fix(runtime): release 2.7.1 with consistent ADB and failure diagnostics

tests/unit/test_llm_gateway.py

@@ -1,6 +1,7 @@
 """Tests for the LLM gateway: complete(), classified recovery, and fallback."""
 
 from types import SimpleNamespace
+from unittest.mock import AsyncMock
 
 from langchain_core.messages import AIMessage, AIMessageChunk, HumanMessage, SystemMessage
 import pytest
@@ -16,6 +17,55 @@ from artemis.services.llm import (
 )
 
 
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    "invalid",
+    [
+        None,
+        {"raw": AIMessage(content=""), "parsed": None, "parsing_error": None},
+        {
+            "raw": AIMessage(content=""),
+            "parsed": None,
+            "parsing_error": ValueError("Invalid schema"),
+        },
+    ],
+)
+async def test_missing_structured_output_retries_inside_existing_budget(invalid):
+    expected = {"found": True, "output": "com.example.app", "reason": "Matched"}
+    base = SimpleNamespace(
+        ainvoke=AsyncMock(
+            side_effect=[
+                invalid,
+                {"raw": AIMessage(content=""), "parsed": expected, "parsing_error": None},
+            ]
+        )
+    )
+    model = RobustChatModelWrapper(base, structured_mode="parsed")
+    assert await model.ainvoke([]) == expected
+    assert base.ainvoke.await_count == 2
+
+
+@pytest.mark.asyncio
+async def test_empty_structured_output_exhausts_then_uses_fallback():
+    primary = SimpleNamespace(
+        ainvoke=AsyncMock(return_value={"raw": AIMessage(content=""), "parsed": None})
+    )
+    expected = {"found": False, "output": None, "reason": "No match"}
+    secondary = SimpleNamespace(
+        ainvoke=AsyncMock(return_value={"raw": AIMessage(content=""), "parsed": expected})
+    )
+    main = RobustChatModelWrapper(
+        primary, endpoint=ModelEndpoint(model_name="primary"), structured_mode="parsed"
+    )
+    fallback = RobustChatModelWrapper(
+        secondary, endpoint=ModelEndpoint(model_name="fallback"), structured_mode="parsed"
+    )
+    result = await with_fallback(lambda: main.ainvoke([]), lambda: fallback.ainvoke([]))
+    assert result == expected
+    assert primary.ainvoke.await_count == 2
+    assert secondary.ainvoke.await_count == 1
+
+
 @pytest.fixture(autouse=True)
 def _isolate_gateway_state(monkeypatch, tmp_path):
     llm_service._ENDPOINT_BREAKER._states.clear()