diff --git a/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py b/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
index e8708bb3..cf0924b5 100644
--- a/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
+++ b/src/tamperbench/whitebox/evals/mt_bench/mt_bench.py
@@ -5,8 +5,9 @@
 
 from dataclasses import dataclass
 from functools import partial
-from typing import Any
+from typing import Any, cast
 
+import pandera.polars as pa
 import polars as pl
 import torch
 from openai import OpenAI
@@ -66,23 +67,23 @@ class MTBenchEvaluationConfig(WhiteBoxEvaluationConfig):
 class MTBenchInferenceSchema(InferenceSchema):
     """Schema for MT-Bench multi-turn inferences."""
 
-    question_id: int
-    category: str
-    turn_1_question: str
-    turn_1_response: str
-    turn_2_question: str
-    turn_2_response: str
+    question_id: int = cast("int", pa.Field(nullable=False))
+    category: str = cast("str", pa.Field(nullable=False))
+    turn_1_question: str = cast("str", pa.Field(nullable=False))
+    turn_1_response: str = cast("str", pa.Field(nullable=False))
+    turn_2_question: str = cast("str", pa.Field(nullable=False))
+    turn_2_response: str = cast("str", pa.Field(nullable=False))
 
 
 class MTBenchScoreSchema(ScoreSchema):
     """Schema for MT-Bench scores with per-turn details."""
 
-    question_id: int
-    category: str
-    turn_1_score: float
-    turn_2_score: float
-    judge_response_1: str
-    judge_response_2: str
+    question_id: int = cast("int", pa.Field(nullable=False))
+    category: str = cast("str", pa.Field(nullable=False))
+    turn_1_score: float = cast("float", pa.Field(nullable=False, coerce=True))
+    turn_2_score: float = cast("float", pa.Field(nullable=False, coerce=True))
+    judge_response_1: str = cast("str", pa.Field(nullable=False))
+    judge_response_2: str = cast("str", pa.Field(nullable=False))
 
 
 class MTBenchEvaluation(WhiteBoxEvaluation[MTBenchEvaluationConfig]):
@@ -151,11 +152,11 @@ def _generate_response(
             max_length=2048,
         )
 
-        input_ids: Tensor = encoding["input_ids"].to(model.device)
-        attention_mask: Tensor = encoding["attention_mask"].to(model.device)
+        input_ids: Tensor = cast(Tensor, encoding["input_ids"]).to(model.device)
+        attention_mask: Tensor = cast(Tensor, encoding["attention_mask"]).to(model.device)
 
         with torch.no_grad():
-            outputs: GenerateOutput = model.generate(
+            outputs: GenerateOutput = model.generate(  # pyright: ignore[reportCallIssue]
                 input_ids=input_ids,
                 attention_mask=attention_mask,
                 max_new_tokens=max_new_tokens,
@@ -261,7 +262,7 @@ def _score_row(self, row: dict[str, Any], question_map: dict[int, MTBenchQuestio
         return {
             InferenceSchema.prompt: row[InferenceSchema.prompt],
             InferenceSchema.response: row[InferenceSchema.response],
-            ScoreSchema.score: (score_1 + score_2) / 2,
+            ScoreSchema.score: (score_1 + score_2) / 2,  # pyright: ignore[reportReturnType]
             "question_id": row["question_id"],
             "category": row["category"],
             "turn_1_score": score_1,
@@ -305,11 +306,11 @@ def compute_results(self, scores: DataFrame[ScoreSchema]) -> DataFrame[Evaluatio
         scores_df = ScoreSchema.validate(scores)
 
         # Overall average score
-        mean_score: float = float(scores_df[ScoreSchema.score].mean())
+        mean_score: float = float(scores_df[ScoreSchema.score].mean())  # pyright: ignore[reportArgumentType,reportCallIssue]
 
         # Per-turn averages
-        mean_turn_1: float = float(scores_df["turn_1_score"].mean())
-        mean_turn_2: float = float(scores_df["turn_2_score"].mean())
+        mean_turn_1: float = float(scores_df["turn_1_score"].mean())  # pyright: ignore[reportArgumentType]
+        mean_turn_2: float = float(scores_df["turn_2_score"].mean())  # pyright: ignore[reportArgumentType]
 
         _metrics_df: pl.DataFrame = pl.from_dict(
             data={
@@ -318,7 +319,7 @@ def compute_results(self, scores: DataFrame[ScoreSchema]) -> DataFrame[Evaluatio
                     str(MetricName.MT_BENCH_TURN_1_SCORE),
                     str(MetricName.MT_BENCH_TURN_2_SCORE),
                 ],
-                EvaluationSchema.metric_value: [mean_score, mean_turn_1, mean_turn_2],
+                EvaluationSchema.metric_value: [mean_score, mean_turn_1, mean_turn_2],  # pyright: ignore[reportArgumentType]
             }
         )
         return EvaluationSchema.validate(_metrics_df)