Shepardizing

Citator services assign treatment signals to each citation instance, categorizing how a subsequent court has treated the cited authority. These signals form the backbone of Shepardizing:

• Red/Stop: Negative treatment—overruled, abrogated, or superseded
• Yellow/Caution: Limited, criticized, or distinguished
• Green/Go: Positive treatment—followed, affirmed, or cited with approval
• Blue: Neutral citing references with no precedential impact

Each signal is algorithmically determined through treatment type classification NLP models trained on judicial language patterns.

Shepardizing constructs a directed 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. This traversal reveals:

• Direct history: Decisions in the same litigation (affirmed, reversed, remanded)
• Indirect history: Citations by courts in unrelated cases
• Depth analysis: How many layers of subsequent treatment exist

The resulting precedent chain allows attorneys to assess whether intervening decisions have eroded the original holding's authority.

The most critical function of Shepardizing is the automated identification of overruling events—instances where a higher court or later panel explicitly invalidates a prior decision's legal holding. Detection relies on:

• Explicit language patterns: Phrases like 'we overrule,' 'no longer good law,' or 'cannot stand'
• Implicit overruling: Situations where a holding is functionally invalidated without express language
• Jurisdictional scoping: Ensuring the overruling court has hierarchical authority over the cited case

Overruling detection directly feeds negative treatment signals and triggers alerts in citation monitoring systems.

Distinguishing occurs when a court declines to apply a precedent by finding material factual or legal differences. In computational Shepardizing, this is modeled as an edge attribute in the citation network:

• Factual distinction: Different parties, circumstances, or evidence
• Legal distinction: Different statutory framework or procedural posture
• Temporal distinction: Changed circumstances since the original decision

Distinguishing does not invalidate precedent but limits its applicability scope. Citator systems flag these instances with caution signals, indicating the authority remains valid but may not control in specific contexts.

Shepardizing applies jurisdictional constraints to ensure that treatment analysis reflects only legally relevant precedent. The filtering logic considers:

• Vertical hierarchy: Higher courts bind lower courts within the same jurisdiction
• Horizontal relationships: Courts of equal rank may follow but are not bound
• Sovereign boundaries: Federal vs. state, and cross-state authority
• Circuit splits: Conflicting interpretations across federal circuits

This filtering prevents false positives where a case from an unrelated jurisdiction appears to weaken authority that remains binding in the relevant court system.

Beyond binary treatment signals, modern Shepardizing incorporates citation sentiment—the polarity of a citing reference toward the cited authority. Sentiment analysis assigns:

• Strongly supportive: 'We reaffirm,' 'remains controlling'
• Weakly supportive: 'See also,' 'cf.'
• Neutral: String citations without commentary
• Weakly negative: 'But see,' 'questioned in dicta'
• Strongly negative: 'Expressly overruled,' 'no longer viable'

These sentiment weights feed into authority propagation algorithms, producing more nuanced precedential influence scores than simple citation counting.

A directed network structure where nodes represent legal cases or statutes and edges represent citation relationships. This forms the foundational data structure for computational precedent analysis.

• Nodes: Cases, statutes, regulations, or constitutional provisions
• Edges: Citation links weighted by treatment type and sentiment
• Directionality: Flows from citing authority to cited source
• Scale: Modern graphs contain millions of nodes and tens of millions of edges

An NLP task that automatically categorizes how a citing case legally treats a cited authority. This classification is essential for determining whether a precedent remains good law.

• Overruled: Explicitly invalidated by a higher court
• Distinguished: Found materially different from the current case
• Followed: Applied as controlling authority
• Criticized: Questioned but not explicitly overruled
• Harmonized: Reconciled with apparently conflicting authority

A graph algorithm that iteratively distributes precedential influence scores across a citation network. Often implemented using PageRank variants adapted for legal hierarchies.

• Weighted edges: Treatment sentiment modifies propagation strength
• Jurisdictional constraints: Authority flows only within binding hierarchies
• Temporal decay: Older citations may receive reduced weight
• Output: Ranked list of most legally significant nodes in the graph

The automated identification of citation instances where a higher court or later panel explicitly invalidates the legal holding of a prior decision. This is a critical signal for maintaining accurate authority graphs.

• Direct overruling: Express statement that prior case is no longer good law
• Implied overruling: Inconsistent holding that functionally supersedes precedent
• Partial overruling: Specific holdings invalidated while others survive
• Detection methods: Combine citator signals with NLP classification of judicial language

The algorithmic identification of landmark decisions that serve as origin points for major legal doctrines. These cases exhibit distinctive structural signatures in the citation graph.

• High out-degree centrality: Cited by many subsequent cases
• Sustained citation velocity: Continued relevance over decades
• Bridge between clusters: Connects previously separate doctrinal areas
• Applications: Legal research prioritization, curriculum design, and doctrinal mapping

A graph traversal constraint that limits citation analysis to courts within a specific sovereign or geographic hierarchy. This ensures authority scores reflect only legally relevant precedent.

• Vertical filtering: Supreme Court → Circuit Courts → District Courts
• Horizontal filtering: Excludes persuasive authority from other jurisdictions
• Temporal scope: Limits analysis to decisions within relevant time windows
• Subject matter: Filters by legal domain when jurisdictions overlap