Public data, instantly queryable

O*NET 31.0: occupations, skills and work context, with scale references and precision flags (468,857 rows; CC BY 4.0). QCEW 2025 annual: covered employment and wages, with all disclosure flags and a suppression-aware view (3.1M rows; BLS public domain). FARS 2024: the first Annual Report File, with 36,297 fatal crashes and views that aggregate people and vehicles before joining.

Each page includes Python, R and SQL examples, links to table and field dictionaries, researcher caveats, source provenance and a changelog. The pinned public artifacts have been checked against their sources and queried from Python and R.

All three are now in the 19-database catalog. Find them in the database cards above or use normal Python and R package discovery. Their immutable direct URLs remain available on the dataset pages.

open-policing: every standardized stop released by the Stanford Open Policing Project, 35 state patrols and 53 city or county departments in one table (230.0M stops), with the project's standardized columns typed and each location's raw source fields kept, plus a locations table with observed and stated coverage. courtlistener: CourtListener's judge database (16,190 judges, 51,289 positions), its registry of 3,353 courts, and 32,336 federal judicial financial disclosure reports with their gifts, debts, reimbursements and income line items, joinable to the FJC directory. Also this week: an independent audit of every repository and database, answered in four rounds (see the changelog). Full changelog →

fjc: the Federal Judicial Center's Integrated Database, every federal civil case since 1988, criminal defendant since FY1970 and appeal since 1971 (21.1M rows), with the two file eras unioned and invalid source dates audited rather than silently dropped. cbp: the Deportation Data Project's CBP FOIA releases, Border Patrol apprehensions, encounters, port-of-entry inadmissibles, Title 42 expulsions and CBP One (51.3M rows), 121 differently-shaped spreadsheets harmonised into one schema per record type with every source file documented. ussc: the U.S. Sentencing Commission's individual offender datafiles, every federal defendant sentenced FY2002-FY2025 (1.72M defendants, 14.0M rows), reshaped from a 22,000-variable fixed-width record into a sentences table plus long tables per count, guideline computation, drug type and departure reason. cook-sao: the Cook County (Chicago) State's Attorney's felony case datasets, 2011 until the Office closed the series at the end of 2024 (3.2M rows): felony review, charges and bond at initiation, dispositions with judge and court, sentences, diversion referrals. scdb: the Supreme Court Database, every decision 1791-present at the case and justice level with its 2,949-entry code book loaded as a table. fjc-judges: the FJC's biographical directory of every Article III judge since 1789, with each appointment's nomination-to-commission timeline. Each ships with a data dictionary, researcher caveats, and works from Python and R today. Full changelog →

Six databases rebuilt from their latest sources: immigration courts through the August 2026 DOJ file (173M rows), ICE enforcement through August 6 from the Deportation Data Project's new release (23.6M rows, with a new arrest city field), ClinicalTrials.gov as of September 14 (59M rows), DOL H-1B/PERM through FY2026 Q3 (with a fix for date columns that had been empty), and FEC campaign finance through the current 2026 cycle (355M rows), plus IPEDS through the 2025-26 provisional release (27.9M rows). The July audit's repairs now run as regression checks inside every build. New: an R package with the same API as the Python client, and datapond 0.1.3 on PyPI. Full changelog →

Every existing database was rebuilt from its latest source (ICE gains a new encounters table and the 2026 FOIA release; CMS Medicare adds CY2024; FEC covers the 2026 cycle), then independently audited from a researcher's perspective and repaired — including fixes to views, lookups, and column typing that affected published-number accuracy. Two databases were added the next day: CMS Open Payments (172.5M rows) and DOL H-1B/PERM visa disclosures (8.5M rows). Each README now carries a researcher-caveats section, and every HuggingFace dataset page includes the changelog. Full changelog →