Update app.py

app.py

@@ -10,8 +10,7 @@ Architecture:
 """
 
 import os, time, threading, json, re, textwrap, copy
-import traceback, html
-import requests
+import traceback, html, itertools, math
 from collections import deque, defaultdict
 from typing import Dict, List, Tuple, Any, Optional, Set
 from dataclasses import dataclass, field
@@ -180,23 +179,45 @@ def build_core_problem(axl_def: AXLProblemDef) -> core.Problem:
 # ══════════════════════════════════════════════════════════════════════
 
 def _extract_json_from_text(text: str) -> str:
-    """Bulletproof JSON extractor to handle conversational AI padding."""
+    """Robust, lexer-style JSON extractor that handles LaTeX groupings."""
     text = text.strip()
     if not text:
         return ""
         
-    # 1. Try to find explicit markdown json blocks
-    match = re.search(r'```(?:json)?\s*(\{.*?\})\s*```', text, re.DOTALL | re.IGNORECASE)
-    if match:
-        return match.group(1).strip()
+    # 1. First, search for explicit markdown blocks
+    blocks = re.findall(r'```(?:json)?\s*(\{.*?\})\s*```', text, re.DOTALL | re.IGNORECASE)
+    for b in blocks:
+        cleaned = b.strip()
+        # Verify it looks like our AXL/Collapser schema
+        if any(k in cleaned for k in ['"variables"', '"constraints"', '"hypothesis_markdown"', '"name"']):
+            return cleaned
+    if blocks:
+        return blocks[0].strip()
         
-    # 2. Try to find the outermost curly braces
-    start = text.find('{')
-    end = text.rfind('}')
-    if start != -1 and end != -1 and end >= start:
-        return text[start:end+1].strip()
+    # 2. Scope-matching parser to isolate valid candidates
+    candidates = []
+    for match in re.finditer(r'\{', text):
+        start_idx = match.start()
+        depth = 0
+        for i in range(start_idx, len(text)):
+            if text[i] == '{':
+                depth += 1
+            elif text[i] == '}':
+                depth -= 1
+                if depth == 0:
+                    candidates.append(text[start_idx:i+1])
+                    break
+                    
+    # Prioritize candidates containing key schema elements
+    for cand in candidates:
+        cleaned = cand.strip()
+        if any(k in cleaned for k in ['"variables"', '"constraints"', '"hypothesis_markdown"', '"name"']):
+            return cleaned
+            
+    # Fallback to the largest bracket scope
+    if candidates:
+        return max(candidates, key=len).strip()
         
-    # 3. Fallback to raw text
     return text
 
 def call_external_ai(url: str, model: str, system_prompt: str, prompt: str) -> str: