Citation-Aware Retrieval

Assigns a numerical weight to each legal document based on its position in the judicial hierarchy and subsequent treatment history.

• Court Hierarchy Weighting: Decisions from a supreme court receive a higher base score than those from a trial court.
• Treatment History Analysis: A case that has been positively treated or followed gains authority; a case that has been overruled or questioned is penalized.
• Jurisdictional Relevance: Binding authority within the query's jurisdiction is weighted higher than persuasive authority from a foreign jurisdiction.

Maps the entire corpus of case law as a directed graph where nodes are cases and edges are citations. This structure enables algorithmic traversal to compute authority metrics.

• In-Degree Centrality: Foundational cases like Marbury v. Madison have a high in-degree, signaling their importance.
• PageRank Variants: Algorithms adapted from web search identify the most authoritative nodes in the legal citation network.
• Temporal Analysis: The graph structure reveals how a precedent's influence grows, stabilizes, or decays over time.

The computational process of automatically mapping the subsequent treatment history of a case to determine its current precedential value.

• Negative Treatment Flags: Automatically detects signals like 'Overruled', 'Reversed', or 'Disapproved' in citing decisions.
• Depth of Treatment: Distinguishes between a passing citation and an extensive, substantive discussion of the precedent.
• Risk Classification: Assigns a 'red flag' warning to cases with negative treatment, preventing a model from relying on bad law.

A scoring function that modulates relevance based on the age of a legal document, reflecting the evolution of statutory and judicial interpretation.

• Statutory Obsolescence: Older interpretations of frequently amended statutes are decayed more aggressively.
• Constitutional Persistence: Foundational constitutional decisions are exempt from decay, as their authority does not diminish with time.
• Configurable Half-Life: The decay curve is a tunable parameter, allowing the system to be optimized for fast-moving regulatory areas versus stable common law doctrines.

Combines the authority score from the citation network with the semantic relevance score from vector search to produce a final ranked list.

• Weighted Linear Combination: The final score is a blend: FinalScore = α * SemanticRelevance + β * AuthorityScore.
• Reciprocal Rank Fusion (RRF): A non-parametric method that merges the ranked lists from semantic and authority indexes without needing to normalize disparate score distributions.
• Re-Ranking Cascade: An initial semantic retrieval fetches a broad candidate set, and a subsequent authority-aware re-ranker orders them to place binding precedent at the top.

A hard constraint applied during retrieval that limits the candidate document set to a specific sovereign entity or geographic court system.

• Prevents Contamination: Ensures a query about California contract law is not answered with a New York precedent.
• Hierarchical Scoping: Can be set to a specific district court, a circuit court, or a state supreme court.
• Mandatory vs. Persuasive Partitioning: Retrieves binding authority first, then supplements with persuasive authority from other jurisdictions only if explicitly requested or if binding authority is sparse.

A weighting algorithm that assigns numerical value to legal documents based on court hierarchy, treatment history, and jurisdictional relevance. The scoring function ensures that binding authority from higher courts is surfaced before persuasive authority from lower or foreign tribunals.

• Weights Supreme Court decisions above appellate rulings
• Depreciates scores for cases flagged as 'overruled' or 'questioned'
• Incorporates depth of treatment metrics from citing cases

The computational process of automatically mapping the subsequent treatment history of a case to determine if its holdings remain good law. This is a critical pre-retrieval or post-retrieval filter that prevents citation-aware systems from surfacing authority that has been implicitly or explicitly overturned.

• Detects negative treatment signals: overruled, superseded, questioned
• Identifies positive treatment signals: followed, affirmed, relied upon
• Integrates with temporal decay weighting for recency-aware scoring

The computational mapping and traversal of legal authority graphs to identify the most central and influential precedents. Algorithms like PageRank variants adapted for legal citation networks surface foundational cases that serve as hubs in the jurisprudential landscape.

• Measures in-degree centrality to identify landmark cases
• Detects citation clusters representing doctrinal lines
• Enables discovery of seminal authority even when not lexically similar to the query

A retrieval constraint that limits search results to legal documents originating from a specific sovereign entity or geographic court system. This prevents cross-jurisdictional contamination where a persuasive case from one circuit is mistakenly treated as binding in another.

• Filters by court level, circuit, and state/federal distinction
• Maintains separate authority indices per jurisdiction
• Critical for multi-jurisdictional practice areas requiring precise venue-aware retrieval

A scoring function that reduces the relevance of older legal documents to account for the evolution of statutory law and judicial interpretation. The decay curve is not uniform—binding precedents that have never been overturned retain authority regardless of age, while persuasive authority depreciates more aggressively.

• Applies exponential or logarithmic decay based on document type
• Exempts landmark constitutional cases from decay
• Integrates with Shepardizing signals to override decay for overruled cases

A retrieval-augmented generation approach that uses a knowledge graph of legal entities and citations to retrieve community summaries of related documents rather than raw text chunks. This shifts retrieval from lexical or semantic similarity to structural authority proximity.

• Traverses citation edges to gather precedent clusters
• Retrieves graph-community summaries synthesized from related cases
• Combines graph-based authority signals with vector similarity for hybrid ranking