A new dataset called ImmigrationReason has been created by researchers. It contains information on 12,375 non-precedent decisions made by the U.S. Citizenship and Immigration Services (USCIS) between 2005 and 2026. The dataset includes details such as applicable legal frameworks, evidence sufficiency findings, adjudicator criticism quotes, citations, and final dispositions. This dataset can be used for research on legal reasoning, including outcome prediction and analysis of adjudicator errors. --- Why it matters: This dataset matters to researchers in AI because it provides a large-scale structured dataset for studying administrative adjudication, which is an underaddressed area in legal NLP resources. It also enables analysis of nearly 9,000 instances of AAO-identified legal errors, making it a valuable resource for improving outcome prediction and agent design in high-stakes regulatory domains. Source: https://arxiv.org/abs/2608.20391