fix(governance): stabilize adr100 km growth slo

ops/monitoring/slo-rules.yml

@@ -45,9 +45,12 @@ groups:
           /
           sum(rate(approval_records_high_confidence_total[1h]))
 
-      # SLO 4: KM 增長率 = 24h increase (絕對值，不做 rate)
+      # SLO 4: KM 增長率 = DB-derived 24h gauge；fallback 給舊 counter history
       - record: sli:km_growth_rate:24h
-        expr: increase(knowledge_entries_total[24h])
+        expr: |
+          max(knowledge_entries_created_24h)
+          or
+          increase(knowledge_entries_total[24h])
 
       # -----------------------------------------------------------------------
       # Error Budget Recording Rules（輔助 Grafana 顯示）
@@ -248,7 +251,7 @@ groups:
           summary: "SLO KM 增長率嚴重不足（< 5 筆/day）— 疑似 KM 鏈斷裂"
           description: "過去 24h KM 新增 {{ $value }} 筆，遠低於目標 20 筆/day，飛輪學習迴圈疑似中斷。"
           runbook: |
-            1. 確認 knowledge_entries_total counter 是否正常遞增
+            1. 確認 knowledge_entries_created_24h gauge 與 knowledge_entries_total counter 是否正常遞增
             2. 查 governance_agent 日誌中 governance_km_growth_slo_violation
             3. 確認 auto_execute 後 KM 寫入路徑（feedback_flywheel_km_write_gap.md）
             4. 手動執行 POST /api/v1/governance/check

ops/monitoring/tests/test_slo_rules.yaml

@@ -35,7 +35,8 @@ tests:
       - expr: sli:autonomy_rate:5m
         eval_time: 15m
         exp_samples:
-          - value: 0.8
+          - labels: '{__name__="sli:autonomy_rate:5m"}'
+            value: 0.8
 
   # ---- SLI 1: 自主化率 = 100%（無 human_required）----
   - interval: 1m
@@ -47,7 +48,8 @@ tests:
       - expr: sli:autonomy_rate:5m
         eval_time: 15m
         exp_samples:
-          - value: 1.0
+          - labels: '{__name__="sli:autonomy_rate:5m"}'
+            value: 1.0
 
   # ---- SLI 2: 決策準確率 = 90% (success=9, auto_executed=10) ----
   - interval: 1m
@@ -61,7 +63,8 @@ tests:
       - expr: sli:decision_accuracy:5m
         eval_time: 15m
         exp_samples:
-          - value: 0.9
+          - labels: '{__name__="sli:decision_accuracy:5m"}'
+            value: 0.9
 
   # ---- SLI 4: KM 增長率（24h increase）----
   - interval: 1m
@@ -74,7 +77,23 @@ tests:
         eval_time: 25h
         exp_samples:
           # increase over 24h = 1440 samples × 1/min
-          - value: 1440
+          - labels: '{__name__="sli:km_growth_rate:24h"}'
+            value: 1440
+
+  # ---- SLI 4: DB-derived gauge 優先，避免 counter 新上線暖機誤報 0 ----
+  - interval: 1m
+    name: "sli:km_growth_rate:24h 應優先使用 knowledge_entries_created_24h"
+    input_series:
+      - series: "knowledge_entries_created_24h"
+        values: "25x30"
+      - series: "knowledge_entries_total"
+        values: "100x30"
+    promql_expr_test:
+      - expr: sli:km_growth_rate:24h
+        eval_time: 15m
+        exp_samples:
+          - labels: '{__name__="sli:km_growth_rate:24h"}'
+            value: 25
 
   # ============================================================
   # Alert Tests — SLO 1: 自主化率
@@ -120,6 +139,10 @@ tests:
               burn_window: 3d
               team: ai
               auto_repair: "false"
+            exp_annotations:
+              summary: "SLO 自主化率 slow burn（長期趨勢偏低）"
+              description: "自主化率長期低於目標，累積 error budget 消耗率偏高，建議本週 review。"
+              runbook: "分析近 7d 數據，是否需要重訓或調整 confidence threshold。"
 
   # ---- 負測: 自主化率 = 85% → SlowBurn 不觸發 ----
   - interval: 1m
@@ -157,6 +180,10 @@ tests:
               burn_window: 3d
               team: ai
               auto_repair: "false"
+            exp_annotations:
+              summary: "SLO 決策準確率 slow burn（長期趨勢偏低）"
+              description: "決策準確率長期低於目標，累積 error budget 消耗偏高。"
+              runbook: "近 7d verifier 失敗分析，考慮 playbook fine-tune。"
 
   # ---- 負測: 決策準確率 = 92% → SlowBurn 不觸發 ----
   - interval: 1m
@@ -192,6 +219,14 @@ tests:
               slo_name: km_growth_rate
               team: ai
               auto_repair: "false"
+            exp_annotations:
+              summary: "SLO KM 增長率嚴重不足（< 5 筆/day）— 疑似 KM 鏈斷裂"
+              description: "過去 24h KM 新增 0 筆，遠低於目標 20 筆/day，飛輪學習迴圈疑似中斷。"
+              runbook: |
+                1. 確認 knowledge_entries_created_24h gauge 與 knowledge_entries_total counter 是否正常遞增
+                2. 查 governance_agent 日誌中 governance_km_growth_slo_violation
+                3. 確認 auto_execute 後 KM 寫入路徑（feedback_flywheel_km_write_gap.md）
+                4. 手動執行 POST /api/v1/governance/check
 
   # ---- 正測: KM 增長率 = 3/day → Critical 觸發（< 5）----
   - interval: 30m
@@ -210,6 +245,14 @@ tests:
               slo_name: km_growth_rate
               team: ai
               auto_repair: "false"
+            exp_annotations:
+              summary: "SLO KM 增長率嚴重不足（< 5 筆/day）— 疑似 KM 鏈斷裂"
+              description: "過去 24h KM 新增 2.9999999999999996 筆，遠低於目標 20 筆/day，飛輪學習迴圈疑似中斷。"
+              runbook: |
+                1. 確認 knowledge_entries_created_24h gauge 與 knowledge_entries_total counter 是否正常遞增
+                2. 查 governance_agent 日誌中 governance_km_growth_slo_violation
+                3. 確認 auto_execute 後 KM 寫入路徑（feedback_flywheel_km_write_gap.md）
+                4. 手動執行 POST /api/v1/governance/check
 
   # ---- 負測: KM 增長率 = 30/day → Critical 不觸發 ----
   - interval: 1m
@@ -240,3 +283,7 @@ tests:
               slo_name: km_growth_rate
               team: ai
               auto_repair: "false"
+            exp_annotations:
+              summary: "SLO KM 增長率偏低（< 20 筆/day）"
+              description: "過去 24h KM 新增 14.976000000000393 筆，低於目標 20 筆/day。"
+              runbook: "查 KM 寫入路徑（auto_execute 後 _write_execution_result_to_km），確認飛輪 KM 閉環正常。"