fix ut

Author: mayinghan

client/evaluation_logs/trajectory_data_driven_20250805_144913.jsonl

@@ -0,0 +1,2 @@
+{"type": "summary", "task_id": "data_driven", "timestamp": "2025-08-05T14:49:13.783963", "num_rollouts": 1, "successful_rollouts": 1, "failed_rollouts": 0, "success_rate": 1.0, "avg_score": 1.0, "std_dev": 0.0, "min_score": 1.0, "max_score": 1.0}
+{"type": "individual_result", "task_id": "data_driven", "rollout_index": 0, "timestamp": "2025-08-05T14:49:13.784559", "score": 1.0}

client/evaluation_logs/trajectory_test_task_20250805_145011.jsonl

@@ -0,0 +1,3 @@
+{"type": "summary", "task_id": "test_task", "timestamp": "2025-08-05T14:50:11.743811", "num_rollouts": 2, "successful_rollouts": 2, "failed_rollouts": 0, "success_rate": 1.0, "avg_score": 0.5, "std_dev": 0.7071067811865476, "min_score": 0.0, "max_score": 1.0}
+{"type": "individual_result", "task_id": "test_task", "rollout_index": 0, "timestamp": "2025-08-05T14:50:11.744330", "score": 1.0, "sample_data": {"id": "run_001", "seed": 42}}
+{"type": "individual_result", "task_id": "test_task", "rollout_index": 1, "timestamp": "2025-08-05T14:50:11.744367", "score": 0.0, "sample_data": {"id": "run_002", "seed": 123}}

client/evaluation_logs/trajectory_traditional_20250805_144913.jsonl

@@ -0,0 +1,2 @@
+{"type": "summary", "task_id": "traditional", "timestamp": "2025-08-05T14:49:13.784961", "num_rollouts": 1, "successful_rollouts": 1, "failed_rollouts": 0, "success_rate": 1.0, "avg_score": 0.5, "std_dev": 0.0, "min_score": 0.5, "max_score": 0.5}
+{"type": "individual_result", "task_id": "traditional", "rollout_index": 0, "timestamp": "2025-08-05T14:49:13.785457", "score": 0.5}

tests/test_url_handling.py

@@ -1,5 +1,4 @@
-import asyncio
-
+from unittest.mock import AsyncMock, patch
 import httpx
 import pytest
 from werkzeug.wrappers import Response
@@ -17,11 +16,17 @@ async def test_mcp_env_make_appends_trailing_slash():
     base_url = "http://localhost:8000/mcp"
     corrected_url = "http://localhost:8000/mcp/"
 
-    # Use n and seeds to avoid needing a full dataset
-    envs = await ep.make(base_url, n=1, seeds=[42])
+    with patch(
+        "eval_protocol.mcp.client.connection.MCPConnectionManager.initialize_session",
+        new_callable=AsyncMock,
+    ) as mock_init:
+        mock_init.return_value = None
+
+        envs = await ep.make(base_url, n=1, seeds=[42])
+
+        mock_init.assert_awaited_once()
 
     assert len(envs.sessions) == 1
-    # The session's base_url should have the trailing slash
     assert envs.sessions[0].base_url == corrected_url
 
 
@@ -32,8 +37,15 @@ async def test_mcp_env_make_keeps_existing_trailing_slash():
     """
     base_url = "http://localhost:8000/mcp/"
 
-    # Use n and seeds to avoid needing a full dataset
-    envs = await ep.make(base_url, n=1, seeds=[42])
+    with patch(
+        "eval_protocol.mcp.client.connection.MCPConnectionManager.initialize_session",
+        new_callable=AsyncMock,
+    ) as mock_init:
+        mock_init.return_value = None
+
+        envs = await ep.make(base_url, n=1, seeds=[42])
+
+        mock_init.assert_awaited_once()
 
     assert len(envs.sessions) == 1
     # The session's base_url should remain unchanged