Binding Precedent

Binding precedent operates exclusively along a vertical axis within a court hierarchy. A decision from a higher court compels all lower courts within the same jurisdictional chain. Computationally, this is modeled as a directed edge in a citation graph with a mandatory attribute, constrained by a jurisdictional scope filter. For example, a ruling by the U.S. Supreme Court binds all federal circuit and district courts, while a decision from the Ninth Circuit binds only district courts within that circuit. This verticality is the primary differentiator from persuasive authority, which operates horizontally across jurisdictions.

The binding force of a precedent is strictly territorially bounded. A decision is only mandatory within the sovereign or geographic jurisdiction of the issuing court. In computational systems, this requires a jurisdictional filter—a graph traversal constraint that prunes nodes outside the relevant hierarchy. Key containment rules:

• Federal circuits bind only their constituent districts
• State supreme courts bind all lower state courts within that state
• Specialized courts (e.g., Tax Court) bind only within their subject-matter domain Failure to apply jurisdictional filtering produces legally invalid authority scores.

Not all text within a judicial opinion carries binding force. Only the ratio decidendi—the essential legal principle necessary to resolve the dispute—constitutes binding precedent. All other statements are obiter dicta, which may be persuasive but never mandatory. This distinction creates a critical NLP challenge: ratio extraction requires identifying the minimal logical chain linking material facts to the holding. Computational models must distinguish:

• Holding: The court's answer to the legal question presented
• Reasoning: The logical steps connecting facts to holding
• Dicta: Incidental commentary, hypotheticals, or background discussion

A precedent is only binding when the material facts of the instant case substantially match those of the prior decision. If a court finds a material distinction, it may distinguish the precedent and decline to apply it. This is modeled computationally as a similarity threshold between case fact embeddings. The distinguishing process involves:

• Fact extraction: Identifying legally significant facts from both cases
• Analogical mapping: Aligning fact patterns to detect structural congruence
• Materiality weighting: Assigning higher weight to facts that were dispositive in the prior holding A high similarity score triggers the binding constraint; a low score permits distinguishing.

The doctrine of stare decisis—'to stand by things decided'—is the jurisprudential foundation of binding precedent. In computational systems, it functions as a hard constraint on prediction models: when a binding precedent exists, the model must predict adherence unless an overruling or distinguishing event is detected. This is implemented through:

• Constraint satisfaction: Binding nodes in the citation graph restrict the output space of case outcome predictors
• Precedent chain traversal: Following the lineage of a legal principle through successive applications
• Treatment monitoring: Detecting negative treatment signals that weaken or nullify the constraint Without stare decisis modeling, legal prediction systems produce jurisprudentially incoherent outputs.

Binding precedent is not permanent. It can be overruled by a higher court or by the same court sitting en banc, which severs the mandatory edge in the citation graph. Computational systems must detect and propagate nullification events:

• Express overruling: A court explicitly states that a prior holding is no longer good law
• Implied overruling: A later decision contradicts the prior holding without explicitly naming it, requiring inference
• Abrogation: A statute or constitutional amendment supersedes the judicial precedent After nullification, the node's authority score drops to zero for binding purposes, though it may retain persuasive or historical value.

The computational representation of the doctrine requiring courts to follow precedent. This involves encoding hierarchical court structures and jurisdictional scopes into a rules engine that determines when a prior decision imposes a mandatory constraint on a current matter. The model must distinguish between vertical stare decisis (higher court binds lower court) and horizontal stare decisis (court follows its own prior rulings), assigning different constraint weights accordingly.

A graph traversal constraint that limits authority analysis to courts within a specific sovereign hierarchy. For a given legal question, the system must identify the controlling jurisdiction and traverse only the subgraph of courts whose decisions carry binding force. This requires a precise ontology of court relationships, including geographic boundaries, appellate paths, and subject-matter divisions.

A quantitative measure of a decision's binding or persuasive force, computed from multiple signals:

• Court hierarchy level: Supreme Court > Appellate > Trial
• Jurisdictional relevance: In-circuit vs. out-of-circuit
• Subsequent treatment: Followed, distinguished, or criticized
• Temporal factors: Age of decision and citation velocity This score feeds directly into authority propagation algorithms.

Graph algorithms, often variants of PageRank, that iteratively distribute precedential influence across a citation network. A case's authority score is a function of both the number and the authority of citing cases. Propagation must account for treatment sentiment—a negative citation from a high-authority court should decrease, not increase, the target's score. This enables identification of the most legally significant nodes in the graph.

The automated identification of citation instances where a higher court or later panel explicitly invalidates a prior decision's holding. This is the highest-priority signal in a citation network, as it severs the precedential chain. Detection requires fine-grained NLP to distinguish express overruling from mere criticism or factual distinguishing, often using treatment type classification models trained on annotated citator data.

A sequential path through the citation graph tracing the logical lineage of a legal principle from its seminal case through all subsequent applying, interpreting, and modifying decisions. Traversal algorithms must handle:

• Branching: When a principle diverges across circuits
• Termination: When an overruling event severs the chain
• Weakening: When negative treatment reduces authority without full invalidation