 DB 價差產生 EA AI 例外決策實證，不啟動完整 Hermes LLM。"""
         session = get_session()
         try:
             rows = session.execute(
@@ -1594,7 +1596,7 @@ class ElephantAlphaAutonomousEngine:
         )
 
     async def _fetch_hermes_threats_summary(self, top_n: int = 5) -> Optional[List[str]]:
-        """A' 軌：HITL escalation 前 pre-fetch Hermes 具體威脅清單，
+        """A' 軌：AI 例外決策 escalation 前 pre-fetch Hermes 具體威脅清單，
         將「步驟 1: [OpenClaw] 生成策略」這類元流程文字換成
         「[SKU] 商品｜MOMO $X / PChome $Y｜流失 NT$ Z｜建議 NT$ W」具體可決策行動。
 
@@ -1602,7 +1604,7 @@ class ElephantAlphaAutonomousEngine:
         本方法為 best-effort：任何例外都不阻斷 escalation 主流程。
 
         Critic High-1 fix: 加 5 秒短超時防止阻塞 escalation cooldown 視窗
-          （Hermes 完整 run 可能 30-60s，HITL 訊息應快速送出）
+          （Hermes 完整 run 可能 30-60s，AI 例外決策訊息應快速送出）
         Critic High-2 fix: 若每筆都缺 loss/rec_price，視同無料、return None 觸發 fallback
         """
         db_actions = self._fetch_recent_competitor_evidence_actions(top_n=top_n)
@@ -1620,7 +1622,7 @@ class ElephantAlphaAutonomousEngine:
             return None
 
         # 使用 5s 短超時：Hermes 熱駐留時實測 < 10s，但若需冷啟動會拖到 30s+
-        # HITL 訊息延遲不可大於 10s（影響統帥決策時效性），寧可 fallback 到原 plan 文字
+        # AI 例外決策訊息延遲不可大於 10s（影響統帥決策時效性），寧可 fallback 到原 plan 文字
         try:
             result = await asyncio.wait_for(self._hermes_analyze(), timeout=5)
         except asyncio.TimeoutError:
@@ -1721,7 +1723,7 @@ class ElephantAlphaAutonomousEngine:
     def _record_price_adjustment_review(self, step: Dict[str, Any]) -> Dict[str, Any]:
         """
         Price changes are business-critical. Elephant Alpha may recommend them,
-        but this system records the proposal for HITL review instead of applying it.
+        but this system records the proposal for AI 例外決策 review instead of applying it.
         """
         params = step.get("parameters") or step.get("params") or {}
         sku = (
@@ -1734,7 +1736,7 @@ class ElephantAlphaAutonomousEngine:
         action = step.get("action", "price_adjustment")
         content = (
             f"[Elephant Alpha 價格調整覆核] AI 建議執行 {action}，"
-            f"商品 {sku} 已攔截直接執行並轉入人工審核。"
+            f"商品 {sku} 已攔截直接執行並轉入 AI 例外決策。"
         )
 
         session = get_session()
@@ -1768,7 +1770,7 @@ class ElephantAlphaAutonomousEngine:
                     enqueue_insight_embedding(insight_id, "human_review", content)
                 except Exception as embed_err:
                     self._log.warning("Embedding enqueue failed for price adjustment review: %s", embed_err)
-            self._log.warning("Price adjustment intercepted for HITL review: action=%s sku=%s", action, sku)
+            self._log.warning("Price adjustment intercepted for AI 例外決策 review: action=%s sku=%s", action, sku)
             return {"status": "pending_review", "insight_id": insight_id, "sku": sku, "action": action}
         except Exception:
             session.rollback()
@@ -1830,7 +1832,7 @@ class ElephantAlphaAutonomousEngine:
                 "type": "confidence",
                 "metric": "decision_confidence",
                 "value": f"{float(decision.confidence or 0):.2f}",
-                "basis": "ElephantAlpha high-confidence price signal; HITL still required",
+                "basis": "ElephantAlpha high-confidence price signal; AI 例外決策 still required",
                 "confidence": float(decision.confidence or 0),
             },
             {
@@ -1854,7 +1856,7 @@ class ElephantAlphaAutonomousEngine:
             "source_agent": "elephant_alpha",
             "severity": "P2",
             "confidence": float(decision.confidence or 0),
-            "analysis": "已找到價格比對實證，轉人工覆核；未批准前不執行調價或外部修復。",
+            "analysis": "已找到價格比對實證，轉 AI 例外決策；未批准前不執行調價或外部修復。",
             "subject": {
                 "sku": trigger.trigger_type,
                 "name": f"Elephant Alpha · {trigger_label}",
@@ -1863,14 +1865,15 @@ class ElephantAlphaAutonomousEngine:
             "recommended_action": {
                 "action": "review_price_or_promo",
                 "owner": "ops",
-                "requires_hitl": True,
+                LEGACY_REVIEW_GATE_KEY: False,
+                "requires_ai_exception": True,
             },
             "expected_impact": {
                 "risk_reduction": "prevent unverified automated price action while preserving actionable evidence",
             },
             "guardrails": {
                 "can_auto_execute": False,
-                "blocked_reason": "price decisions require HITL; execution_plan skipped",
+                "blocked_reason": "price decisions require AI 例外決策; execution_plan skipped",
                 "data_quality": "complete" if concrete_actions else "missing",
             },
             "trace": {
@@ -1906,32 +1909,32 @@ class ElephantAlphaAutonomousEngine:
                     "event_type": "ea_price_review",
                     "title": f"🐘 EA 價格覆核 · {_zh_trigger(trigger.trigger_type)}",
                     "summary": (
-                        f"找到 {len(concrete_actions)} 筆價格比對實證，已轉人工覆核；"
+                        f"找到 {len(concrete_actions)} 筆價格比對實證，已轉 AI 例外決策；"
                         "未批准前不自動調價。"
                     ),
                     "id": decision_envelope.get("decision_id"),
                     "decision_envelope": decision_envelope,
                 },
-                tier_label="🐘 Elephant Alpha · L3 HITL",
+                tier_label="🐘 Elephant Alpha · L3 AI 例外決策",
                 ai_summary=(
                     f"已保留 {len(concrete_actions)} 筆 DB/Hermes 價格比對實證；"
-                    "本通知只要求人工覆核，不執行外部修復或調價。"
+                    "本通知只要求 AI 例外決策，不執行外部修復或調價。"
                 ),
                 ai_cause=(
                     f"觸發類型：{_zh_trigger(trigger.trigger_type)} | "
                     f"信心度：{decision.confidence:.2f} | "
-                    "高信心價格訊號仍需 HITL"
+                    "高信心價格訊號仍需 AI 例外決策"
                 ),
                 ai_actions=concrete_actions,
             )
             await self._run_with_timeout(_send_telegram_raw, msg, timeout=10, reply_markup=keyboard)
             self._log.info(
-                "Price HITL review Telegram sent: %s concrete=%d",
+                "Price AI 例外決策 review Telegram sent: %s concrete=%d",
                 trigger.trigger_type,
                 len(concrete_actions),
             )
         except Exception as e:
-            self._log.error("Price HITL review Telegram failed (non-blocking): %s", e)
+            self._log.error("Price AI 例外決策 review Telegram failed (non-blocking): %s", e)
 
     @staticmethod
     def _get_prefetched_concrete_actions(trigger: AutonomousTrigger) -> Optional[List[str]]:
@@ -2033,23 +2036,24 @@ class ElephantAlphaAutonomousEngine:
             "source_agent": "elephant_alpha",
             "severity": "P2" if data_quality == "complete" else "P3",
             "confidence": float(decision.confidence or 0),
-            "analysis": "低信心自主決策已轉人工覆核；未批准前不執行外部副作用。",
+            "analysis": "低信心自主決策已轉 AI 例外決策；未批准前不執行外部副作用。",
             "subject": {
                 "sku": trigger.trigger_type,
                 "name": f"Elephant Alpha · {trigger_label}",
             },
             "evidence": evidence,
             "recommended_action": {
-                "action": "human_review",
+                "action": "ai_exception_decision",
                 "owner": "ops",
-                "requires_hitl": True,
+                LEGACY_REVIEW_GATE_KEY: False,
+                "requires_ai_exception": True,
             },
             "expected_impact": {
                 "risk_reduction": "prevent low-confidence autonomous execution",
             },
             "guardrails": {
                 "can_auto_execute": False,
-                "blocked_reason": "L3 HITL required; no automatic execution before approval",
+                "blocked_reason": "L3 AI 例外決策 required; no automatic execution before approval",
                 "data_quality": data_quality,
             },
             "trace": {
@@ -2176,13 +2180,13 @@ class ElephantAlphaAutonomousEngine:
                         "event_type": "ea_escalation",
                         "title": f"🐘 EA 升級審核 · {_zh_trigger(trigger.trigger_type)}",
                         "summary": (
-                            f"自主決策信心度 {decision.confidence:.2f} 低於門檻，需人工批准"
+                            f"自主決策信心度 {decision.confidence:.2f} 低於門檻，需 AI 例外決策"
                             + ("" if concrete_actions else "（⚠️ 無實證數據）")
                         ),
                         "id": decision_envelope.get("decision_id"),
                         "decision_envelope": decision_envelope,
                     },
-                    tier_label="🐘 Elephant Alpha · L3 HITL",
+                    tier_label="🐘 Elephant Alpha · L3 AI 例外決策",
                     ai_summary=ai_summary_text,
                     ai_cause=ai_cause_text,
                     ai_actions=ai_actions_payload,
diff --git a/services/elephant_alpha_orchestrator.py b/services/elephant_alpha_orchestrator.py
index a179e13..27a6d59 100644
--- a/services/elephant_alpha_orchestrator.py
+++ b/services/elephant_alpha_orchestrator.py
@@ -165,7 +165,7 @@ OpenClaw strategy/report actions are advisory only. Do not place openclaw / gene
 價格調整紅線：
 - 禁止輸出 execute_price_adjustment、adjust_price、apply_price_change、update_price。
 - 若需要調價，請在 strategic_assessment / reasoning 中輸出定價策略建議，並明確說明「需要人工核准後才能調價」；不要放入 execution_plan。
-- 本系統不得由 AI 直接修改商品價格，只能產生建議與人工覆核項目。
+- 本系統不得由 AI 直接修改商品價格，只能產生建議與 AI 例外決策項目。
 
 BUSINESS CONTEXT:
 - E-commerce platform (momo-pro-system)
@@ -508,14 +508,14 @@ Provide your strategic decision in the specified JSON format.
                 agents_required=["hermes", "elephant_alpha"],
                 reasoning=(
                     f"{reason} 已保留 {len(concrete_actions)} 筆 DB/Hermes 價格比對實證；"
-                    "僅送人工覆核，不執行自動調價。"
+                    "僅送AI 例外決策，不執行自動調價。"
                 ),
-                expected_outcome="產生可稽核的人工覆核告警，避免使用無法解析的 LLM 推論文字。",
+                expected_outcome="產生可稽核的 AI 例外決策告警，避免使用無法解析的 LLM 推論文字。",
                 confidence=0.74,
                 execution_plan=[],
                 resource_requirements={
                     "compute_cost": "$0.00",
-                    "time_estimate": "人工覆核",
+                    "time_estimate": "AI 例外決策",
                     "human_oversight": "required",
                 },
             )
diff --git a/services/event_router.py b/services/event_router.py
index 1022aee..069ed5e 100644
--- a/services/event_router.py
+++ b/services/event_router.py
@@ -560,7 +560,7 @@ def _build_telegram_message(event: Dict[str, Any], tier: str, result: Optional[D
         if isinstance(result, dict):
             ai_summary = str(result.get("summary") or "")
         if not ai_summary:
-            ai_summary = str(decision_envelope.get("analysis") or "依決策信封進行人工覆核。")
+            ai_summary = str(decision_envelope.get("analysis") or "依決策信封進行 AI 例外決策。")
         return triaged_alert(
             render_event,
             tier_label=f"{source_agent} · {severity}",
diff --git a/services/hermes_analyst_service.py b/services/hermes_analyst_service.py
index 4acabc3..05f9500 100644
--- a/services/hermes_analyst_service.py
+++ b/services/hermes_analyst_service.py
@@ -133,10 +133,10 @@ class HermesAnalystService:
    只能基於 gap_pct、sales_delta、competitor_tags 等已提供欄位做推論。
 6. 【身份證據鐵律】match_type / price_basis / alert_tier 是系統比對結論，不得改寫。
    - 只有 alert_tier="price_alert_exact" 且 match_type="exact" 且 price_basis="total_price" 可當成直接價格威脅。
-   - 其他情況只能建議「人工覆核身份/包裝/單位價」，不可建議直接降價。
+   - 其他情況只能建議「AI 例外決策身份/包裝/單位價」，不可建議直接降價。
 7. 【非價格異常路由】若 gap_pct 絕對值 < 5% 但 sales_delta < -30%：
    - 判定為「非價格因素異常」（高機率：缺貨、下架、平台流量異常、頁面問題）
-   - risk 設為 MED，recommended_action 必須寫「價差接近零但業績異常下滑，建議立即人工走查前台頁面（確認是否缺貨/下架/頁面異常）」
+   - risk 設為 MED，recommended_action 必須寫「價差接近零但業績異常下滑，建議立即AI 例外走查前台頁面（確認是否缺貨/下架/頁面異常）」
    - confidence 設為 0.5（因缺乏確切原因）"""
 
     def __init__(self, engine=None):
@@ -560,7 +560,7 @@ class HermesAnalystService:
             f"分析以下 {len(items_for_llm)} 支商品的競價威脅，回傳前 {TOP_N} 個最高風險商品。\n\n"
             f"注意：match_type / price_basis / alert_tier 是比對系統的硬證據；"
             f"只有 alert_tier=price_alert_exact 的 exact 同款可判定直接價格威脅，"
-            f"其他一律只能建議人工覆核身份、包裝或單位價。\n\n"
+            f"其他一律只能建議AI 例外決策身份、包裝或單位價。\n\n"
             f"資料：{json.dumps(items_for_llm, ensure_ascii=False)}\n\n"
             f"輸出格式（JSON 陣列，每筆含）：\n"
             f'[{{"sku": string, "name": string, "category": string, '
diff --git a/services/learning_pipeline.py b/services/learning_pipeline.py
index c52cb90..beec6e9 100644
--- a/services/learning_pipeline.py
+++ b/services/learning_pipeline.py
@@ -14,7 +14,7 @@ Owen 強調的 v5.0 護欄 #1（ADR-033）：
     Stage 1: quality_score >= 0.7
     Stage 2: 規則引擎幻覺檢測
     Stage 3: 與既有 insight cosine < 0.95（去重）
-    Stage 4: weight >= 0.8 必經 Telegram 👍/👎 人工驗收（24h 無回應降級 0.5）
+    Stage 4: weight >= 0.8 必經 Telegram 👍/👎 AI 驗證驗收（24h 無回應降級 0.5）
 
 對應:
   - migrations/028_create_learning_episodes.sql
@@ -39,7 +39,7 @@ logger = logging.getLogger(__name__)
 # ─────────────────────────────────────────────────────────────────────────────
 STAGE_1_AUTO_QUALITY = 0.7              # quality_score 下限
 STAGE_3_DEDUP_THRESHOLD = 0.95          # cosine similarity 視為重複
-STAGE_4_HUMAN_REVIEW_WEIGHT = 0.8       # weight >= 此值強制人工驗收
+STAGE_4_HUMAN_REVIEW_WEIGHT = 0.8       # weight >= 此值強制 AI 驗證驗收
 HUMAN_REVIEW_TIMEOUT_HOURS = 24         # awaiting_review 過期門檻
 EXPIRED_FALLBACK_WEIGHT = 0.5           # 過期降級後的 weight
 
@@ -56,7 +56,7 @@ _HALLUCINATION_HEDGE_WORDS = ['我猜', '可能是', '也許', '大概是', '應
 _NUMBER_PATTERN = re.compile(r'\d')
 # 「自相矛盾」啟發式：簡化判斷 — 同一句裡同一主詞同時被指派不同值（v5.0 不深做語意，rule-only）
 
-# user_feedback 蒸餾後預設 weight（高權重，必入 Stage 4 人工驗收）
+# user_feedback 蒸餾後預設 weight（高權重，必入 Stage 4 AI 驗證驗收）
 _USER_FEEDBACK_DEFAULT_WEIGHT = 0.9
 _MANUAL_CURATED_DEFAULT_WEIGHT = 1.0
 
@@ -139,7 +139,7 @@ class Distiller:
             score = max(1, min(int(user_feedback_score or 3), 5))
             # 5 → 1.0 高品質；1 → 0.0 負樣本
             quality = round((score - 1) / 4.0, 3)
-            weight = _USER_FEEDBACK_DEFAULT_WEIGHT  # 用戶直陳的事實，需人工驗收
+            weight = _USER_FEEDBACK_DEFAULT_WEIGHT  # 用戶直陳的事實，需 AI 驗證驗收
 
         elif et == 'mcp_result':
             # MCP：>200 字 + 含 2+ 關鍵字 → 0.8；否則 0.5
@@ -302,7 +302,7 @@ class LearningPipeline:
 class PromotionGate:
     """learning_episodes → ai_insights 4 階段晉升閘。
 
-    Owen 強調：高權重必經人工驗收，避免幻覺污染 RAG。
+    Owen 強調：高權重必經AI 驗證驗收，避免幻覺污染 RAG。
 
     使用範例:
         gate = PromotionGate()
@@ -417,7 +417,7 @@ class PromotionGate:
             return None
 
     # ──────────────────────────────────────────────────────────────────────
-    # Stage 4: 高權重強制人工驗收
+    # Stage 4: 高權重強制 AI 驗證驗收
     # ──────────────────────────────────────────────────────────────────────
     def _stage_4_review(self, episode: Dict[str, Any]) -> GateDecision:
         """weight >= 0.8 → awaiting_review；否則自動晉升 approved。"""
@@ -426,7 +426,7 @@ class PromotionGate:
             return GateDecision(
                 can_promote=False,
                 reason='awaiting_review',
-                detail=f'weight={weight:.3f} >= {STAGE_4_HUMAN_REVIEW_WEIGHT} 強制人工驗收',
+                detail=f'weight={weight:.3f} >= {STAGE_4_HUMAN_REVIEW_WEIGHT} 強制 AI 驗證驗收',
             )
         return GateDecision(can_promote=True, reason='approved')
 
@@ -543,7 +543,7 @@ class PromotionGate:
         Args:
             reason: rejected_quality / rejected_hallucination / rejected_duplicate / rejected_human
             detail: 補充說明（會與 reason 拼成 rejected_reason 文本）
-            human_approver: 已 hash 後的人工審核者識別；不存 PII 原文。
+            human_approver: 已 hash 後的 AI 例外決策者識別；不存 PII 原文。
         """
         valid_statuses = (
             'rejected_quality',
diff --git a/services/llm_caller_registry.py b/services/llm_caller_registry.py
index 8c17721..964901f 100644
--- a/services/llm_caller_registry.py
+++ b/services/llm_caller_registry.py
@@ -26,7 +26,7 @@ CALLER_REGISTRY: frozenset = frozenset({
     # Hermes 競價分析（hermes_analyst_service）
     'hermes_analyst',           # _call_hermes_batch
     'hermes_intent',            # intent_classify (L1 NLP)
-    'hermes_ea_prefetch',       # EA HITL pre-fetch (ADR-021)
+    'hermes_ea_prefetch',       # EA AI 例外決策 pre-fetch (ADR-021)
 
     # KM Embedding（openclaw_learning_service）
     'km_embedding_worker',      # 60s retry queue worker
diff --git a/services/llm_model_router.py b/services/llm_model_router.py
index 2a4ff4b..0cffd8c 100644
--- a/services/llm_model_router.py
+++ b/services/llm_model_router.py
@@ -80,7 +80,7 @@ ROUTING_RULES: Dict[str, list] = {
          'minicpm-v:latest'),
     ],
 
-    # 推理增強場景（EA HITL 戰略決策；Gemini 不可作為預設模型）
+    # 推理增強場景（EA AI 例外決策戰略決策；Gemini 不可作為預設模型）
     'ea_engine': [
         (lambda ctx: bool(ctx.get('require_chain_of_thought', False)),
          'deepseek-r1:14b'),
diff --git a/services/nemoton_dispatcher_service.py b/services/nemoton_dispatcher_service.py
index 5437670..2094b1d 100644
--- a/services/nemoton_dispatcher_service.py
+++ b/services/nemoton_dispatcher_service.py
@@ -14,7 +14,7 @@ NemoTron 行動派發器 (Module 2 — Dispatcher)
 工具清單（扁平化，避免 NIM JSON 截斷）：
   trigger_price_alert      → Telegram 競價高危險預警
   add_to_recommendation    → 寫入前台推薦商品 + Telegram 通知
-  flag_for_human_review    → Telegram 人工覆核請求
+  flag_for_human_review    → Telegram AI 例外決策請求
 """
 
 import json
@@ -29,6 +29,7 @@ from services.mcp_context_service import build_mcp_context
 
 from config import HERMES_URL  # ADR-008 集中化：禁止硬編碼 IP
 from services.ai_call_logger import log_ai_call  # Operation Ollama-First v5.0 P1
+from services.ai_exception_contract import LEGACY_REVIEW_GATE_KEY
 
 logger = logging.getLogger(__name__)
 
@@ -67,14 +68,14 @@ def _has_repeated_phrase(text: str, min_len: int = 4, min_count: int = 2) -> boo
     return False
 
 
-def _sanitize_text(text: str, fallback: str = "請人工確認", max_len: int = 200) -> str:
+def _sanitize_text(text: str, fallback: str = "請 AI 自動驗證確認", max_len: int = 200) -> str:
     """防止 LLM 幻覺文字輸出到 Telegram。
 
     [2026-04-18 台北] Bug-3.1 三層檢測升級 — Claude Opus 4.7
       L1: 連續 ≥15 字中文且整段無標點 → 幻覺（原規則）
       L2: 命中簡體字/異體字黑名單 → 簡繁污染
       L3: 連續中文片段內 4+ 字子串重複 ≥2 次 → 語意坍塌
-    任一層命中 → fallback 替換為「請人工確認」（或呼叫端指定字串）
+    任一層命中 → fallback 替換為「請 AI 自動驗證確認」（或呼叫端指定字串）
     """
     if not text or not isinstance(text, str):
         return fallback
@@ -208,7 +209,7 @@ TOOLS = [
         "type": "function",
         "function": {
             "name": "flag_for_human_review",
-            "description": "當情況複雜、AI 信心不足，或需要人工決策時，發送 Telegram 請求人工覆核",
+            "description": "當情況複雜、AI 信心不足，或需要 AI 例外決策時，發送 Telegram 請求AI 例外決策",
             "parameters": {
                 "type": "object",
                 "properties": {
@@ -325,7 +326,7 @@ def _compute_business_impact(threat) -> dict:
 HEADER_DISPATCHER = "⚡ NemoTron 派發器"
 # 風險級別
 ICON_CRITICAL  = "🚨"   # 高危險/立即行動
-ICON_WARNING   = "⚠️"   # 中風險/人工覆核
+ICON_WARNING   = "⚠️"   # 中風險/AI 例外決策
 ICON_INSIGHT   = "💡"   # 低風險/策略建議
 ICON_REPORT    = "📊"   # 例行報告
 # 業務屬性
@@ -345,7 +346,7 @@ MATCH_TYPE_LABELS = {
 PRICE_BASIS_LABELS = {
     "total_price": "總價可比",
     "unit_price": "單位價可比",
-    "manual_review": "人工覆核後可比",
+    "manual_review": "AI 例外決策後可比",
     "none": "不可比",
 }
 ALERT_TIER_LABELS = {
@@ -528,9 +529,9 @@ def _build_price_decision_envelope(
 
     action = "price_follow_review" if decision_type == "price_alert" else "identity_or_price_review"
     blocked_reason = (
-        "價格調整需人工覆核；不得自動寫入或覆蓋正式競品價格"
+        "價格調整需 AI 例外決策；不得自動寫入或覆蓋正式競品價格"
         if decision_type == "price_alert"
-        else "身份、包裝、單位價或前台狀態需人工確認"
+        else "身份、包裝、單位價或前台狀態需 AI 自動驗證確認"
     )
     risk_reduction = "high" if severity == "P1" else ("medium" if severity == "P2" else "watch")
 
@@ -549,11 +550,12 @@ def _build_price_decision_envelope(
             "competitor_product_name": str(competitor_product_name or "")[:120],
         },
         "evidence": evidence,
-        "analysis": _sanitize_text(analysis, fallback="請人工確認", max_len=300),
+        "analysis": _sanitize_text(analysis, fallback="請 AI 自動驗證確認", max_len=300),
         "recommended_action": {
             "action": action,
             "owner": "營運",
-            "requires_hitl": True,
+            LEGACY_REVIEW_GATE_KEY: False,
+            "requires_ai_exception": True,
         },
         "expected_impact": {
             "momo_price": momo_value if momo_value > 0 else None,
@@ -771,7 +773,7 @@ class NemotronDispatcher:
                     "1. match_type 不是 exact，或 price_basis 不是 total_price，或 alert_tier 不是 price_alert_exact "
                     "→ 不可直接價格告警，呼叫 flag_for_human_review，concern 說明需覆核身份、包裝或單位價。\n"
                     "2. gap_pct < 5% 且 sales_delta < -30% → 非價格異常，呼叫 flag_for_human_review，"
-                    "concern 說明『價差接近 0 但銷量大幅下滑，疑似缺貨/下架/平台流量異常，請人工走查前台』。\n"
+                    "concern 說明『價差接近 0 但銷量大幅下滑，疑似缺貨/下架/平台流量異常，請 AI 例外走查前台』。\n"
                     "3. gap_pct ≥ 5% 且 risk=HIGH → trigger_price_alert（填入 momo_price, comp_price）。\n"
                     "4. 我方價格低於競品且銷量正成長 → add_to_recommendation。\n"
                     "5. confidence < 0.6 或其他複雜情況 → flag_for_human_review。\n"
@@ -918,7 +920,7 @@ class NemotronDispatcher:
             "1. match_type 不是 exact，或 price_basis 不是 total_price，或 alert_tier 不是 price_alert_exact "
             "→ 不可直接價格告警，呼叫 flag_for_human_review，concern 說明需覆核身份、包裝或單位價。\n"
             "2. gap_pct < 5% 且 sales_delta < -30% → 非價格異常，呼叫 flag_for_human_review，"
-            "concern 說明『價差接近 0 但銷量大幅下滑，疑似缺貨/下架/平台流量異常，請人工走查前台』。\n"
+            "concern 說明『價差接近 0 但銷量大幅下滑，疑似缺貨/下架/平台流量異常，請 AI 例外走查前台』。\n"
             "3. gap_pct ≥ 5% 且 risk=HIGH → trigger_price_alert（填入 momo_price, comp_price）。\n"
             "4. 我方價格低於競品且銷量正成長 → add_to_recommendation。\n"
             "5. confidence < 0.6 或其他複雜情況 → flag_for_human_review。\n"
@@ -1076,13 +1078,13 @@ class NemotronDispatcher:
                     continue
 
                 if t.gap_pct < 5 and t.sales_7d_delta_pct < -30:
-                    # Rule 1：價差微小但銷量大跌 → 非定價問題，人工確認
+                    # Rule 1：價差微小但銷量大跌 → 非定價問題，AI 自動驗證確認
                     self._exec_flag_for_human_review(
                         sku=t.sku, name=t.name,
                         concern=(
                             f"🟡 [規則引擎] 價差僅 {t.gap_pct:+.1f}% 但銷量大跌 "
                             f"{t.sales_7d_delta_pct:+.1f}%，疑似缺貨/下架/平台流量異常，"
-                            "請人工走查前台。"
+                            "請 AI 例外走查前台。"
                         ),
                         confidence=0.80,
                         footprint=footprint,
@@ -1116,7 +1118,7 @@ class NemotronDispatcher:
                         threat=t,
                     )
                 else:
-                    # Rule 4：其餘複雜情況 → 人工覆核
+                    # Rule 4：其餘複雜情況 → AI 例外決策
                     self._exec_flag_for_human_review(
                         sku=t.sku, name=t.name,
                         concern=(
@@ -1236,7 +1238,7 @@ class NemotronDispatcher:
         competitor_product_name: str = "",
     ) -> str:
         """
-        類別二：人工覆核
+        類別二：AI 例外決策
         客觀數據由 Python 注入（防幻覺），AI 診斷隔離在獨立欄位
         B' 軌：補金額影響欄位
         """
@@ -1268,7 +1270,7 @@ class NemotronDispatcher:
         )
 
         return (
-            f"{ICON_WARNING} [{HEADER_DISPATCHER}] 異常波動需人工覆核\n\n"
+            f"{ICON_WARNING} [{HEADER_DISPATCHER}] 異常波動需 AI 例外決策\n\n"
             f"🔍 待查商品：[{sku}] {name}\n\n"
             f"{data_block}"
             f"{match_block}"
@@ -1460,7 +1462,7 @@ class NemotronDispatcher:
         competitor_product_id: str = "",
         competitor_product_name: str = "",
     ):
-        """發送語意化人工覆核請求"""
+        """發送語意化AI 例外決策請求"""
         concern = _sanitize_text(concern, fallback=f"數據走勢違背常理，疑似缺貨或前台異常。")
         msg = self._fmt_human_review(
             sku, name, concern, footprint,
@@ -1496,13 +1498,13 @@ class NemotronDispatcher:
         )
         self._send_telegram(msg, decision_envelope=decision_envelope)
         logger.info(
-            f"[Dispatcher] 人工覆核請求 → {sku} loss=${revenue_loss_7d:,.0f}"
+            f"[Dispatcher] AI 例外決策請求 → {sku} loss=${revenue_loss_7d:,.0f}"
         )
         # ADR-007 雙寫
         self._sink_insight_to_km(
             insight_type="human_review",
             sku=sku, name=name,
-            content=f"[人工覆核] {name}。疑慮：{concern}",
+            content=f"[AI 例外決策] {name}。疑慮：{concern}",
             metadata={"confidence": confidence, "gap_pct": gap_pct, "sales_delta": sales_delta,
                       "momo_price": momo_price, "comp_price": comp_price,
                       "revenue_loss_7d": revenue_loss_7d,
@@ -1740,7 +1742,7 @@ class NemotronDispatcher:
           閘門 A：銷量 ≤ -95% 絕對斷崖 → 100% 是缺貨/下架/前台異常，不論價差
                   (2026-04-18 傍晚升級：真實案例 sku=7440662 sales=-100% gap=6.1% 被 NemoTron 錯派降價)
           閘門 B：銷量 ≤ -80% 且 |價差| < 5% → 中度斷崖 + 價差微小，定價非主因
-          命中任一閘門 → 強制走人工覆核，不進 NIM — Claude Opus 4.7
+          命中任一閘門 → 強制走 AI 例外決策，不進 NIM — Claude Opus 4.7
         """
         if not threats:
             return {"dispatched": 0, "skipped": 0, "errors": [], "nim_stats": {}}
@@ -2102,7 +2104,7 @@ if __name__ == "__main__":
             recommended_price=980,       # B' 軌驗證：跟進競品價
         ))
         print()
-        print("=== 類別二：人工覆核 ===")
+        print("=== 類別二：AI 例外決策 ===")
         # [2026-04-18 台北] CLI 測試修正：_fmt_human_review 用 kwargs，避免位置錯位 — Claude Opus 4.7
         print(NemotronDispatcher._fmt_human_review(
             sku="A001", name="玻尿酸面膜10片裝",
@@ -2145,7 +2147,7 @@ if __name__ == "__main__":
                 forced.append(t)
             else:
                 nim_list.append(t)
-        print(f"強制人工覆核 {len(forced)} 筆（預期 3：閘門 A + 閘門 B × 2）: "
+        print(f"強制 AI 例外決策 {len(forced)} 筆（預期 3：閘門 A + 閘門 B × 2）: "
               f"{[t.sku for t in forced]}")
         print(f"送入 NIM 決策 {len(nim_list)} 筆（預期 1）: "
               f"{[t.sku for t in nim_list]}")
diff --git a/services/pchome_growth_momo_backfill_service.py b/services/pchome_growth_momo_backfill_service.py
index 8fe632f..c1835aa 100644
--- a/services/pchome_growth_momo_backfill_service.py
+++ b/services/pchome_growth_momo_backfill_service.py
@@ -115,7 +115,7 @@ def run_pchome_growth_momo_backfill(
     """補高業績 PChome 商品的 MOMO 對應。
 
     不呼叫 LLM，只搜尋 MOMO 候選，並只把可自動判斷的 total_price / unit_price
-    寫入 external_offers；需人工確認的候選會以 needs_review 保存，不進價格判斷。
+    寫入 external_offers；需 AI 自動驗證確認的候選會以 needs_review 保存，不進價格判斷。
     """
     limit = max(1, min(int(limit or 12), 20))
     build_payload = build_payload_func or _default_build_payload
diff --git a/services/telegram_bot_service.py b/services/telegram_bot_service.py
index d777b4c..e317de1 100644
--- a/services/telegram_bot_service.py
+++ b/services/telegram_bot_service.py
@@ -13,6 +13,8 @@ pip install python-telegram-bot>=20.0
 - 每日趨勢摘要推播
 """
 
+from __future__ import annotations
+
 import os
 import json
 import asyncio
@@ -509,7 +511,7 @@ class TrendTelegramBot:
 
         await query.answer()
 
-        # ADR-012 / A' 軌：HITL escalation 按鈕（momo:eig:{event_id} = event_ignore）
+        # ADR-012 / A' 軌：AI 例外事件忽略按鈕（momo:eig:{event_id} = event_ignore）
         # triaged_alert 鍵盤按鈕「🛑 忽略此事件」會 callback momo:eig:<id>
         if data.startswith('momo:eig:'):
             await self._handle_event_ignore_callback(query, data)
@@ -586,7 +588,7 @@ class TrendTelegramBot:
             await self._handle_settings_callback(query, data)
 
     async def _handle_event_ignore_callback(self, query, data: str):
-        """A' 軌：HITL 升級審核「🛑 忽略此事件」按鈕處理。
+        """A' 軌：AI 例外決策「🛑 忽略此事件」按鈕處理。
 
         callback_data 格式：momo:eig:<event_id>
         動作：
@@ -594,7 +596,7 @@ class TrendTelegramBot:
           2. 編輯原訊息：標記「已忽略 by <user> @ <ts>」
           3. 失敗 best-effort，不阻斷 user 體驗
 
-        ADR-012 §③：人工決策必有 audit trail；ai_insights 為 SOT。
+        ADR-012 §③：AI 例外決策必有 audit trail；ai_insights 為 SOT。
         Critic Critical-1 fix: user_label / ts_label 寫入 HTML 前須 escape；
         Critic Medium-2 fix: 空 event_id 直接拒絕，避免污染 audit 資料。
         """
@@ -609,7 +611,7 @@ class TrendTelegramBot:
                 await query.answer("event_id 缺失，忽略動作未生效", show_alert=False)
             except Exception:
                 logger.debug("empty event_id callback answer failed", exc_info=True)
-            logger.warning("[EA HITL] empty event_id callback rejected: %r", data)
+            logger.warning("[EA AI Exception] empty event_id callback rejected: %r", data)
             return
 
         user = getattr(query, 'from_user', None)
@@ -632,8 +634,8 @@ class TrendTelegramBot:
                         VALUES (:type, :content, :conf, :by, :status, :meta)
                     """),
                     {
-                        "type": "human_review",
-                        "content": f"[EA HITL] 事件 {event_id} 由 {user_label_raw} 忽略",
+                        "type": "ai_exception_decision",
+                        "content": f"[EA AI Exception] 事件 {event_id} 由 {user_label_raw} 忽略",
                         "conf": 1.0,
                         "by": f"telegram:{user_label_raw}",
                         "status": "ignored",
@@ -642,6 +644,8 @@ class TrendTelegramBot:
                             "decided_by": user_label_raw,
                             "decided_at": ts_label_raw,
                             "decision": "ignored",
+                            "decision_mode": "ai_exception_decision",
+                            "legacy_insight_type": "human_review",
                         }, ensure_ascii=False),
                     },
                 )
@@ -649,7 +653,7 @@ class TrendTelegramBot:
             finally:
                 session.close()
         except Exception as audit_err:
-            logger.warning(f"[EA HITL] ai_insights audit 寫入失敗（不阻斷 UI）: {audit_err}")
+            logger.warning(f"[EA AI Exception] ai_insights audit 寫入失敗（不阻斷 UI）: {audit_err}")
 
         # Critic Critical-1: user_label / ts_label 須 HTML escape，避免攻擊者
         # 透過 Telegram username 注入 <a>/<pre>/破版標籤
@@ -663,9 +667,9 @@ class TrendTelegramBot:
                 parse_mode='HTML',
             )
         except Exception as ui_err:
-            logger.debug(f"[EA HITL] edit_message 失敗（不阻斷）: {ui_err}")
+            logger.debug(f"[EA AI Exception] edit_message 失敗（不阻斷）: {ui_err}")
 
-        logger.info(f"[EA HITL] event_ignore event_id={event_id} by={user_label_raw}")
+        logger.info(f"[EA AI Exception] event_ignore event_id={event_id} by={user_label_raw}")
 
     async def _handle_openclaw_callback(self, query, context, data: str):
         """轉接 OpenClaw 完整菜單 callback，避免長輪詢 Bot 吃掉 /menu。"""