math_sanitizer.py

# math_sanitizer.py - V1.1 ProductionMathSanitizer
import re
import logging

logger = logging.getLogger(__name__)

class ProductionMathSanitizer:
    @staticmethod
    def normalize_latex(latex_str: str) -> str:
        """
        V1.1: Standardizes LaTeX for SymPy and LLM comparison.
        """
        if not latex_str: return ""
        
        # 1. Basic Cleaning
        clean = latex_str.strip()
        clean = clean.replace(r'\ ', '')
        clean = clean.replace(r'\times', '*')
        clean = clean.replace(r'\cdot', '*')
        
        # 2. Bracket Normalization
        clean = clean.replace(r'\left(', '(').replace(r'\right)', ')')
        clean = clean.replace(r'\left[', '[').replace(r'\right]', ']')
        clean = clean.replace('{', '(').replace('}', ')')
        
        # 3. Fractions
        while r'\frac' in clean:
            clean = re.sub(r'\\frac\s*\((.*?)\)\((.*?)\)', r'(\1)/(\2)', clean)
            if r'\frac' in clean and '(' not in clean: # Fallback for simple fractions
                 clean = re.sub(r'\\frac\s*(.*?)\s*(.*?)', r'(\1)/(\2)', clean)
        
        # 4. Implicit Multiplication Guard (V1.1)
        clean = re.sub(r'(\d)([a-zA-Z(])', r'\1*\2', clean)
        clean = re.sub(r'\)([a-zA-Z0-9(])', r')*\1', clean)
        
        return clean

    @staticmethod
    def validate_semantic_completeness(anchor_data: dict, formula_tokens: list[str]) -> bool:
        """
        V1.1: Partial Semantic Recovery Check.
        Returns True if the missing tokens are non-critical.
        """
        # Logic to check if critical variables/values are missing
        # For now, a simple check if the main function key is present.
        critical_keys = ['function_equations', 'equations']
        for key in critical_keys:
            if key in anchor_data and anchor_data[key]:
                return True
        return False

    @staticmethod
    def get_symbolic_bridge(proof_graph) -> str:
        """
        V1.1: Zero Hallucination Bridge.
        Converts the Immutable ProofGraph to a clean mathematical context for the LLM.
        """
        bridge = "════════════════════════════════════════\n"
        bridge += "📜 VERIFIED SYMBOLIC BRIDGE (V1.1):\n"
        bridge += "════════════════════════════════════════\n"
        for step in proof_graph.steps:
            bridge += f"Step {step.step_id}: {step.math_content} ({step.logic_description or ''})\n"
            
            # V6 Ontology Injection
            if hasattr(step, 'allowed_concepts') and getattr(step, 'allowed_concepts'):
                concepts_str = ", ".join(step.allowed_concepts)
                tag = getattr(step, 'pedagogical_tag', 'כללי')
                bridge += f"For step {step.step_id}, your pedagogical_tag is '{tag}'. You MUST build your explanation using ONLY the concepts from this list: [{concepts_str}]. Do NOT introduce any other mathematical concepts. Keep it under 2 sentences.\n"

        bridge += "════════════════════════════════════════\n"
        bridge += "RULE: USE ONLY THE DATA ABOVE. DO NOT HALLUCINATE OR CHANGE MATH.\n"
        return bridge

def sanitize_math_ocr_hotfix(text: str) -> str:
    """
    V1.1.1 Aggressive Sanitizer: Removes all spaces and fixes frac regex.
    Fixes failures caused by leading spaces or visual artifacts.
    """
    if not text: return ""

    # תיקון קריטי: הסרת כל הרווחים למניעת כשלי Regex (פתרון לשאלה 2 ו-3)
    text = text.replace(" ", "")

    # ניקוי שאריות ויזואליות
    text = text.replace("\\left", "").replace("\\right", "")

    # נרמול שברים (עובד עכשיו על מחרוזת נקייה מרווחים)
    import re
    text = re.sub(
        r"frac\(([^()]+)\)\(([^()]+)\)",
        lambda m: f"(({m.group(1)})/({m.group(2)}))",
        text
    )
    return text.strip()