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

local_deep_research/advanced_search_system/strategies/improved_evidence_based_strategy.py

@@ -0,0 +1,782 @@
+"""
+Improved evidence-based search strategy for complex query resolution.
+
+Key improvements:
+1. Multi-stage candidate discovery with adaptive query generation
+2. Dynamic constraint combination for cross-constraint searches
+3. Query adaptation based on partial results
+4. Enhanced source diversity management
+"""
+
+import itertools
+from collections import defaultdict
+from dataclasses import dataclass
+from datetime import datetime
+from typing import Any, Dict, List, Set
+
+from langchain_core.language_models import BaseChatModel
+
+from ...utilities.search_utilities import remove_think_tags
+from ..candidates.base_candidate import Candidate
+from ..constraints.base_constraint import Constraint, ConstraintType
+from ..constraints.constraint_analyzer import ConstraintAnalyzer
+from ..evidence.base_evidence import EvidenceType
+from ..evidence.evaluator import EvidenceEvaluator
+from ..findings.repository import FindingsRepository
+from .base_strategy import BaseSearchStrategy
+
+
+@dataclass
+class SearchAttempt:
+    """Track search attempts for query adaptation."""
+
+    query: str
+    constraint_ids: List[str]
+    results_count: int
+    candidates_found: int
+    timestamp: str
+    strategy_type: str  # 'single', 'combined', 'exploratory'
+
+
+class ImprovedEvidenceBasedStrategy(BaseSearchStrategy):
+    """
+    Improved evidence-based strategy with adaptive search capabilities.
+
+    Key improvements:
+    1. Multi-stage candidate discovery
+    2. Adaptive query generation based on results
+    3. Cross-constraint search optimization
+    4. Source diversity tracking and enhancement
+    """
+
+    def __init__(
+        self,
+        model: BaseChatModel,
+        search: Any,
+        all_links_of_system: List[str],
+        max_iterations: int = 20,
+        confidence_threshold: float = 0.85,
+        candidate_limit: int = 15,  # Increased for better diversity
+        evidence_threshold: float = 0.6,
+        max_search_iterations: int = 3,
+        questions_per_iteration: int = 3,
+        min_source_diversity: int = 3,  # Minimum different sources
+        adaptive_query_count: int = 3,  # Number of adaptive queries per stage
+    ):
+        """Initialize the improved evidence-based strategy."""
+        super().__init__(all_links_of_system)
+        self.model = model
+        self.search = search
+        self.max_iterations = max_iterations
+        self.confidence_threshold = confidence_threshold
+        self.candidate_limit = candidate_limit
+        self.evidence_threshold = evidence_threshold
+        self.max_search_iterations = max_search_iterations
+        self.questions_per_iteration = questions_per_iteration
+        self.min_source_diversity = min_source_diversity
+        self.adaptive_query_count = adaptive_query_count
+
+        # Initialize components
+        self.constraint_analyzer = ConstraintAnalyzer(model)
+        self.evidence_evaluator = EvidenceEvaluator(model)
+        self.findings_repository = FindingsRepository(model)
+
+        # State tracking
+        self.constraints: List[Constraint] = []
+        self.candidates: List[Candidate] = []
+        self.search_history: List[Dict] = []
+        self.search_attempts: List[SearchAttempt] = []
+        self.failed_queries: Set[str] = set()
+        self.successful_patterns: List[Dict[str, Any]] = []
+        self.source_types: Dict[str, Set[str]] = defaultdict(set)
+        self.iteration: int = 0
+
+    def analyze_topic(self, query: str) -> Dict:
+        """Analyze a topic using improved evidence-based approach."""
+        # Initialize
+        self.all_links_of_system.clear()
+        self.questions_by_iteration = []
+        self.findings = []
+        self.iteration = 0
+        self.search_attempts.clear()
+        self.failed_queries.clear()
+        self.successful_patterns.clear()
+
+        # Step 1: Extract initial constraints
+        if self.progress_callback:
+            self.progress_callback(
+                "Analyzing query for constraint extraction...",
+                2,
+                {"phase": "constraint_analysis", "status": "starting"},
+            )
+
+        self.constraints = self.constraint_analyzer.extract_constraints(query)
+
+        # Step 2: Multi-stage candidate discovery
+        self._multi_stage_candidate_discovery()
+
+        # Step 3: Main evidence-gathering loop with adaptive search
+        while (
+            self.iteration < self.max_iterations
+            and not self._has_sufficient_answer()
+        ):
+            self.iteration += 1
+
+            if self.progress_callback:
+                progress = 15 + int((self.iteration / self.max_iterations) * 70)
+                self.progress_callback(
+                    f"Iteration {self.iteration}/{self.max_iterations} - {self._get_iteration_status()}",
+                    progress,
+                    {
+                        "phase": "iteration_start",
+                        "iteration": self.iteration,
+                        "candidates_count": len(self.candidates),
+                        "search_attempts": len(self.search_attempts),
+                        "successful_patterns": len(self.successful_patterns),
+                    },
+                )
+
+            # Adaptive evidence gathering
+            self._adaptive_evidence_gathering()
+
+            # Score and prune with diversity consideration
+            self._score_with_diversity()
+
+            # Adaptive candidate discovery if needed
+            if len(self.candidates) < 3 or self._needs_diversity():
+                self._adaptive_candidate_search()
+
+        # Step 4: Cross-validation and final verification
+        self._cross_validate_candidates()
+
+        # Step 5: Generate final answer
+        return self._synthesize_final_answer(query)
+
+    def _multi_stage_candidate_discovery(self):
+        """Multi-stage candidate discovery with different strategies."""
+        stages = [
+            ("distinctive", self._discover_with_distinctive_constraints),
+            ("combined", self._discover_with_combined_constraints),
+            ("exploratory", self._discover_with_exploratory_search),
+            ("pattern_based", self._discover_with_pattern_matching),
+        ]
+
+        for stage_name, discovery_method in stages:
+            if self.progress_callback:
+                self.progress_callback(
+                    f"Stage {stage_name}: Discovering candidates...",
+                    5 + stages.index((stage_name, discovery_method)) * 3,
+                    {"phase": "candidate_discovery", "stage": stage_name},
+                )
+
+            new_candidates = discovery_method()
+
+            # Add unique candidates
+            existing_names = {c.name.lower() for c in self.candidates}
+            for candidate in new_candidates:
+                if candidate.name.lower() not in existing_names:
+                    self.candidates.append(candidate)
+                    existing_names.add(candidate.name.lower())
+
+            # Stop if we have enough diverse candidates
+            if len(self.candidates) >= self.candidate_limit // 2:
+                break
+
+    def _discover_with_distinctive_constraints(self) -> List[Candidate]:
+        """Discover candidates using most distinctive constraints."""
+        distinctive = self._get_adaptive_distinctive_constraints()
+        candidates = []
+
+        for constraint_combo in self._generate_constraint_combinations(
+            distinctive, max_size=3
+        ):
+            query = self._create_adaptive_search_query(constraint_combo)
+            if query not in self.failed_queries:
+                results = self._execute_tracked_search(
+                    query, constraint_combo, "distinctive"
+                )
+                candidates.extend(
+                    self._extract_candidates_with_context(results, query)
+                )
+
+                if candidates:  # Track successful patterns
+                    self.successful_patterns.append(
+                        {
+                            "constraints": [c.id for c in constraint_combo],
+                            "query_pattern": query,
+                            "candidates_found": len(candidates),
+                        }
+                    )
+
+        return candidates
+
+    def _discover_with_combined_constraints(self) -> List[Candidate]:
+        """Discover candidates using strategic constraint combinations."""
+        # Combine constraints from different types for better results
+        type_groups = defaultdict(list)
+        for c in self.constraints:
+            type_groups[c.type].append(c)
+
+        candidates = []
+        # Cross-type combinations
+        for type1, type2 in itertools.combinations(type_groups.keys(), 2):
+            for c1, c2 in itertools.product(
+                type_groups[type1][:2], type_groups[type2][:2]
+            ):
+                query = self._create_cross_constraint_query([c1, c2])
+                results = self._execute_tracked_search(
+                    query, [c1, c2], "combined"
+                )
+                candidates.extend(
+                    self._extract_candidates_with_context(results, query)
+                )
+
+        return candidates
+
+    def _discover_with_exploratory_search(self) -> List[Candidate]:
+        """Use exploratory searches to find unexpected candidates."""
+        candidates = []
+
+        # Generate exploratory queries
+        exploratory_prompt = f"""
+Based on these constraints, generate 3 exploratory search queries that might find relevant candidates:
+
+Constraints:
+{self._format_constraints_for_prompt(self.constraints[:5])}
+
+Create queries that:
+1. Use alternative phrasings or related concepts
+2. Explore edge cases or unusual combinations
+3. Look for historical or contextual matches
+
+Return only the queries, one per line.
+"""
+
+        response = self.model.invoke(exploratory_prompt)
+        queries = remove_think_tags(response.content).strip().split("\n")
+
+        for query in queries[:3]:
+            if query.strip() and query not in self.failed_queries:
+                results = self._execute_tracked_search(
+                    query, self.constraints[:3], "exploratory"
+                )
+                candidates.extend(
+                    self._extract_candidates_with_context(results, query)
+                )
+
+        return candidates
+
+    def _discover_with_pattern_matching(self) -> List[Candidate]:
+        """Use pattern matching based on successful patterns."""
+        if not self.successful_patterns:
+            return []
+
+        candidates = []
+
+        # Adapt successful patterns
+        for pattern in self.successful_patterns[:3]:
+            constraint_ids = pattern["constraints"]
+            constraints = [
+                c for c in self.constraints if c.id in constraint_ids
+            ]
+
+            # Create variations of successful queries
+            adapted_query = self._adapt_successful_query(
+                pattern["query_pattern"], constraints
+            )
+            results = self._execute_tracked_search(
+                adapted_query, constraints, "pattern_based"
+            )
+            candidates.extend(
+                self._extract_candidates_with_context(results, adapted_query)
+            )
+
+        return candidates
+
+    def _adaptive_evidence_gathering(self):
+        """Gather evidence with adaptive query generation."""
+        for candidate in self.candidates[:5]:
+            unverified = candidate.get_unverified_constraints(self.constraints)
+
+            if not unverified:
+                continue
+
+            # Sort by weight and group by type
+            unverified.sort(key=lambda c: c.weight, reverse=True)
+            type_groups = defaultdict(list)
+            for c in unverified:
+                type_groups[c.type].append(c)
+
+            # Try different evidence gathering strategies
+            for constraint_type, constraints in type_groups.items():
+                # Try single constraint
+                for c in constraints[:2]:
+                    query = self._create_evidence_query(candidate, [c])
+                    results = self._execute_tracked_search(
+                        query, [c], "evidence"
+                    )
+                    evidence = self.evidence_evaluator.extract_evidence(
+                        results.get("current_knowledge", ""), candidate.name, c
+                    )
+                    candidate.add_evidence(c.id, evidence)
+
+                # Try combined constraints of same type
+                if len(constraints) > 1:
+                    query = self._create_evidence_query(
+                        candidate, constraints[:2]
+                    )
+                    results = self._execute_tracked_search(
+                        query, constraints[:2], "evidence_combined"
+                    )
+
+                    for c in constraints[:2]:
+                        evidence = self.evidence_evaluator.extract_evidence(
+                            results.get("current_knowledge", ""),
+                            candidate.name,
+                            c,
+                        )
+                        if (
+                            c.id not in candidate.evidence
+                            or evidence.confidence
+                            > candidate.evidence[c.id].confidence
+                        ):
+                            candidate.add_evidence(c.id, evidence)
+
+    def _create_adaptive_search_query(
+        self, constraints: List[Constraint]
+    ) -> str:
+        """Create adaptive search queries based on past performance."""
+        # Check if similar constraint combinations have been successful
+        constraint_ids = {c.id for c in constraints}
+
+        for pattern in self.successful_patterns:
+            if (
+                len(constraint_ids.intersection(pattern["constraints"]))
+                >= len(constraint_ids) // 2
+            ):
+                # Adapt successful pattern
+                return self._adapt_successful_query(
+                    pattern["query_pattern"], constraints
+                )
+
+        # Check failed queries to avoid repetition
+        base_query = self._create_base_search_query(constraints)
+        if base_query in self.failed_queries:
+            # Modify query to avoid failure
+            return self._modify_failed_query(base_query, constraints)
+
+        return base_query
+
+    def _create_cross_constraint_query(
+        self, constraints: List[Constraint]
+    ) -> str:
+        """Create queries that leverage relationships between constraints."""
+        prompt = f"""
+Create a search query that finds candidates satisfying BOTH/ALL of these constraints:
+
+{self._format_constraints_for_prompt(constraints)}
+
+The query should:
+1. Find entities that match both/all constraints
+2. Use operators to require both/all conditions
+3. Focus on finding specific names or entities
+
+Return only the search query.
+"""
+
+        response = self.model.invoke(prompt)
+        return remove_think_tags(response.content).strip()
+
+    def _create_evidence_query(
+        self, candidate: Candidate, constraints: List[Constraint]
+    ) -> str:
+        """Create targeted evidence queries."""
+        constraint_desc = self._format_constraints_for_prompt(constraints)
+
+        prompt = f"""
+Create a search query to verify if "{candidate.name}" satisfies these constraints:
+
+{constraint_desc}
+
+The query should:
+1. Include the candidate name
+2. Target the specific constraint requirements
+3. Find factual evidence, not opinions
+
+Return only the search query.
+"""
+
+        response = self.model.invoke(prompt)
+        return remove_think_tags(response.content).strip()
+
+    def _score_with_diversity(self):
+        """Score candidates considering source diversity."""
+        for candidate in self.candidates:
+            # Base score from evidence
+            candidate.calculate_score(self.constraints)
+
+            # Diversity bonus
+            diversity_score = self._calculate_diversity_score(candidate)
+            candidate.score = 0.8 * candidate.score + 0.2 * diversity_score
+
+            # Track source types
+            for evidence in candidate.evidence.values():
+                if hasattr(evidence, "source"):
+                    self.source_types[candidate.name].add(evidence.source)
+
+        # Sort by adjusted score
+        self.candidates.sort(key=lambda c: c.score, reverse=True)
+
+        # Prune while maintaining some diversity
+        self._prune_with_diversity()
+
+    def _cross_validate_candidates(self):
+        """Cross-validate top candidates using different approaches."""
+        if not self.candidates:
+            return
+
+        top_candidates = self.candidates[:3]
+
+        for candidate in top_candidates:
+            # Validate using different search engines or approaches
+            validation_queries = self._generate_validation_queries(candidate)
+
+            for query in validation_queries:
+                results = self._execute_tracked_search(
+                    query, self.constraints, "validation"
+                )
+
+                # Update evidence if better found
+                for constraint in self.constraints:
+                    evidence = self.evidence_evaluator.extract_evidence(
+                        results.get("current_knowledge", ""),
+                        candidate.name,
+                        constraint,
+                    )
+
+                    if (
+                        constraint.id not in candidate.evidence
+                        or evidence.confidence
+                        > candidate.evidence[constraint.id].confidence
+                    ):
+                        candidate.add_evidence(constraint.id, evidence)
+
+    def _execute_tracked_search(
+        self, query: str, constraints: List[Constraint], strategy_type: str
+    ) -> Dict:
+        """Execute search with tracking for adaptation."""
+        results = self._execute_search(query)
+
+        # Track the attempt
+        candidates_found = len(
+            self._extract_candidates_with_context(results, query)
+        )
+        attempt = SearchAttempt(
+            query=query,
+            constraint_ids=[c.id for c in constraints],
+            results_count=len(results.get("all_links_of_system", [])),
+            candidates_found=candidates_found,
+            timestamp=datetime.utcnow().isoformat(),
+            strategy_type=strategy_type,
+        )
+        self.search_attempts.append(attempt)
+
+        # Mark as failed if no results
+        if candidates_found == 0:
+            self.failed_queries.add(query)
+
+        return results
+
+    def _needs_diversity(self) -> bool:
+        """Check if we need more diverse candidates."""
+        if len(self.candidates) < 3:
+            return True
+
+        # Check source diversity
+        top_sources = self.source_types.get(self.candidates[0].name, set())
+        return len(top_sources) < self.min_source_diversity
+
+    def _generate_constraint_combinations(
+        self, constraints: List[Constraint], max_size: int = 3
+    ) -> List[List[Constraint]]:
+        """Generate strategic constraint combinations."""
+        combinations = []
+
+        # Single constraints
+        combinations.extend([[c] for c in constraints])
+
+        # Pairs
+        for size in range(2, min(len(constraints), max_size) + 1):
+            for combo in itertools.combinations(constraints, size):
+                combinations.append(list(combo))
+
+        return combinations
+
+    def _format_constraints_for_prompt(
+        self, constraints: List[Constraint]
+    ) -> str:
+        """Format constraints for LLM prompts."""
+        formatted = []
+        for c in constraints:
+            formatted.append(
+                f"- {c.type.value}: {c.description} (weight: {c.weight:.2f})"
+            )
+        return "\n".join(formatted)
+
+    def _adapt_successful_query(
+        self, pattern_query: str, constraints: List[Constraint]
+    ) -> str:
+        """Adapt a successful query pattern with new constraints."""
+        prompt = f"""
+Adapt this successful search query pattern with new constraints:
+
+Original query: {pattern_query}
+
+New constraints:
+{self._format_constraints_for_prompt(constraints)}
+
+Create a similar query structure but with the new constraint values.
+Return only the adapted query.
+"""
+
+        response = self.model.invoke(prompt)
+        return remove_think_tags(response.content).strip()
+
+    def _modify_failed_query(
+        self, failed_query: str, constraints: List[Constraint]
+    ) -> str:
+        """Modify a failed query to try a different approach."""
+        prompt = f"""
+This search query returned no results: {failed_query}
+
+Constraints we're trying to satisfy:
+{self._format_constraints_for_prompt(constraints)}
+
+Create an alternative query that:
+1. Uses different keywords or phrases
+2. Tries a different search approach
+3. Still targets the same constraints
+
+Return only the modified query.
+"""
+
+        response = self.model.invoke(prompt)
+        return remove_think_tags(response.content).strip()
+
+    def _calculate_diversity_score(self, candidate: Candidate) -> float:
+        """Calculate diversity score for a candidate."""
+        if not candidate.evidence:
+            return 0.0
+
+        # Source diversity
+        sources = self.source_types.get(candidate.name, set())
+        source_score = min(len(sources) / self.min_source_diversity, 1.0)
+
+        # Evidence type diversity
+        evidence_types = {e.type for e in candidate.evidence.values()}
+        type_score = len(evidence_types) / len(EvidenceType)
+
+        # Confidence distribution (prefer balanced confidence)
+        confidences = [e.confidence for e in candidate.evidence.values()]
+        if confidences:
+            variance = sum((c - 0.7) ** 2 for c in confidences) / len(
+                confidences
+            )
+            confidence_score = 1.0 / (1.0 + variance)
+        else:
+            confidence_score = 0.0
+
+        return (source_score + type_score + confidence_score) / 3.0
+
+    def _prune_with_diversity(self):
+        """Prune candidates while maintaining diversity."""
+        if len(self.candidates) <= self.candidate_limit:
+            return
+
+        # Keep top candidates
+        kept = self.candidates[: self.candidate_limit // 2]
+        remaining = self.candidates[self.candidate_limit // 2 :]
+
+        # Add diverse candidates from remaining
+        for candidate in remaining:
+            if len(kept) >= self.candidate_limit:
+                break
+
+            # Check if this candidate adds diversity
+            if self._adds_diversity(candidate, kept):
+                kept.append(candidate)
+
+        # Fill remaining slots with highest scoring
+        for candidate in remaining:
+            if len(kept) >= self.candidate_limit:
+                break
+            if candidate not in kept:
+                kept.append(candidate)
+
+        self.candidates = kept
+
+    def _adds_diversity(
+        self, candidate: Candidate, existing: List[Candidate]
+    ) -> bool:
+        """Check if a candidate adds diversity to the existing set."""
+        # Check source diversity
+        candidate_sources = self.source_types.get(candidate.name, set())
+        existing_sources = set()
+        for c in existing:
+            existing_sources.update(self.source_types.get(c.name, set()))
+
+        new_sources = candidate_sources - existing_sources
+        if new_sources:
+            return True
+
+        # Check constraint coverage
+        candidate_constraints = set(candidate.evidence.keys())
+        existing_constraints = set()
+        for c in existing:
+            existing_constraints.update(c.evidence.keys())
+
+        new_constraints = candidate_constraints - existing_constraints
+        return len(new_constraints) > 0
+
+    def _generate_validation_queries(self, candidate: Candidate) -> List[str]:
+        """Generate validation queries for cross-checking."""
+        queries = []
+
+        # Query combining multiple constraints
+        high_weight_constraints = sorted(
+            self.constraints, key=lambda c: c.weight, reverse=True
+        )[:3]
+        combined_query = f'"{candidate.name}" ' + " ".join(
+            c.to_search_terms() for c in high_weight_constraints
+        )
+        queries.append(combined_query)
+
+        # Query with alternative phrasing
+        alt_prompt = f"""
+Create an alternative search query to validate "{candidate.name}" as the answer.
+Use different keywords but same intent.
+
+Return only the query.
+"""
+        response = self.model.invoke(alt_prompt)
+        queries.append(remove_think_tags(response.content).strip())
+
+        # Source-specific query
+        if self.source_types.get(candidate.name):
+            source_query = f'"{candidate.name}" site:{list(self.source_types[candidate.name])[0]}'
+            queries.append(source_query)
+
+        return queries
+
+    def _extract_candidates_with_context(
+        self, results: Dict, query: str
+    ) -> List[Candidate]:
+        """Extract candidates with context awareness."""
+        # Use the original extraction method but with context
+        candidates = self._extract_candidates_from_results(results, query)
+
+        # Add context metadata
+        for candidate in candidates:
+            candidate.metadata["discovery_query"] = query
+            candidate.metadata["discovery_stage"] = self.iteration
+
+        return candidates
+
+    def _create_base_search_query(self, constraints: List[Constraint]) -> str:
+        """Create a base search query from constraints."""
+        # Use an improved prompt that considers constraint relationships
+        prompt = f"""
+Create a search query that finds specific entities satisfying these constraints:
+
+{self._format_constraints_for_prompt(constraints)}
+
+Guidelines:
+1. Focus on finding names/entities, not general information
+2. Use the most distinctive constraints
+3. Combine constraints effectively
+4. Keep the query concise but comprehensive
+
+Return only the search query.
+"""
+
+        response = self.model.invoke(prompt)
+        return remove_think_tags(response.content).strip()
+
+    def _adaptive_candidate_search(self):
+        """Adaptively search for more candidates based on current state."""
+        # Analyze what types of candidates we're missing
+        covered_constraints = set()
+        for candidate in self.candidates[:5]:
+            covered_constraints.update(candidate.evidence.keys())
+
+        uncovered = [
+            c for c in self.constraints if c.id not in covered_constraints
+        ]
+
+        if uncovered:
+            # Search specifically for uncovered constraints
+            queries = []
+            for constraint_group in self._generate_constraint_combinations(
+                uncovered[:5], max_size=2
+            ):
+                query = self._create_adaptive_search_query(constraint_group)
+                queries.append((query, constraint_group))
+
+            for query, constraints in queries[: self.adaptive_query_count]:
+                results = self._execute_tracked_search(
+                    query, constraints, "adaptive"
+                )
+                new_candidates = self._extract_candidates_with_context(
+                    results, query
+                )
+
+                # Add unique candidates
+                existing_names = {c.name.lower() for c in self.candidates}
+                for candidate in new_candidates:
+                    if candidate.name.lower() not in existing_names:
+                        self.candidates.append(candidate)
+                        existing_names.add(candidate.name.lower())
+
+    def _get_adaptive_distinctive_constraints(self) -> List[Constraint]:
+        """Get distinctive constraints with adaptive prioritization."""
+        # Start with basic prioritization
+        priority_order = [
+            ConstraintType.NAME_PATTERN,
+            ConstraintType.LOCATION,
+            ConstraintType.EVENT,
+            ConstraintType.STATISTIC,
+            ConstraintType.COMPARISON,
+            ConstraintType.PROPERTY,
+            ConstraintType.TEMPORAL,
+            ConstraintType.EXISTENCE,
+        ]
+
+        # Adjust based on successful patterns
+        if self.successful_patterns:
+            # Count successful constraint types
+            type_success = defaultdict(int)
+            for pattern in self.successful_patterns:
+                for constraint_id in pattern["constraints"]:
+                    constraint = next(
+                        (c for c in self.constraints if c.id == constraint_id),
+                        None,
+                    )
+                    if constraint:
+                        type_success[constraint.type] += pattern[
+                            "candidates_found"
+                        ]
+
+            # Sort by success rate
+            priority_order = sorted(
+                priority_order,
+                key=lambda t: type_success.get(t, 0),
+                reverse=True,
+            )
+
+        # Sort constraints by adjusted priority
+        sorted_constraints = sorted(
+            self.constraints,
+            key=lambda c: (priority_order.index(c.type), -c.weight),
+        )
+
+        return sorted_constraints[:5]