[claude] Wire QuotaMonitor.select_model() into cascade router (#1106) (#1113)

src/infrastructure/router/cascade.py

@@ -485,18 +485,26 @@ class CascadeRouter:
     def _quota_allows_cloud(self, provider: Provider) -> bool:
         """Check quota before routing to a cloud provider.
 
-        Uses the metabolic protocol: cloud calls are gated by 5-hour quota.
+        Uses the metabolic protocol via select_model(): cloud calls are only
+        allowed when the quota monitor recommends a cloud model (BURST tier).
         Returns True (allow cloud) if quota monitor is unavailable or returns None.
         """
         if _quota_monitor is None:
             return True
         try:
-            # Map provider type to task_value heuristic
-            task_value = "high"  # conservative default
-            status = _quota_monitor.check()
-            if status is None:
-                return True  # No credentials — caller decides based on config
-            return _quota_monitor.should_use_cloud(task_value)
+            suggested = _quota_monitor.select_model("high")
+            # Cloud is allowed only when select_model recommends the cloud model
+            allows = suggested == "claude-sonnet-4-6"
+            if not allows:
+                status = _quota_monitor.check()
+                tier = status.recommended_tier.value if status else "unknown"
+                logger.info(
+                    "Metabolic protocol: %s tier — downshifting %s to local (%s)",
+                    tier,
+                    provider.name,
+                    suggested,
+                )
+            return allows
         except Exception as exc:
             logger.warning("Quota check failed, allowing cloud: %s", exc)
             return True