pyright round 4

eval_protocol/benchmarks/test_livebench_data_analysis.py

@@ -258,6 +258,7 @@ def _read_jsonl_table_from_text(text: str, header_cols: List[str]):
 
     reader = _read_df_v1 if version == "v1" else _read_df_v2
     gt_df = reader(output_fmt, ground_truth)
+    assert gt_df is not None, "GT dataframe is None"
 
     llm_clean = _clean_llm_output(llm_answer)
     llm_clean = _remove_initial_phrase(llm_clean)

eval_protocol/benchmarks/test_tau_bench_airline.py

@@ -198,6 +198,7 @@ def test_tau_bench_airline_evaluation(row: EvaluationRow) -> EvaluationRow:
     task = Task(
         id="Filler", evaluation_criteria=evaluation_criteria, user_scenario=UserScenario(instructions="Filler")
     )  # id and user_scenario are required for the Task type but not used in calculating reward
+    assert task.evaluation_criteria is not None, "Task evaluation criteria is None"
 
     if RewardType.DB in task.evaluation_criteria.reward_basis:
         env_reward_info = EnvironmentEvaluator.calculate_reward(

eval_protocol/execution/pipeline.py

@@ -212,6 +212,7 @@ async def _execute_standard_generation(
         if system_prompt_content:
             current_messages_for_rollout.append({"role": "system", "content": system_prompt_content})
         current_messages_for_rollout.append({"role": "user", "content": user_query})
+        assert self.model_client is not None, "at this point model client needs to be initialized"
 
         generation_output_std = await self.model_client.generate(
             messages=current_messages_for_rollout,

eval_protocol/mcp/client/connection.py

@@ -11,7 +11,7 @@
 import logging
 import time
 from contextlib import AsyncExitStack
-from typing import Any, Dict, List, Optional, Tuple
+from typing import Any, Dict, List, Optional, Tuple, cast
 
 import httpx
 from mcp.client.session import ClientSession
@@ -276,7 +276,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
         try:
             # List available resources - this is where initial state should come from
             logger.debug(f"Session {session.session_id}: Discovering MCP resources for initial state...")
-            resources_response = await mcp_session.list_resources()
+            mcp_session_local = session._mcp_session
+            if mcp_session_local is None:
+                raise RuntimeError("Session not initialized while listing resources")
+            resources_response = await mcp_session_local.list_resources()
             resources = resources_response.resources if hasattr(resources_response, "resources") else []
             logger.debug(f"Session {session.session_id}: Found {len(resources)} MCP resources")
             for resource in resources:
@@ -303,7 +306,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
                     f"Session {session.session_id}: Reading initial state from resource: {initial_state_resource.uri}"
                 )
 
-                resource_content = await mcp_session.read_resource(initial_state_resource.uri)
+                mcp_session_for_read = session._mcp_session
+                if mcp_session_for_read is None:
+                    raise RuntimeError("Session not initialized while reading resource")
+                resource_content = await mcp_session_for_read.read_resource(initial_state_resource.uri)
 
                 # Handle the new ResourceContents format
                 text_value = getattr(resource_content, "text", None)
@@ -348,7 +354,10 @@ async def _get_initial_state_from_mcp_resource(self, session: MCPSession) -> Any
                         f"Session {session.session_id}: About to call mcp_session.read_resource with fallback URI: {first_resource.uri}"
                     )
 
-                    resource_content = await mcp_session.read_resource(first_resource.uri)
+                    mcp_session_for_fallback_read = session._mcp_session
+                    if mcp_session_for_fallback_read is None:
+                        raise RuntimeError("Session not initialized while reading fallback resource")
+                    resource_content = await mcp_session_for_fallback_read.read_resource(first_resource.uri)
 
                     logger.debug(
                         f"Session {session.session_id}: fallback read_resource returned type: {type(resource_content)}"