Upload tests/test_pipeline_e2e.py

tests/test_pipeline_e2e.py

@@ -100,7 +100,7 @@ class TestProcessCandidateDirectly:
             ],
             neutralization=Neutralization.SUBINDUSTRY,
             decay=5,
-            novelty_claim="Test proven candidate",
+            novelty_claim="Test proven candidate with sufficient length",
             academic_anchor=None,
             anomaly_tag=AnomalyTag.PEAD,
         )
@@ -173,7 +173,7 @@ class TestLLMPathWithMocking:
             ],
             neutralization=Neutralization.SUBINDUSTRY,
             decay=5,
-            novelty_claim="Mocked novel alpha",
+            novelty_claim="Mocked novel alpha with enough chars to pass validation",
             academic_anchor=None,
             anomaly_tag=AnomalyTag.VALUE,
         )
@@ -255,7 +255,7 @@ class TestLLMPathWithMocking:
             components=[Component(name="main", fields=[field.id], operators=["rank"], horizon_days=252, weight=1.0, sign_direction="long_high")],
             neutralization=Neutralization.SUBINDUSTRY,
             decay=5,
-            novelty_claim="Test",
+            novelty_claim="Test novelty claim that is long enough",
             academic_anchor=None,
             anomaly_tag=AnomalyTag.PEAD,
         )
@@ -263,9 +263,6 @@ class TestLLMPathWithMocking:
 
         async def _run_with_kill():
             # Mock crowd scout to return KILL
-            original_scout = pipeline.__class__._process_candidate
-            # We need to mock the internal LLM calls
-            # Actually, let's use monkeypatch on the personas module
             with patch("alpha_factory.personas.crowd_scout.scout_novelty", new_callable=AsyncMock) as mock_scout:
                 mock_scout.return_value = CrowdScoutResult(
                     max_corr_to_library=0.9,
@@ -510,7 +507,8 @@ class TestAcceptanceChecklist:
         )
 
         assert not result.all_passed
-        assert any("RETURNS-CORR" in k for k in result.blocking_failures or [])
+        # The blocking failure message contains the threshold explanation
+        assert any("0.95" in f for f in result.blocking_failures or [])
 
 
 class TestLLMClient:
@@ -534,13 +532,15 @@ class TestLLMClient:
         from alpha_factory.infra.llm_client import LLMClient, LLMConfig, TokenBudgetExceeded
         config = LLMConfig()
         client = LLMClient(config)
-        client._token_count = 4_999_999
+        # 4,999,000 + 10 = 4,999,010 < 5,000,000 → should NOT raise
+        client._token_count = 4_999_000
         client._check_budget(estimated_tokens=10)
-        assert client._token_count == 4_999_999  # Should not raise yet
+        assert client._token_count == 4_999_000  # Should not raise
 
-        client._token_count = 5_000_001
+        # 4,999,999 + 10 = 5,000,009 >= 5,000,000 → SHOULD raise
+        client._token_count = 4_999_999
         with pytest.raises(TokenBudgetExceeded):
-            client._check_budget()
+            client._check_budget(estimated_tokens=10)
 
 
 class TestBrainClient: