Update app.py

app.py

@@ -81,23 +81,33 @@ def analyze_world_model(model_name, dataset_key, num_samples=25):
         # **State S{cluster_id} [Label]**: [One sentence explanation of the shared logic/context].
         # """
         prompt = f"""
-        Act as a Mechanistic Interpretability Researcher. You are analyzing latent cluster S{cluster_id} from the '{dataset_key}' dataset.
+        Act as a Mechanistic Interpretability Researcher. You are decoding the latent 'Newtonian' world model 
+        within a Transformer for the '{dataset_key}' dataset.
         
-        The model has clustered these specific snippets because they represent a 'Coherent World State'—an internal map it uses to navigate the data.
+        You are analyzing Cluster S{cluster_id}. This cluster represents a specific 'Equivalence Class' where 
+        the model treats different sequences as functionally identical.
         
-        ### DATASET SNIPPETS:
+        ### RAW DATASET SNIPPETS FOR CLUSTER S{cluster_id}:
         {context_payload}
         
-        ### 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').
+        ### YOUR MANDATORY TASK:
+        Provide a comprehensive, high-fidelity analysis for THIS CLUSTER ALONE. You must deliver the same 
+        level of depth as seen in previous successful state interpretations.
         
-        ### 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]
+        ### REQUIRED OUTPUT FORMAT:
+        **State S{cluster_id} [Structural State Label]**
+        
+        - **Internal World Model**: Explain the CORE 'Law' or 'Invariant' here. What logical map has the model activated? 
+          Describe how this state interconnects lore, timelines, or mechanics into a single 'Coherent World State'.
+          
+        - **Dataset Sensor**: List the specific 'Triggers' that push the model into this state. 
+          Identify: (1) Proper Nouns/Identifiers, (2) Domain-Specific Terminology, (3) Syntax Patterns (e.g., lists, headers).
+          
+        - **Predictive Function**: Explain how being in this state constrains the model's future. 
+          What next-tokens are now 'Biased' or 'Anticipated'? How does this state filter out irrelevant topics?
+          
+        ---
+        (Ensure your response is dense, professional, and strictly follows the bolded sections above.)
         """
         
         try: