local-deep-research 0.4.4__py3-none-any.whl → 0.5.2__py3-none-any.whl

local_deep_research/advanced_search_system/constraint_checking/constraint_checker.py

@@ -0,0 +1,223 @@
+"""
+Main constraint checker that orchestrates constraint validation.
+
+This module provides the primary interface for checking candidates against constraints.
+"""
+
+from dataclasses import dataclass
+from typing import Dict, List, Optional
+
+from langchain_core.language_models import BaseChatModel
+from loguru import logger
+
+from ..candidates.base_candidate import Candidate
+from ..constraints.base_constraint import Constraint
+from .evidence_analyzer import EvidenceAnalyzer
+from .rejection_engine import RejectionEngine, RejectionResult
+
+
+@dataclass
+class ConstraintCheckResult:
+    """Result of checking a candidate against all constraints."""
+
+    candidate: Candidate
+    total_score: float
+    constraint_scores: Dict[str, Dict]
+    rejection_result: Optional[RejectionResult]
+    detailed_results: List[Dict]
+
+
+class ConstraintChecker:
+    """
+    Main constraint checker that validates candidates against constraints.
+
+    This checker:
+    1. Gathers evidence for each constraint
+    2. Analyzes evidence using dual confidence scoring
+    3. Makes rejection decisions based on evidence
+    4. Provides detailed scoring breakdown
+    """
+
+    def __init__(
+        self,
+        model: BaseChatModel,
+        evidence_gatherer=None,  # Will be passed in from strategy
+        negative_threshold: float = 0.25,
+        positive_threshold: float = 0.4,
+        uncertainty_penalty: float = 0.2,
+        negative_weight: float = 0.5,
+    ):
+        """
+        Initialize the constraint checker.
+
+        Args:
+            model: Language model for evidence analysis
+            evidence_gatherer: Function to gather evidence (from strategy)
+            negative_threshold: Rejection threshold for negative evidence
+            positive_threshold: Minimum positive evidence required
+            uncertainty_penalty: Penalty for uncertain evidence
+            negative_weight: Weight for negative evidence in scoring
+        """
+        self.model = model
+        self.evidence_gatherer = evidence_gatherer
+
+        # Initialize components
+        self.evidence_analyzer = EvidenceAnalyzer(model)
+        self.rejection_engine = RejectionEngine(
+            negative_threshold, positive_threshold
+        )
+
+        # Scoring parameters
+        self.uncertainty_penalty = uncertainty_penalty
+        self.negative_weight = negative_weight
+
+    def check_candidate(
+        self, candidate: Candidate, constraints: List[Constraint]
+    ) -> ConstraintCheckResult:
+        """
+        Check a candidate against all constraints.
+
+        Args:
+            candidate: The candidate to check
+            constraints: List of constraints to check against
+
+        Returns:
+            ConstraintCheckResult: Complete evaluation result
+        """
+        logger.info(f"Checking candidate: {candidate.name}")
+
+        constraint_results = {}
+        constraint_scores = {}
+        detailed_results = []
+        total_score = 0.0
+
+        for constraint in constraints:
+            # Gather evidence for this constraint
+            evidence_list = self._gather_evidence_for_constraint(
+                candidate, constraint
+            )
+
+            if evidence_list:
+                # Analyze evidence with dual confidence
+                dual_evidence = [
+                    self.evidence_analyzer.analyze_evidence_dual_confidence(
+                        e, constraint
+                    )
+                    for e in evidence_list
+                ]
+
+                constraint_results[constraint] = dual_evidence
+
+                # Calculate average scores for this constraint
+                avg_positive = sum(
+                    e.positive_confidence for e in dual_evidence
+                ) / len(dual_evidence)
+                avg_negative = sum(
+                    e.negative_confidence for e in dual_evidence
+                ) / len(dual_evidence)
+                avg_uncertainty = sum(
+                    e.uncertainty for e in dual_evidence
+                ) / len(dual_evidence)
+
+                # Calculate constraint score
+                score = self.evidence_analyzer.evaluate_evidence_list(
+                    evidence_list,
+                    constraint,
+                    self.uncertainty_penalty,
+                    self.negative_weight,
+                )
+
+                # Store results
+                constraint_scores[constraint.value] = {
+                    "total": score,
+                    "positive": avg_positive,
+                    "negative": avg_negative,
+                    "uncertainty": avg_uncertainty,
+                    "weight": constraint.weight,
+                }
+
+                detailed_results.append(
+                    {
+                        "constraint": constraint.value,
+                        "score": score,
+                        "positive": avg_positive,
+                        "negative": avg_negative,
+                        "uncertainty": avg_uncertainty,
+                        "weight": constraint.weight,
+                        "type": constraint.type.value,
+                    }
+                )
+
+                # Log result
+                symbol = "✓" if score >= 0.8 else "○" if score >= 0.5 else "✗"
+                logger.info(
+                    f"{symbol} {candidate.name} | {constraint.value}: {int(score * 100)}% "
+                    f"(+{int(avg_positive * 100)}% -{int(avg_negative * 100)}% ?{int(avg_uncertainty * 100)}%)"
+                )
+
+            else:
+                # No evidence found
+                score = 0.5 - self.uncertainty_penalty
+
+                constraint_scores[constraint.value] = {
+                    "total": score,
+                    "positive": 0.0,
+                    "negative": 0.0,
+                    "uncertainty": 1.0,
+                    "weight": constraint.weight,
+                }
+
+                detailed_results.append(
+                    {
+                        "constraint": constraint.value,
+                        "score": score,
+                        "positive": 0.0,
+                        "negative": 0.0,
+                        "uncertainty": 1.0,
+                        "weight": constraint.weight,
+                        "type": constraint.type.value,
+                    }
+                )
+
+                logger.info(
+                    f"? {candidate.name} | {constraint.value}: No evidence found"
+                )
+
+        # Check for rejection
+        rejection_result = self.rejection_engine.check_all_constraints(
+            candidate, constraint_results
+        )
+
+        if rejection_result and rejection_result.should_reject:
+            # Candidate should be rejected
+            total_score = 0.0
+        else:
+            # Calculate weighted average score
+            if detailed_results:
+                weights = [r["weight"] for r in detailed_results]
+                scores = [r["score"] for r in detailed_results]
+                total_score = sum(s * w for s, w in zip(scores, weights)) / sum(
+                    weights
+                )
+
+        logger.info(f"Final score for {candidate.name}: {total_score:.2%}")
+
+        return ConstraintCheckResult(
+            candidate=candidate,
+            total_score=total_score,
+            constraint_scores=constraint_scores,
+            rejection_result=rejection_result,
+            detailed_results=detailed_results,
+        )
+
+    def _gather_evidence_for_constraint(
+        self, candidate: Candidate, constraint: Constraint
+    ) -> List[Dict]:
+        """Gather evidence for a constraint using the provided evidence gatherer."""
+        if self.evidence_gatherer:
+            return self.evidence_gatherer(candidate, constraint)
+        else:
+            logger.warning(
+                "No evidence gatherer provided - cannot gather evidence"
+            )
+            return []

local_deep_research/advanced_search_system/constraint_checking/constraint_satisfaction_tracker.py

@@ -0,0 +1,387 @@
+"""
+Multi-Stage Constraint Verification with Intelligent Weighting
+
+This module implements the constraint satisfaction tracking system outlined in
+BROWSECOMP_IMPROVEMENT_STRATEGY.md to improve BrowseComp performance.
+
+Key features:
+1. Intelligent constraint weighting based on reliability
+2. Progressive constraint verification
+3. Partial satisfaction scoring
+4. Constraint difficulty analysis
+"""
+
+import logging
+from dataclasses import dataclass
+from typing import Dict, List
+
+logger = logging.getLogger(__name__)
+
+
+@dataclass
+class ConstraintVerificationResult:
+    """Result of constraint verification process."""
+
+    total_score: float
+    should_accept: bool
+    constraint_details: Dict[str, Dict]
+    weighted_score: float
+    reliability_adjusted_score: float
+
+
+class ConstraintSatisfactionTracker:
+    """
+    Multi-stage constraint verification with intelligent weighting.
+
+    Implements the strategy from BROWSECOMP_IMPROVEMENT_STRATEGY.md to:
+    - Weight constraints by reliability and importance
+    - Allow partial satisfaction instead of requiring perfect matches
+    - Prioritize reliable constraints over problematic ones
+    """
+
+    def __init__(self, satisfaction_threshold: float = 0.7):
+        self.satisfaction_threshold = satisfaction_threshold
+
+        # Constraint type weights based on reliability and importance
+        self.constraint_weights = {
+            "PROPERTY": 1.0,  # Basic properties, moderately reliable
+            "NAME_PATTERN": 1.5,  # Names are highly identifying
+            "STATISTIC": 0.8,  # Often imprecise/outdated
+            "EVENT": 1.2,  # Events are good identifiers
+            "TEMPORAL": 1.1,  # Dates are useful but sometimes fuzzy
+            "LOCATION": 1.3,  # Geographic info is reliable
+            "COMPARISON": 0.6,  # Relative comparisons often problematic
+            "EXISTENCE": 1.0,  # Basic existence checks
+            "RELATIONSHIP": 0.9,  # Relationships can be complex
+        }
+
+        # Difficulty multipliers (easier constraints get higher weight)
+        self.difficulty_multipliers = {
+            "EASY": 1.2,  # Simple property checks
+            "MEDIUM": 1.0,  # Numerical comparisons
+            "HARD": 0.7,  # Complex relationships, ratios
+        }
+
+    def verify_constraints_progressively(
+        self, candidate: object, constraints: List[object]
+    ) -> ConstraintVerificationResult:
+        """
+        Score each constraint independently with intelligent weighting.
+
+        Args:
+            candidate: The candidate entity to verify
+            constraints: List of constraints to check
+
+        Returns:
+            ConstraintVerificationResult with scoring details
+        """
+        scores = {}
+        total_weight = 0
+
+        for constraint in constraints:
+            # Analyze constraint difficulty
+            difficulty = self.analyze_constraint_difficulty(constraint)
+
+            # Calculate constraint weight
+            weight = self.calculate_constraint_weight(constraint, difficulty)
+
+            # Verify constraint against candidate
+            satisfaction = self.verify_single_constraint(candidate, constraint)
+
+            # Store detailed results
+            scores[
+                constraint.id if hasattr(constraint, "id") else str(constraint)
+            ] = {
+                "satisfaction": satisfaction,
+                "weight": weight,
+                "difficulty": difficulty,
+                "constraint_type": self._get_constraint_type(constraint),
+                "raw_score": satisfaction,
+                "weighted_score": satisfaction * weight,
+            }
+
+            total_weight += weight
+
+        # Calculate weighted satisfaction score
+        if total_weight > 0:
+            weighted_score = (
+                sum(
+                    score["satisfaction"] * score["weight"]
+                    for score in scores.values()
+                )
+                / total_weight
+            )
+        else:
+            weighted_score = 0.0
+
+        # Apply reliability adjustment
+        reliability_adjusted_score = self._apply_reliability_adjustment(
+            weighted_score, scores
+        )
+
+        # Determine if candidate should be accepted
+        should_accept = (
+            reliability_adjusted_score >= self.satisfaction_threshold
+        )
+
+        return ConstraintVerificationResult(
+            total_score=reliability_adjusted_score,
+            should_accept=should_accept,
+            constraint_details=scores,
+            weighted_score=weighted_score,
+            reliability_adjusted_score=reliability_adjusted_score,
+        )
+
+    def analyze_constraint_difficulty(self, constraint: object) -> str:
+        """
+        Classify constraints by difficulty level.
+
+        Returns:
+            'EASY', 'MEDIUM', or 'HARD'
+        """
+        constraint_text = str(constraint).lower()
+
+        # HARD: Complex relationships and ratios
+        hard_indicators = [
+            "times more",
+            "times larger",
+            "times bigger",
+            "ratio",
+            "proportion",
+            "percentage of",
+            "compared to",
+            "relative to",
+            "in relation to",
+            "between",
+            "among",
+            "relationship",
+        ]
+
+        if any(indicator in constraint_text for indicator in hard_indicators):
+            return "HARD"
+
+        # MEDIUM: Numerical comparisons and ranges
+        medium_indicators = [
+            "number",
+            "count",
+            "amount",
+            "quantity",
+            "more than",
+            "less than",
+            "at least",
+            "at most",
+            "approximately",
+            "around",
+            "about",
+            "million",
+            "billion",
+            "thousand",
+        ]
+
+        if any(indicator in constraint_text for indicator in medium_indicators):
+            return "MEDIUM"
+
+        # EASY: Simple property checks
+        return "EASY"
+
+    def calculate_constraint_weight(
+        self, constraint: object, difficulty: str
+    ) -> float:
+        """
+        Calculate the weight for a constraint based on type and difficulty.
+
+        Args:
+            constraint: The constraint object
+            difficulty: Difficulty level ('EASY', 'MEDIUM', 'HARD')
+
+        Returns:
+            Float weight value
+        """
+        # Get base weight from constraint type
+        constraint_type = self._get_constraint_type(constraint)
+        base_weight = self.constraint_weights.get(constraint_type, 1.0)
+
+        # Apply difficulty multiplier
+        difficulty_multiplier = self.difficulty_multipliers.get(difficulty, 1.0)
+
+        # Calculate final weight
+        final_weight = base_weight * difficulty_multiplier
+
+        logger.debug(
+            f"Constraint weight: {constraint_type} × {difficulty} = "
+            f"{base_weight} × {difficulty_multiplier} = {final_weight}"
+        )
+
+        return final_weight
+
+    def verify_single_constraint(
+        self, candidate: object, constraint: object
+    ) -> float:
+        """
+        Verify a single constraint against a candidate.
+
+        This is a placeholder - implement actual verification logic
+        based on your constraint checking system.
+
+        Returns:
+            Float score from 0.0 to 1.0
+        """
+        # TODO: Implement actual constraint verification
+        # This should call your existing constraint checker
+
+        # For now, return a placeholder score
+        # In practice, this would call something like:
+        # return self.constraint_checker.verify(candidate, constraint)
+
+        return 0.5  # Placeholder
+
+    def _get_constraint_type(self, constraint: object) -> str:
+        """Extract constraint type from constraint object."""
+        if hasattr(constraint, "type"):
+            if hasattr(constraint.type, "value"):
+                return constraint.type.value
+            else:
+                return str(constraint.type)
+        elif hasattr(constraint, "constraint_type"):
+            return constraint.constraint_type
+        else:
+            # Try to infer from constraint text
+            constraint_text = str(constraint).lower()
+
+            if any(
+                word in constraint_text
+                for word in ["name", "called", "known as"]
+            ):
+                return "NAME_PATTERN"
+            elif any(
+                word in constraint_text
+                for word in ["location", "country", "city"]
+            ):
+                return "LOCATION"
+            elif any(
+                word in constraint_text
+                for word in ["year", "date", "when", "time"]
+            ):
+                return "TEMPORAL"
+            elif any(
+                word in constraint_text
+                for word in ["number", "count", "amount"]
+            ):
+                return "STATISTIC"
+            elif any(
+                word in constraint_text
+                for word in ["event", "happened", "occurred"]
+            ):
+                return "EVENT"
+            elif any(
+                word in constraint_text
+                for word in ["than", "more", "less", "compared"]
+            ):
+                return "COMPARISON"
+            else:
+                return "PROPERTY"
+
+    def _apply_reliability_adjustment(
+        self, weighted_score: float, constraint_scores: Dict[str, Dict]
+    ) -> float:
+        """
+        Apply reliability adjustment based on constraint mix.
+
+        Boosts score if high-reliability constraints are satisfied,
+        reduces score if only low-reliability constraints match.
+        """
+        # Count high vs low reliability constraints
+        high_reliability_count = 0
+        high_reliability_satisfied = 0
+        low_reliability_count = 0
+        low_reliability_satisfied = 0
+
+        high_reliability_types = {
+            "NAME_PATTERN",
+            "LOCATION",
+            "EVENT",
+            "TEMPORAL",
+        }
+        low_reliability_types = {"COMPARISON", "STATISTIC"}
+
+        for constraint_id, score_data in constraint_scores.items():
+            constraint_type = score_data.get("constraint_type", "PROPERTY")
+            satisfaction = score_data.get("satisfaction", 0.0)
+
+            if constraint_type in high_reliability_types:
+                high_reliability_count += 1
+                if satisfaction > 0.5:
+                    high_reliability_satisfied += 1
+            elif constraint_type in low_reliability_types:
+                low_reliability_count += 1
+                if satisfaction > 0.5:
+                    low_reliability_satisfied += 1
+
+        # Calculate reliability bonus/penalty
+        reliability_adjustment = 0.0
+
+        # Bonus for satisfying high-reliability constraints
+        if high_reliability_count > 0:
+            high_reliability_ratio = (
+                high_reliability_satisfied / high_reliability_count
+            )
+            reliability_adjustment += high_reliability_ratio * 0.1
+
+        # Small penalty if only low-reliability constraints are satisfied
+        if low_reliability_count > 0 and high_reliability_satisfied == 0:
+            low_reliability_ratio = (
+                low_reliability_satisfied / low_reliability_count
+            )
+            if low_reliability_ratio > 0.7:  # Many low-reliability matches
+                reliability_adjustment -= 0.05
+
+        # Apply adjustment
+        adjusted_score = weighted_score + reliability_adjustment
+
+        # Ensure score stays in valid range
+        return max(0.0, min(1.0, adjusted_score))
+
+    def get_constraint_priorities(self) -> Dict[str, int]:
+        """
+        Get constraint priorities for relaxation strategies.
+
+        Returns priorities from 1 (relax first) to 10 (never relax).
+        """
+        return {
+            "COMPARISON": 1,  # Frequently relax
+            "STATISTIC": 3,  # Often relax
+            "PROPERTY": 6,
+            "EVENT": 5,
+            "TEMPORAL": 7,
+            "LOCATION": 8,
+            "EXISTENCE": 9,  # Rarely relax
+            "NAME_PATTERN": 10,  # Never relax
+        }
+
+    def suggest_constraint_relaxation(
+        self, constraints: List[object], current_results: List[object]
+    ) -> List[List[object]]:
+        """
+        Suggest progressive constraint relaxation based on results.
+
+        Returns list of relaxed constraint sets to try.
+        """
+        if len(current_results) > 5:
+            return [constraints]  # Enough results, no relaxation needed
+
+        priorities = self.get_constraint_priorities()
+
+        # Sort constraints by relaxation priority (lowest first)
+        relaxable_constraints = sorted(
+            constraints,
+            key=lambda c: priorities.get(self._get_constraint_type(c), 5),
+        )
+
+        # Generate progressive relaxation sets
+        relaxed_sets = []
+        for i in range(1, min(len(constraints), 4)):  # Max 3 relaxation levels
+            relaxed_set = relaxable_constraints[:-i]  # Remove i lowest priority
+            if len(relaxed_set) >= 2:  # Keep at least 2 constraints
+                relaxed_sets.append(relaxed_set)
+
+        return relaxed_sets