domain/risk_engine.py

# domain/risk_engine.py
import logging
from typing import Dict, Any, List
from .curriculum_classifier import CurriculumClassifier

logger = logging.getLogger(__name__)

class CognitiveRiskEngine:
    """
    V6.1 Phase 4: Risk Scoring
    Evaluates the cognitive risk of a proposal based on AST complexity, 
    step count, retries, and symbolic stability.
    """
    
    WEIGHTS = {
        "ast_complexity": 0.30,
        "reasoning_depth": 0.20,
        "retry_penalty": 0.25,
        "symbolic_instability": 0.25
    }
    
    @classmethod
    def calculate_risk_score(cls, 
                             math_input: str, 
                             draft_steps: List[Dict[str, Any]], 
                             retry_count: int,
                             validation_errors: int) -> Dict[str, Any]:
        """
        Calculates the Cognitive Risk Score (CRS) between 0.0 and 1.0.
        """
        # 1. AST Complexity (normalized 0-1)
        # We reuse CurriculumClassifier's internal complexity estimation if possible, 
        # but here we normalize it against a ceiling of 20.
        raw_complexity = CurriculumClassifier.estimate_complexity(math_input)
        ast_risk = min(raw_complexity / 20.0, 1.0)
        
        # 2. Reasoning Depth (normalized 0-1)
        # 10 steps is considered high depth for our MVP
        depth_risk = min(len(draft_steps) / 10.0, 1.0)
        
        # 3. Retry Penalty
        # 0 retries = 0.0, 1 retry = 1.0 (since max retry is 1 in V6.1)
        retry_risk = 1.0 if retry_count > 0 else 0.0
        
        # 4. Symbolic Instability
        # Based on validation failures during the process
        instability_risk = min(validation_errors / 5.0, 1.0)
        
        # Weighted Final Score
        crs = (ast_risk * cls.WEIGHTS["ast_complexity"] +
               depth_risk * cls.WEIGHTS["reasoning_depth"] +
               retry_risk * cls.WEIGHTS["retry_penalty"] +
               instability_risk * cls.WEIGHTS["symbolic_instability"])
        
        result = {
            "risk_score": round(crs, 3),
            "features": {
                "ast_risk": round(ast_risk, 3),
                "depth_risk": round(depth_risk, 3),
                "retry_risk": retry_risk,
                "instability_risk": round(instability_risk, 3)
            }
        }
        
        logger.info(f"🧠 [RISK_ENGINE] Calculated CRS: {result['risk_score']} (AST: {ast_risk}, Depth: {depth_risk}, Retry: {retry_risk})")
        return result