feat: carry the reasoning verdict through to LLMResponse

Both assembly paths fill it, streaming and non-streaming alike. Filling
only one is exactly the divergence this issue exposed: M3 returns
reasoning prose over SSE and nothing at all over the plain endpoint, so
a verdict computed on one path says nothing about the other.

The field defaults to UNKNOWN on both TransportResult and LLMResponse.
A transport that does not judge should not get to declare absence on
the provider's behalf, and a default that stays silent is the only one
that cannot lie.

tests/unit/test_openai_compat.py

@@ -22,7 +22,7 @@ from polygateway.transports.openai_compat import (
     _iter_sse_deltas,
     _sse_data_payload,
 )
-from polygateway.types import ChatRequest, LLMResponse, SourceConfig
+from polygateway.types import ChatRequest, LLMResponse, SourceConfig, ThinkingObservation
 
 
 def _source(**overrides):
@@ -471,6 +471,81 @@ class TestReasoningTokens:
         assert result.reasoning_tokens is None
 
 
+class TestThinkingObservationVerdict:
+    """issue #16/#17: 两条组装路径都必须裁定"推理到底发生没发生"。
+
+    流式与非流式各测一遍是刻意的——只填一条路径正是本 issue 的根因形态:
+    库在其中一条路径上悄悄给出了不同的可观测性,下游无从分辨。
+    """
+
+    def _reasoning_usage(self, reasoning):
+        return {**_USAGE, "completion_tokens_details": {"reasoning_tokens": reasoning}}
+
+    async def test_stream_reasoning_content_is_observed(self):
+        def handler(request):
+            return _sse_stream(
+                _chunk(reasoning="想一下"), _chunk(content="ok"), _chunk(usage=_USAGE)
+            )
+
+        result = await _complete(_transport_for(handler), _source())
+        assert result.thinking_observation is ThinkingObservation.OBSERVED
+
+    async def test_stream_without_any_signal_is_unknown(self):
+        """无正文、无 details: 库不知道,就如实说不知道。"""
+
+        def handler(request):
+            return _sse_stream(_chunk(content="ok"), _chunk(usage=_USAGE))
+
+        result = await _complete(_transport_for(handler), _source())
+        assert result.thinking_observation is ThinkingObservation.UNKNOWN
+
+    async def test_stream_zero_reasoning_tokens_is_absent(self):
+        """上游明确上报 0 才算 ABSENT——这是唯一的"确实没推理"证据。"""
+
+        def handler(request):
+            return _sse_stream(_chunk(content="ok"), _chunk(usage=self._reasoning_usage(0)))
+
+        result = await _complete(_transport_for(handler), _source())
+        assert result.thinking_observation is ThinkingObservation.ABSENT
+
+    async def test_non_stream_reasoning_content_is_observed(self):
+        def handler(request):
+            return httpx.Response(
+                200,
+                json={
+                    "choices": [{"message": {"content": "42", "reasoning_content": "想一下"}}],
+                    "usage": _USAGE,
+                },
+            )
+
+        result = await _complete(_transport_for(handler), _source(), stream=False)
+        assert result.thinking_observation is ThinkingObservation.OBSERVED
+
+    async def test_non_stream_without_any_signal_is_unknown(self):
+        """M3 非流式实测形态: 推理已计费却既不回传正文也不回传 details。"""
+
+        def handler(request):
+            return httpx.Response(
+                200, json={"choices": [{"message": {"content": "42"}}], "usage": _USAGE}
+            )
+
+        result = await _complete(_transport_for(handler), _source(), stream=False)
+        assert result.thinking_observation is ThinkingObservation.UNKNOWN
+
+    async def test_non_stream_zero_reasoning_tokens_is_absent(self):
+        def handler(request):
+            return httpx.Response(
+                200,
+                json={
+                    "choices": [{"message": {"content": "42"}}],
+                    "usage": self._reasoning_usage(0),
+                },
+            )
+
+        result = await _complete(_transport_for(handler), _source(), stream=False)
+        assert result.thinking_observation is ThinkingObservation.ABSENT
+
+
 class TestNonStreamFastPath:
     async def test_non_stream_parses_message(self):
         def handler(request):