Abstract
<p>Federal courts sentence tens of thousands of people every year, yet the public sentencing records released by the United States Sentencing Commission (USSC) deliberately omit the identity of the sentencing judge, limiting research on and public oversight of judicial discretion. We present JUSTFAIR 2.0, a public dataset that links 140,299 federal criminal sentencing records from fiscal years 2018 through 2023 to 969 individual district judges across 87 judicial districts, using only publicly available sources. We construct the dataset in three linkage stages: (1) linkage of USSC sentencing records to case records in the Federal Judicial Center Integrated Database (FJCID) using a graph-based method with offense-code similarity scoring; (2) extraction of judge identifiers embedded in docket numbers scraped from the Public Access to Court Electronic Records (PACER) system; and (3) matching of these identifiers to the Biographical Directory of Article III Federal Judges, which supplies each judge's full name, demographic information, appointment history, and appointing president. We validate the record linkage internally by comparing offense-code agreement between unique and non-unique matches (Jensen-Shannon divergence = 0.006) and validate judge matching by manually auditing a stratified sample of matches against PACER dockets, finding no errors. We document attrition at every pipeline stage and assess representativeness by comparing retained and excluded records on demographic, offense, and sentencing characteristics. The dataset updates and extends the original JUSTFAIR database, the first large-scale public resource connecting federal sentences to judges. We describe known structural limitations, including districts whose docket numbers lack judge identifiers, and offer guidance for appropriate use.</p>