[V10.318] 收緊 EA HITL 告警證據與排版治理

TODO_NEXT_STEPS.txt

@@ -4,6 +4,7 @@
 ================================================================================
 
 【已完成】
+   - V10.318 收緊 Elephant Alpha HITL 告警治理：`ea_escalation` 只有真正含 SKU/價格比較的 actions 才排成 TOP 待審 SKU 卡片；非 SKU 診斷改為「待確認事項」，並用測試鎖住價格類低信心但無 DB/Hermes 實證時只 suppress、不寫 human_review、不發 Telegram，避免空泛告警打擾人工審核。
    - V10.317 修正 PChome 比價覆蓋率分子：`fetch_competitor_coverage()` 的 valid_matches 改成 `ACTIVE + 有 MOMO 最新價` 商品與有效 PChome `identity_v2` 價格的交集，不再把非活躍或無 MOMO 現價的舊 competitor_prices 列入覆蓋率，避免 daily/growth/PPT/AI 報表高估比價資料品質。
    - V10.315 修正競品簡報/報表指定日期取價：`fetch_competitor_comparison_results()` 在有 start/end date 時改讀 `competitor_price_history` 的期間快照，MOMO 價格也取期間結束前最新價；沒有指定日期才使用目前有效 `competitor_prices`，避免把今天的 PChome 快取價塞回歷史 daily/growth/PPT 判讀。
    - V10.314 擴大 PChome 候選池與搜尋韌性：PChome 搜尋 API 改為依 limit 掃多頁並對 429/5xx/timeout 做有限重試；feeder 預設每個商品最多 5 組搜尋詞、每詞 20 候選、2 頁，且搜尋清理不再刪掉括號/方括號內的品牌與規格，讓正確候選更有機會進 matcher，而不是長期停在「待對比」。

config.py

@@ -320,7 +320,7 @@ YOUTUBE_API_KEY = os.getenv('YOUTUBE_API_KEY', '')
 # ==========================================
 # 系統版本與路徑
 # ==========================================
-SYSTEM_VERSION = "V10.317"
+SYSTEM_VERSION = "V10.318"
 LOG_FILE_PATH = os.path.join(BASE_DIR, 'logs/system.log')
 public_url = PUBLIC_URL  # 用於模板顯示
 

docs/AI_INTELLIGENCE_MODULE_SOT.md

@@ -2,7 +2,7 @@
 
 > **最後更新**: 2026-05-20 (台北時間)
 > **狀態**: 🟢 四 AI Agent 自動化閉環已落地；LLM 路由紅線升級為 Ollama-first 三主機級聯，Gemini 僅備援 / 鎖定場景
-> **適用版本**: V10.307
+> **適用版本**: V10.318
 
 ---
 
@@ -609,3 +609,4 @@ POSTGRES_HOST=momo-db
 | 2026-05-20 | 正確 PChome 候選常因只掃第一頁或搜尋詞丟失品牌/規格而未進入 matcher | V10.314 起搜尋 API 依 limit 掃多頁、對暫時性錯誤有限重試；feeder 預設 5 組搜尋詞、20 候選、2 頁，並保留括號/方括號內品牌與規格，提升覆核隊列與正式比價的候選品質 |
 | 2026-05-20 | 指定日期競品簡報可能混用目前 `competitor_prices` 快取價 | V10.315 起 `fetch_competitor_comparison_results()` 有 start/end date 時改用 `competitor_price_history` 期間快照，MOMO 價格取報表結束日前最新價；即時報表才使用目前有效 `competitor_prices` |
 | 2026-05-20 | PChome 覆蓋率分子可能被非活躍或無 MOMO 現價 SKU 膨脹 | V10.317 起 `fetch_competitor_coverage()` 的 `valid_matches` 改為 active MOMO latest price 與有效 PChome `identity_v2` 價格交集，確保 daily/growth/PPT/AI 看到的比價資料品質不被舊快取列高估 |
+| 2026-05-20 | EA HITL 告警可能把非 SKU 診斷誤排成待審 SKU，或在缺少 DB/Hermes 實證時打擾人工 | V10.318 起 `ea_escalation` 僅對含 SKU/價格比較的 actions 使用競價卡片；非 SKU 診斷改為「待確認事項」。價格類低信心事件若無 DB/Hermes 實證，測試鎖定只 suppress、不寫 human_review、不發 Telegram |

services/telegram_templates.py

@@ -542,6 +542,15 @@ def _format_ea_risk_summary(actions: List[Dict[str, Any]]) -> List[str]:
     return lines
 
 
+def _is_ea_sku_action(item: Dict[str, Any]) -> bool:
+    return bool(
+        item.get("sku")
+        or item.get("comparison")
+        or item.get("momo_price")
+        or item.get("pchome_price")
+    )
+
+
 def _format_ea_action_card(item: Dict[str, Any], index: int) -> List[str]:
     sku = escape(str(item.get("sku") or ""))
     name = escape(_short_text(item.get("name") or item.get("title") or "", 58))
@@ -575,6 +584,11 @@ def _format_ea_action_card(item: Dict[str, Any], index: int) -> List[str]:
     return lines
 
 
+def _format_ea_generic_action(item: Dict[str, Any], index: int) -> str:
+    text = item.get("raw") or item.get("title") or ""
+    return f"{index}. {escape(_short_text(text, 140))}"
+
+
 def _format_ea_escalation_alert(
     *,
     base_event: Dict[str, Any],
@@ -592,8 +606,10 @@ def _format_ea_escalation_alert(
         if part and part.strip()
     ]
     parsed_actions = [_parse_ea_action(action) for action in (ai_actions or [])]
-    shown_actions = parsed_actions[:5]
-    hidden_count = max(0, len(parsed_actions) - len(shown_actions))
+    sku_actions = [item for item in parsed_actions if _is_ea_sku_action(item)]
+    generic_actions = [item for item in parsed_actions if not _is_ea_sku_action(item)]
+    shown_actions = sku_actions[:5]
+    hidden_count = max(0, len(sku_actions) - len(shown_actions))
 
     lines = [
         f"⚡ <b>{escape(str(tier_label))}</b>",
@@ -614,11 +630,11 @@ def _format_ea_escalation_alert(
             f"• {escape(_short_text(ai_summary, 280))}",
         ]
 
-    if parsed_actions:
+    if sku_actions:
         lines += [
             "",
             "📊 <b>風險摘要</b>",
-            *_format_ea_risk_summary(parsed_actions),
+            *_format_ea_risk_summary(sku_actions),
             "",
             "📋 <b>TOP 待審 SKU</b>",
         ]
@@ -635,6 +651,20 @@ def _format_ea_escalation_alert(
             "• 同款：評估跟價、組合促銷或加強 MOMO 價格優勢曝光",
             "• 非同款：標記待審，避免進入自動調價或簡報決策",
         ]
+    elif generic_actions:
+        lines += [
+            "",
+            "📋 <b>待確認事項</b>",
+            *[
+                _format_ea_generic_action(item, idx)
+                for idx, item in enumerate(generic_actions[:5], start=1)
+            ],
+            "",
+            "✅ <b>建議處置</b>",
+            "• 先確認資料來源、最近錯誤紀錄與觀測台狀態",
+            "• 補齊可審核證據後再批准執行",
+            "• 未取得實證前，不執行自動調價、修復或策略派發",
+        ]
     else:
         lines += [
             "",

tests/test_elephant_alpha_engine.py

@@ -231,6 +231,55 @@ def test_escalate_resource_optimization_without_evidence_is_suppressed(monkeypat
     assert suppressed == [("resource_optimization", "no_concrete_evidence")]
 
 
+def test_escalate_price_alert_without_evidence_is_suppressed(monkeypatch):
+    import services.elephant_alpha_autonomous_engine as engine_module
+    from services.elephant_alpha_autonomous_engine import (
+        AutonomousTrigger,
+        ElephantAlphaAutonomousEngine,
+    )
+    from services.elephant_alpha_orchestrator import StrategicDecision
+
+    engine = ElephantAlphaAutonomousEngine()
+    suppressed = []
+    cooldown = []
+
+    async def _no_concrete_actions(top_n=5):
+        return None
+
+    def _raise_if_db_opened():
+        raise AssertionError("no-concrete price escalation should not write human_review")
+
+    monkeypatch.setattr(engine, "_fetch_hermes_threats_summary", _no_concrete_actions)
+    monkeypatch.setattr(engine_module, "get_session", _raise_if_db_opened)
+    monkeypatch.setattr(engine, "_store_escalation", lambda trigger_type: cooldown.append(trigger_type))
+    monkeypatch.setattr(
+        engine,
+        "_record_suppressed_escalation",
+        lambda decision, trigger, reason: suppressed.append((trigger.trigger_type, reason)),
+    )
+
+    decision = StrategicDecision(
+        priority="medium",
+        agents_required=["openclaw"],
+        reasoning="價格調整建議信心不足",
+        expected_outcome="待人工確認",
+        confidence=0.62,
+        execution_plan=[],
+        resource_requirements={},
+    )
+    trigger = AutonomousTrigger(
+        trigger_type="price_drop_alert",
+        conditions={},
+        threshold=0.8,
+        enabled=True,
+    )
+
+    asyncio.run(engine._escalate_to_human(decision, trigger))
+
+    assert cooldown == ["price_drop_alert"]
+    assert suppressed == [("price_drop_alert", "no_concrete_evidence")]
+
+
 def test_resource_pressure_classifier_does_not_equate_backlog_with_cpu_load():
     from services.elephant_alpha_autonomous_engine import ElephantAlphaAutonomousEngine
 

tests/test_telegram_triaged_alert_format.py

@@ -41,3 +41,26 @@ def test_ea_escalation_uses_structured_incident_brief():
     assert "PChome：<code>DABC53-A9009OEF</code>" in msg
     assert "  • [5900068]" not in msg
     assert keyboard["inline_keyboard"][0][0]["callback_data"] == "momo:eig:ea_review_test"
+
+
+def test_ea_escalation_generic_actions_do_not_render_as_sku_cards():
+    msg, _ = triaged_alert(
+        base_event={
+            "event_type": "ea_escalation",
+            "title": "🐘 EA 升級審核 · 程式碼異常偵測",
+            "summary": "低信心且缺少可格式化的具體行動",
+            "id": "ea_review_generic",
+        },
+        tier_label="🐘 Elephant Alpha · L3 HITL",
+        ai_summary="已隱藏 LLM plan 文字，避免把推測當成事實。",
+        ai_cause="觸發類型：程式碼異常偵測 | 信心度：0.62 | 缺少可直接審核的實證資料",
+        ai_actions=[
+            "檢查觸發條件：{\"scan_containers\": [\"momo-pro-system\"]}",
+            "不執行自動動作；請先在觀測台確認對應資料來源與最近錯誤紀錄。",
+        ],
+    )
+
+    assert "📋 <b>待確認事項</b>" in msg
+    assert "📋 <b>TOP 待審 SKU</b>" not in msg
+    assert "• 待審 SKU" not in msg
+    assert "未取得實證前，不執行自動調價、修復或策略派發" in msg