Citation Graph

A legal citation graph is inherently directed, with edges pointing from a citing case to a cited authority. Critically, it is largely acyclic due to temporal constraints—a 2024 decision cannot be cited by a 1920 precedent. This temporal directionality enables topological sorting algorithms to trace the evolution of a doctrine without infinite loops.

• Temporal Constraint: Edges only point backward in time.
• Computational Benefit: Enables efficient traversal without cycle-detection overhead.
• Exception: Simultaneous decisions within the same court term can create small cycles.

Authority is not binary; it propagates through the graph based on court hierarchy and subsequent treatment. A citation from the U.S. Supreme Court carries more weight than one from a district court. Negative treatment—such as an overruling—acts as a 'weight cut' that severs the authoritative link, a process algorithmically analogous to PageRank but with jurisdictional constraints.

• Hierarchical Weighting: Edges are weighted by the citing court's level.
• Negative Treatment: Acts as a logical NOT gate, invalidating downstream authority.
• Persuasive vs. Binding: Cross-jurisdictional citations form weaker, non-mandatory edges.

Citation graphs naturally form hub and authority nodes, a pattern identified by Kleinberg's HITS algorithm. A seminal case like Marbury v. Madison acts as an authority node, receiving a massive indegree of citations. A treatise or a case that extensively surveys the law acts as a hub, pointing outward to many authorities.

• Seminal Case Detection: High indegree centrality identifies landmark precedents.
• Bibliometric Coupling: Two cases citing the same authority share a doctrinal lineage.
• Co-citation Analysis: Two cases frequently cited together reveal implicit doctrinal clusters.

Citation patterns are not uniform over time. A sudden burst of citations to an old case often signals a doctrinal revival. Conversely, a sharp decline in citations—a 'citation decay curve'—can indicate a case falling into desuetude or being silently superseded. Analyzing the temporal distribution of citing edges allows models to detect shifts in legal interpretation before they are explicitly announced.

• Burst Detection: Identifies sudden relevance of a dormant precedent.
• Decay Curves: Quantifies the obsolescence of authority.
• Time-Series Clustering: Groups cases by their citation lifecycle patterns.

In a sophisticated citation graph, edges are not merely 'cites' but are encoded with treatment vectors. An edge can be labeled as 'Followed,' 'Distinguished,' 'Criticized,' or 'Overruled.' This transforms the graph from a simple topology into a semantic network where the nature of the relationship is computationally accessible, enabling queries like 'Find all cases that followed Roe but were later criticized.'

• Edge Labeling: NLP classifiers assign treatment types to each citation.
• Sentiment Analysis: Positive, negative, or neutral treatment signals.
• Query Enrichment: Enables path-finding through specific treatment chains.

The global legal graph is partitioned into jurisdictional subgraphs (e.g., 9th Circuit, Delaware Chancery). Edges that cross these partitions are 'persuasive' rather than 'binding.' A binding authority check is a graph traversal problem: is there a directed path from the cited authority to the citing court that stays entirely within the same hierarchical appellate chain?

• Subgraph Isolation: Enables filtering by controlling jurisdiction.
• Cross-Partition Edges: Flagged as persuasive, not mandatory.
• Path Validation: Algorithmically confirms binding vs. persuasive status.

Citation graphs enable algorithmic pathfinding between legal authorities, tracing how a specific legal principle flows from a seminal case through subsequent decisions. This allows AI systems to automatically construct the persuasive authority chain for any legal proposition.

• Breadth-first traversal identifies all cases citing a target authority within N degrees of separation
• Shortest-path algorithms find the most direct precedential link between two seemingly unrelated cases
• Temporal weighting ensures recent treatments are prioritized over older, potentially superseded citations

Graph centrality metrics quantify a case's precedential weight by analyzing its position within the citation network. This moves beyond simple citation counts to measure true jurisprudential influence.

• PageRank variants adapted for legal graphs identify seminal cases that are cited by other highly-cited authorities
• Betweenness centrality detects cases that serve as critical bridges between otherwise disconnected doctrinal clusters
• Eigenvector centrality surfaces authorities that derive influence from being cited by influential subsequent decisions

Community detection algorithms partition the citation graph into doctrinal clusters—groups of cases that cite each other around shared legal principles. This enables automated identification of distinct lines of authority.

• Louvain modularity optimization groups cases into coherent doctrinal communities without pre-labeled categories
• Overlapping community detection accommodates cases that contribute to multiple legal doctrines simultaneously
• Cluster evolution tracking reveals how legal doctrines merge, split, or fade over decades of jurisprudence

Citation graphs model how negative treatment cascades through the precedent network. When a case is overruled, its downstream citing cases may also be implicitly weakened—a signal that graph algorithms can propagate automatically.

• Label propagation algorithms flag all cases whose authority depends on an overruled precedent
• Treatment edge typing distinguishes 'criticized by', 'limited by', and 'overruled by' relationships for precise risk scoring
• Real-time graph updates ensure that a new Supreme Court reversal immediately surfaces all affected downstream authorities

The graph structure encodes jurisdictional relationships as weighted edges, allowing AI systems to calculate the binding or persuasive force of any authority for a given legal question in a specific court.

• Vertical stare decisis edges connect lower courts to their controlling appellate authorities with maximum weight
• Horizontal edges between peer courts carry persuasive but non-binding weight
• Cross-jurisdictional edges enable persuasive authority analysis when a jurisdiction has not yet ruled on a novel issue

Time-series analysis of citation graph dynamics reveals the lifecycle of legal authority—identifying rising stars, established precedents, and declining cases before they are formally overruled.

• Citation velocity measures the rate at which a case accumulates new citations, flagging rapidly emerging authorities
• Citation decay curves model how a case's influence diminishes over time, surfacing aging precedents for re-evaluation
• Anomaly detection identifies sudden citation spikes that may indicate a case's relevance to a newly prominent legal issue