fix first seen date feat

data.py

@@ -68,12 +68,21 @@ def generate_fake_dates(num_days: int = 7) -> List[str]:
 
 def parse_json_field(value) -> dict:
     """Safely parse a JSON field that might be a string or dict."""
+    if value is None or pd.isna(value):
+        return {}
     if isinstance(value, str):
         try:
             return json.loads(value)
         except:
             return {}
-    return value if isinstance(value, dict) else {}
+    # Handle dict-like objects (including pandas Series/dict)
+    if isinstance(value, dict):
+        return value
+    # Try to convert to dict if possible
+    try:
+        return dict(value) if hasattr(value, '__iter__') else {}
+    except:
+        return {}
 
 def extract_date_from_path(path: str, pattern: str) -> Optional[str]:
     """Extract date from file path using regex pattern."""
@@ -368,21 +377,37 @@ def get_fake_historical_data(start_date: str, end_date: str) -> pd.DataFrame:
 
 def find_failure_first_seen(historical_df: pd.DataFrame, model_name: str, test_name: str, device: str, gpu_type: str) -> Optional[str]:
     """Find the first date when a specific test failure appeared in historical data."""
-    if historical_df.empty:
+    if historical_df is None or historical_df.empty:
         return None
     
     try:
-        model_data = historical_df[historical_df.index == model_name.lower()].copy()
+        model_name_lower = model_name.lower()
+        # Filter by model name (case-insensitive)
+        model_data = historical_df[historical_df.index.str.lower() == model_name_lower].copy()
         if model_data.empty:
             return None
         
+        # Ensure we have a 'date' column
+        if 'date' not in model_data.columns:
+            return None
+        
         # Check each date (oldest first) for this failure
         for _, row in model_data.sort_values('date').iterrows():
-            failures = parse_json_field(row.get(f'failures_{device}'))
-            if gpu_type in failures:
-                for test in failures[gpu_type]:
-                    if test.get('line', '') == test_name:
-                        return row.get('date')
+            failures_raw = row.get(f'failures_{device}')
+            if failures_raw is None or pd.isna(failures_raw):
+                continue
+            
+            # Parse failures (could be dict, string, or already parsed)
+            failures = parse_json_field(failures_raw)
+            if not isinstance(failures, dict) or gpu_type not in failures:
+                continue
+            
+            # Check each test in this gpu_type
+            for test in failures.get(gpu_type, []):
+                if isinstance(test, dict) and test.get('line', '') == test_name:
+                    date_value = row.get('date')
+                    return date_value if date_value else None
+        
         return None
         
     except Exception as e: