Our Methodology

Data Sources

CourtListener & the Free Law Project

GavelSight's foundation is CourtListener, the Free Law Project's open database of federal court opinions. CourtListener aggregates published opinions from all federal district and circuit courts, making them searchable and machine-readable. This open-data approach means our analysis starts from the same public record available to any attorney — we add structure, patterns, and predictions on top.

CourtListener's strengths: comprehensive coverage of published opinions, open API, active maintenance by the Free Law Project. Its limitations: not every unpublished order appears, coverage can lag days to weeks behind PACER, and docket-level data is available through the RECAP project but not always complete.

PACER Integration

Docket filings from PACER are available through CourtListener's PACER bridge and the RECAP browser extension. Full PACER integration is on our roadmap. Until then, docket-level data supplements our opinion-based analysis where available.

Baseline Data

At launch, judge profiles are built on curated baseline data — structured from public sources, expert-reviewed for accuracy, and clearly labeled in the product as baseline rather than live-synced. As CourtListener integration deepens (pending a commercial licensing agreement with the Free Law Project), profiles will transition to live-updated analytics.

What Counts as a Ruling

We analyze judicial actions that produce a written record in the federal court system:

• Published opinions in CourtListener's database — the core of our analysis
• Orders on motions (when captured in docket data) — motion-to-dismiss rulings, summary judgment decisions, preliminary injunctions
• Sentencing outcomes in criminal cases — deviation from guidelines, patterns across case types

What we exclude: unpublished minute orders, sealed proceedings, settlements without judicial action, and administrative docket entries. These exclusions are documented in every judge profile so attorneys know exactly what the data represents.

Ruling Classification

Case Type Taxonomy

Every ruling is classified into a case type category: Criminal Sentencing, Civil Rights, Government Regulation, Patent/IP, Employment, Contract, Tort, Immigration, Securities, Antitrust, and others. The taxonomy aligns with the federal courts' own Nature of Suit codes where possible, with additional granularity for categories where judicial tendencies diverge significantly.

Directional Tendency Analysis

Rather than reducing a judge to a single "liberal" or "conservative" label, GavelSight maps tendencies per case type. A judge might be plaintiff-favorable in employment discrimination cases but defendant-favorable in contract disputes. These directional tendencies are based on outcome patterns across multiple rulings, not individual decisions.

Partial Grants

Partial grants are treated as distinct outcomes, not forced into a binary grant/deny framework. When a judge grants a motion to dismiss on three counts but denies it on two, we record the partial grant — this nuance matters for litigation strategy.

Prediction Methodology

Our prediction model considers:

• Historical ruling patterns — the judge's own track record on similar case types and motion types
• Case type baseline — how the broader population of federal judges rules on similar matters
• Circuit norms — appellate patterns that may influence district judge behavior
• Temporal trends — whether a judge's tendencies have shifted over time

Predictions built on baseline data are labeled differently from those built on live CourtListener integration. A baseline prediction carries a "Baseline" badge in the UI and displays lower confidence intervals until validated against live data.

The full methodology payload — including the raw data points, weights, and caveats behind every prediction — is visible to subscribers. You can inspect what drove the prediction, not just read the output.

Sample Size and Statistical Thresholds

Small samples produce unreliable patterns. GavelSight enforces minimum thresholds before surfacing any tendency or prediction:

• Tendencies are suppressed below the minimum sample threshold for that case type
• Confidence intervals widen with smaller samples — a prediction based on 8 rulings shows a wider range than one based on 80
• "Insufficient data" is a valid and common output — we never force confidence where the data doesn't support it
• The exact sample size is displayed alongside every tendency and prediction

When GavelSight tells you "we don't have enough data for this judge on patent cases," that is the product working correctly. An honest gap is more useful than a fabricated pattern.

Coverage Today

Federal Courts

GavelSight covers all U.S. federal district courts and circuit courts of appeals. At launch, 20 judges have deep profiles with structured tendency data. Coverage is expanding — the goal is comprehensive federal coverage across all active Article III judges.

State Courts

Florida state trial courts are live. Texas and California are planned for late 2026. State court data comes from different sources than federal, with different coverage characteristics — each state's data quality and completeness is documented in the product.

Known Gaps

We do not currently cover: magistrate judges (who handle discovery disputes and some pretrial matters), bankruptcy courts, immigration courts, or military courts. These are on the roadmap but not yet in production. When a judge isn't in our database, we tell you — we don't guess.

Update Cadence

Baseline data is updated with each product release. When live CourtListener integration is active, profiles will sync on a regular schedule — the frequency will depend on the terms of our commercial licensing agreement with the Free Law Project.

Each judge profile displays a "Last analyzed" date so you always know how current the data is. If a profile hasn't been updated recently, that date tells you.

Browse the judge directory · Read our guide to researching judge history · View pricing