fix(rag): lift chat output cap to 32768 tokens — long answers no longer cut off

tests/unit/test_llm_client.py

@@ -273,11 +273,13 @@ class _FakeCompletions:
 
 
 def _make_stream_client(
-    chunks: list | None = None, fail: Exception | None = None
+    chunks: list | None = None,
+    fail: Exception | None = None,
+    **settings_kwargs: Any,
 ) -> tuple[LLMClient, _FakeCompletions]:
     completions = _FakeCompletions(chunks, fail)
     fake_openai = SimpleNamespace(chat=SimpleNamespace(completions=completions))
-    llm = LLMClient(_settings())
+    llm = LLMClient(_settings(**settings_kwargs))
     llm._client = fake_openai  # pyright: ignore[reportAttributeAccessIssue]
     return llm, completions
 
@@ -302,10 +304,22 @@ def test_chat_stream_uses_locked_generation_params() -> None:
     assert completions.kwargs["model"] == "turbo"
     assert completions.kwargs["stream"] is True
     assert completions.kwargs["temperature"] == 0.4
-    assert completions.kwargs["max_tokens"] == 700
+    # Phase 11: the old hard 700-token cap is gone — answers may run up to
+    # BOR_MAX_OUTPUT_TOKENS (default 32 768) so they are not cut off.
+    assert completions.kwargs["max_tokens"] == 32_768
     assert completions.kwargs["messages"] == messages
 
 
+def test_chat_stream_max_tokens_comes_from_settings() -> None:
+    """The output cap is operator-configurable, not a client constant."""
+    llm, completions = _make_stream_client(
+        [_chunk("x")], max_output_tokens=1234  # pyright: ignore[reportArgumentType]
+    )
+    asyncio.run(_collect(llm, [{"role": "user", "content": "q"}]))
+    assert completions.kwargs is not None
+    assert completions.kwargs["max_tokens"] == 1234
+
+
 def test_chat_stream_skips_empty_deltas_and_choiceless_chunks() -> None:
     llm, _ = _make_stream_client([_chunk("a"), _chunk(empty=True), _chunk(None), _chunk("b")])
     assert asyncio.run(_collect(llm, [{"role": "user", "content": "q"}])) == ["a", "b"]