The K Rule: Authority Selection

Deterministic curation of high-impact precedents using global network centrality scores.

The Selection Problem

A typical Supreme Court opinion contains dozens of citations. Including all of them would create unwieldy context packets, while including too few might omit critical precedents. The K Rule provides a principled method to select the most authoritative citations for inclusion in ResearchPacks.

SCOTUS

Fowler Score

Derived from network analysis of the U.S. Supreme Court citation graph.

Landmark 0.9-1.0

Strong 0.5-0.9

CAP

PageRank Percentile

Centrality measured across the broader Federal reporter citation network.

High Centrality Top 10%

Supplementary Top 25%

The K Values

We analyzed authority score distributions across 20,402 anchors to determine the "sweet spot" where authority quality remains high without bloating the context budget.

Source	K Value	Mean Authority Quality	Rationale
SCOTUS	10	0.76 (Fowler)	Landmark-heavy selection; maintains high precedential density.
CAP	5	0.72 (PageRank)	Focused set of supplemental circuit/district authorities.

How Selection Works

The process is fully deterministic and occurs during the pipeline build-time (Stages 3.7 to 4A).

# Stage 3.7 Process 1. Rank all citations by Fowler / PageRank 2. Generate ranked.jsonl maps per anchor # Stage 4A Build 3. Apply K Rule (Top 10 / Top 5) 4. Record audit trail in doc3.json: { "scotus_k_config": 10, "selected": 8, "shipped": 8 }

Tie-Breaking Policy

In cases of identical authority scores, ties are broken lexicographically by the citation string (e.g., "410 U.S. 113" < "411 U.S. 1"). This ensures 100% deterministic selection across runs.