Methodology

Sources of truth

Different sources are authoritative for different facts, so we resolve each field to the source that owns it:

• Federal Judicial Center (FJC) Biographical Directory: judicial service, dates, seats, and seat succession for every Article III judge since 1789. Public domain.
• CourtListener / Free Law Project: opinions, the citation graph, and financial disclosures (from the public-domain bulk data). Public domain.
• Wikidata: cross-identifiers and portraits. CC0.
• Congress.gov + U.S. Senate roll-call records: nominations and confirmation votes. Public domain.

How sources are reconciled

Official records govern. We enforce a tiered hierarchy, so a fact stated about a person never rests on a source that anyone can edit:

• Tier 1: official records. FJC Biographical Directory; Congress.gov and U.S. Senate confirmation records; court .gov sites. Every appointment, confirmation, seat, succession, date, and career fact rests here.
• Tier 2: Wikidata and Wikipedia, enrichment only. Cross-identifiers, disambiguation, and portrait discovery. Never the sole basis for a fact stated about a person. If a datum exists only in Wikidata, we confirm it against a Tier 1 record or leave it out.
• Tier 3: Wikimedia Commons. Portrait files plus their license and attribution, nothing more.

Connecting the same judge across sources

A judge appears in several datasets under different identifiers. We link them deterministically first, via the ID bridges the sources publish (the FJC nid, CourtListener’s fjc_id, a Wikidata QID), then by an exact normalized key. Anything still ambiguous is proposed, never auto-merged: it goes to a human review queue rather than silently guessing.

Measured vs. computed

We separate what a record states (a confirmation vote, an appointment date) from what we compute from records (years of service, rankings). Computed figures show their formula. We don’t publish outcome predictions.

The Judicial Record (court opinions)

On judge pages for the federal courts and for the state supreme and appellate courts we show a Judicial Record: the opinions a judge authored, their most-cited ruling, and a selection of their rulings. It is built entirely from the Free Law Project’s public-domain CourtListener bulk data. Our primary attribution is CourtListener’s own structured author id: the same identifier we key our federal judges to (through their FJC record) and our state judges to (directly, since we import the state bench from CourtListener). We do not host opinion text or write our own summaries. Every ruling links to the official public-domain full text. Each figure is objective, recomputable, and scoped to what is in our data.

CourtListener leaves many older state supreme-court opinions with only a free-text author name and no structured id. For those we add a second, deliberately conservative attribution: we credit the opinion to a judge only when the court (from CourtListener’s own court record), the filing year, and the author’s surname resolve to exactly one justice on our roster of that court in that year. If two justices ever shared a surname in overlapping years, or the surname isn’t on our roster for that court and era, we leave the opinion unattributed, and never guess. These name-matched opinions are marked † on the judge page and, like every ruling, link to the CourtListener official text so you can verify the author yourself.

• Opinions authored: the count of published opinions for the court attributed to this judge, whether by CourtListener’s author id or by the unique court+date+surname match above (the latter marked †). Concurrences and dissents are counted separately.
• Most-cited / “Cited”: inbound citations to an opinion across the whole CourtListener corpus. This is an objective magnitude, not our judgment of importance.
• Selected / recent rulings: ranked by that same citation count; no editorial hand-picking.
• Which court an opinion belongs to: resolved from the seat the judge held on the day the opinion issued, so an elevated judge’s district years and circuit years are told apart.

Coverage is partial, and we label it as such. CourtListener does not tag an author on every opinion, so these counts are a floor, not a judge’s complete output. The record shows what is attributable in our data, never “every opinion this judge wrote.” For district judges the record is especially selective: most unpublished orders, where much trial-court work lives, are not in the published corpus, so a district record is a sample of notable rulings, not a docket. Our state coverage is concentrated in the supreme and intermediate appellate courts, whose opinions CourtListener attributes well; state trial courts rarely publish authored opinions, so most trial judges carry no record.

Caseload, case types & time to disposition

On district-judge pages we summarize the docket a judge actually handled, from the Free Law Project’s public-domain CourtListener dockets (PACER-derived). Each case is attributed to a judge through CourtListener’s structured assigned-to id, bridged to the judge’s FJC record, the only judge signal available on free data, because the Federal Judicial Center’s Integrated Database redacts every judge field by Judicial Conference policy.

• Cases assigned: federal district-court dockets on which this judge is the assigned judge in our data. It is a partial sample, not a complete caseload: CourtListener does not hold every docket, and the field reflects the judge last assigned a case, so a reassigned or inherited case counts for its final judge. A few mass-tort judges show very large counts because a consolidated multidistrict litigation gathers thousands of member cases under one judge, and we label those.
• Case-type mix: the share of a judge’s docket in each Administrative Office nature-of-suit group (civil rights, prisoner & habeas, contract, torts, intellectual property, and so on), plus a criminal bucket. Percentages are of the cases we can classify.
• Median time to disposition: the median days from filing to termination across the judge’s closed cases in our data. A median (not an average), so a few years-long cases don’t distort it; a descriptive figure, never a rating of speed.

Affirmed and reversed on appeal

For district judges we show counts of how their decisions fared on appeal: “of N reviewed decisions in our data, A affirmed, R reversed or vacated.” It is a two-source join of public records:

• The FJC Integrated Database (appellate) supplies the coded merits outcome (affirmed / reversed or vacated / affirmed in part) and the originating district docket, but its own judge field is redacted.
• CourtListener dockets supply the district judge, matched to the appellate record by court and docket number.

We show counts, never a rate or percentage, and only for judges with at least ten matched merits reviews. Real limits, stated plainly: it covers only appeals decided on the merits from FY 2008 forward that we could match to a district docket with an identified judge; a decision appealed more than once is counted each time; and the “assigned” judge is the last assignee. It is a historical record of outcomes, not a prediction and not a verdict on the judge; an affirmance or reversal turns on many things beyond the trial judge, and we never estimate a number we cannot source.

What we will not do. We describe the record, never the jurist. No “good/bad,” “slow,” or “often wrong.” And we do not build motion-grant analytics or outcome prediction (“how will this judge rule”): those require deep docket parsing, are outside our free-data scope, and are the subject of an active patent family. Out of scope pending counsel review.

Accuracy & corrections

A validation gate runs on every build and blocks deploy if invariants break (an appointment with no judge, impossible overlapping tenures, a dangling link). Pages carry a visible “data last verified” date. Found something wrong? Tell us, and cite the primary source, and we correct against the record.

Informational only; not legal advice. Verify against the primary source before relying on anything here. Not a consumer report (FCRA): not for credit, employment, insurance, or tenant-screening use.