Update app.py

app.py

@@ -70,15 +70,34 @@ def analyze_world_model(model_name, dataset_key, num_samples=25):
         context_payload = "\n".join([f"- {s}" for s in snippets[:4]])
         
         # Proper prompt engineering to decode the 'Equivalence Class'
+        # prompt = f"""
+        # Analyze these text snippets from the '{dataset_key}' dataset that fall into the same latent state cluster.
+        # Identify the CORE structural or semantic theme (e.g., 'Historical Narrative', 'Technical Development', 'Numerical Lists').
+        
+        # Text Snippets:
+        # {context_payload}
+        
+        # Format your response exactly as:
+        # **State S{cluster_id} [Label]**: [One sentence explanation of the shared logic/context].
+        # """
         prompt = f"""
-        Analyze these text snippets from the '{dataset_key}' dataset that fall into the same latent state cluster.
-        Identify the CORE structural or semantic theme (e.g., 'Historical Narrative', 'Technical Development', 'Numerical Lists').
+        Act as a Mechanistic Interpretability Researcher. You are analyzing latent cluster S{cluster_id} from the '{dataset_key}' dataset.
         
-        Text Snippets:
+        The model has clustered these specific snippets because they represent a 'Coherent World State'—an internal map it uses to navigate the data.
+        
+        ### DATASET SNIPPETS:
         {context_payload}
         
-        Format your response exactly as:
-        **State S{cluster_id} [Label]**: [One sentence explanation of the shared logic/context].
+        ### YOUR TASK:
+        1. **Newtonian Logic**: Identify the underlying 'Law' or 'Invariant' here. Why does the model treat these as functionally identical? (e.g., 'The model has a dedicated state for tracking chronological advancement').
+        2. **Dataset Attributes**: Pinpoint the specific text features (keywords, syntax, or formatting) that act as 'Sensors' to trigger this state.
+        3. **Functional Role**: Explain how this state helps the model predict the next token. (e.g., 'Being in this state restricts the search space to numerical dates or phase-related verbs').
+        
+        ### RESPONSE FORMAT (Markdown):
+        **State S{cluster_id} [Structural Label]**
+        - **Internal World Model**: [Explanation of the logic]
+        - **Dataset Sensor**: [Key attributes found in the text]
+        - **Predictive Function**: [How it aids next-token prediction]
         """
         
         try: