Add IMO Bench dataset validation script and documentation

imobench/README.md

@@ -12,6 +12,15 @@ As of 02/12/2026, we updated IMO-Answerbench with `answerbench_v2.csv` to fix
 some problems that had ambiguous problem statements or incorrect answers.
 The previous version [answerbench.csv](answerbench.csv) is now deprecated.
 
+## Dataset Validation
+
+To validate the integrity and schema of the IMO Bench CSV files, run:
+
+```bash
+python3 imobench/validate_imobench.py
+...
+
+
 ## Acknowledgments
 
 We would like to thank Xujie from Tsinghua University, Aiden Jung and Hyunwoo

imobench/validate_imobench.py

@@ -0,0 +1,120 @@
+import csv
+import sys
+import os
+
+
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+
+
+def validate_columns(file_path, required_columns):
+    print(f"Validating columns in {os.path.basename(file_path)}...")
+    with open(file_path, newline="", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        headers = reader.fieldnames
+        missing = [col for col in required_columns if col not in headers]
+        if missing:
+            print(f"  Missing columns: {missing}")
+            return False
+    print("  Columns OK")
+    return True
+
+
+def validate_unique_ids(file_path, id_column):
+    print(f"Checking unique IDs in {os.path.basename(file_path)}...")
+    seen = set()
+    with open(file_path, newline="", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        for row in reader:
+            value = row.get(id_column)
+            if not value:
+                print("  Empty ID detected")
+                return False
+            if value in seen:
+                print(f"  Duplicate ID detected: {value}")
+                return False
+            seen.add(value)
+    print("  Unique IDs OK")
+    return True
+
+
+def validate_numeric_column(file_path, column_name):
+    print(f"Checking numeric values in {column_name} of {os.path.basename(file_path)}...")
+    with open(file_path, newline="", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        for row in reader:
+            value = row.get(column_name)
+            if value is None:
+                print(f"  Missing column: {column_name}")
+                return False
+            try:
+                float(value)
+            except ValueError:
+                print(f"  Non-numeric value found: {value}")
+                return False
+    print(f"  {column_name} numeric OK")
+    return True
+
+
+def validate_enum_column(file_path, column_name, allowed_values):
+    print(f"Checking allowed values in {column_name} of {os.path.basename(file_path)}...")
+    bad = {}
+    with open(file_path, newline="", encoding="utf-8") as f:
+        reader = csv.DictReader(f)
+        for row in reader:
+            value = row.get(column_name)
+            if value is None:
+                print(f"  Missing column: {column_name}")
+                return False
+            value = value.strip()
+            if value not in allowed_values:
+                bad[value] = bad.get(value, 0) + 1
+                if len(bad) >= 10:
+                    break
+    if bad:
+        print(f"  Unexpected values found (showing up to 10): {bad}")
+        return False
+    print(f"  {column_name} values OK")
+    return True
+
+
+def main():
+    success = True
+
+    answerbench = os.path.join(BASE_DIR, "answerbench_v2.csv")
+    proofbench = os.path.join(BASE_DIR, "proofbench.csv")
+    gradingbench = os.path.join(BASE_DIR, "gradingbench.csv")
+
+    success &= validate_columns(
+        answerbench,
+        ["Problem ID", "Problem", "Short Answer", "Category", "Subcategory", "Source"],
+    )
+    success &= validate_unique_ids(answerbench, "Problem ID")
+
+    success &= validate_columns(
+        proofbench,
+        ["Problem ID", "Problem", "Solution", "Grading guidelines", "Category", "Level", "Short Answer", "Source"],
+    )
+    success &= validate_unique_ids(proofbench, "Problem ID")
+
+    success &= validate_columns(
+        gradingbench,
+        ["Grading ID", "Problem ID", "Problem", "Solution", "Grading guidelines", "Response", "Points", "Reward", "Problem Source"],
+    )
+    success &= validate_unique_ids(gradingbench, "Grading ID")
+    success &= validate_numeric_column(gradingbench, "Points")
+    success &= validate_enum_column(
+        gradingbench,
+        "Reward",
+        {"Correct", "Incorrect", "Partial", "Almost"},
+    )
+
+    if not success:
+        print("\nValidation failed.")
+        sys.exit(1)
+
+    print("\nAll validations passed.")
+
+
+if __name__ == "__main__":
+    main()
+