fix(cma): cashflow statements importer + auto_verify/alert_generator observability

- add backend/scripts/import_cashflow_statements.py: load zxt cashflow stat (CNY) into cashflow_statements in 10k CNY, dry-run default, --apply gated

- auto_verify_cron: per-reason skip counters + legacy-condition rule now resolves kpi_code via plan.kpi_id (was unreachable) + --dry-run

- alert_generator: coverage self-check warns when current-month KPI values are absent (0 alerts != no trigger)

task: finance-20260916-cma-health-gaps-3items (items 1/2/3)

backend/scripts/alert_generator.py

@@ -217,6 +217,38 @@ def generate_and_push(db_session) -> dict:
     new_count = run_alert_check(db_session, period)
     logger.info(f"预警检查完成: 新增 {new_count} 条")
 
+    # 1a. 覆盖度自检（2026-09-16 新增，任务 finance-20260916-cma-health-gaps-3items · 项3）：
+    # run_alert_check 的期间口径 = 当月（period=current month），若当月尚无任何 KPIValue，
+    # 则「0 条新预警」并不代表"确无阈值触发"，而是"本次检查未覆盖任何数据"——
+    # 该区别原实现不可见（日志与汇总行都只说 0 条），导致连续多日 0 新增无法归因。
+    try:
+        kpis_with_values = 0
+        active_kpis = db_session.query(KPIDefinition).filter(KPIDefinition.status == "active").all()
+        for k in active_kpis:
+            if not any((k.threshold_red, k.threshold_yellow, k.threshold_green)):
+                continue
+            hit = db_session.query(KPIValue).filter(
+                KPIValue.kpi_id == k.id, KPIValue.period == period,
+                KPIValue.actual_value.isnot(None),
+            ).first()
+            if hit:
+                kpis_with_values += 1
+        coverage = {"period": period, "kpis_with_threshold_and_value": kpis_with_values}
+        if kpis_with_values == 0:
+            latest_period = db_session.query(KPIValue.period).filter(
+                KPIValue.actual_value.isnot(None),
+                KPIValue.period.op("REGEXP")("^[0-9]{4}-[0-9]{2}$"),  # 只取 YYYY-MM 月期间（排除 2026-H1/H2 伪期间）
+            ).order_by(KPIValue.period.desc()).first()
+            logger.warning(
+                f"覆盖度告警: 当月({period})无任何 KPI 实际值 → 本次预警检查未覆盖任何 KPI，"
+                f"「0 条新预警」= 无数据可判，非「确无阈值触发」。库内最新期间="
+                f"{latest_period[0] if latest_period else 'N/A'}"
+            )
+            coverage["latest_period_in_db"] = latest_period[0] if latest_period else None
+    except Exception as e:
+        logger.error(f"覆盖度自检失败: {e}")
+        coverage = {"error": str(e)}
+
     # 1b. 规则引擎检查（预测偏差 / 客户集中度，P2 2026-09-10）
     rule_stats = run_rule_engine_checks(db_session)
 
@@ -231,7 +263,20 @@ def generate_and_push(db_session) -> dict:
             logger.error(f"推送失败: {e}")
 
     return {"period": period, "new_alerts": new_count, "pushed": pushed,
-            "cleaned_green": cleaned, "rule_engine": rule_stats}
+            "cleaned_green": cleaned, "rule_engine": rule_stats,
+            "coverage": coverage}
+
+
+def coverage_note(result: dict) -> str:
+    """汇总行追加提示：当月无 KPI 值时显式说明「0 条 = 无数据可判」而非「确无触发」
+
+    2026-09-16 新增（项3）。独立成函数便于回查（无需触发真实写路径即可断言文案）。
+    """
+    cov = result.get("coverage") or {}
+    if cov.get("kpis_with_threshold_and_value") == 0:
+        return (f" | ⚠️ 当月({result['period']})无KPI值: 本次检查未覆盖任何KPI"
+                f"（库内最新月期间 {cov.get('latest_period_in_db')}）→ 0 条不等于确无触发")
+    return ""
 
 
 if __name__ == "__main__":
@@ -244,6 +289,6 @@ if __name__ == "__main__":
     try:
         result = generate_and_push(db)
         print(f"预警检查: {result['new_alerts']}条新预警 / {result['pushed']}条已推送"
-              f" / 规则引擎{result['rule_engine']}")
+              f" / 规则引擎{result['rule_engine']}{coverage_note(result)}")
     finally:
         db.close()