Methodology

Every number we publish should be traceable, and this page explains how the sausage gets made.

Where the data comes from

Court records. The litigation database is built from public federal docket records via CourtListener and the RECAP archive, supplemented by documents we purchase directly from PACER. Case pages link to the underlying dockets.

Baseline statistics. Historical filing baselines come from the Federal Judicial Center's Integrated Database, using plaintiff-assigned nature-of-suit codes. Those codes carry the same caveats the FJC itself publishes: they reflect what the filing party selected, not a court's characterization.

Supreme Court emergency docket. Emergency application counts are compiled from the Court's public orders and docket entries, cross-checked against academic tallies of prior administrations.

What counts as a "case against the administration"

A federal case in which the United States, an agency, or an official sued in an official capacity is a defendant, and the claims challenge an action of the current administration. Long-running matters that predate January 20, 2025 are tracked when current administration action is at issue; they are counted but excluded from "filed under this administration" date ranges, and we say so where it matters.

How outcomes are coded

Outcome events (TROs, preliminary injunctions, dismissals, appellate results) are coded from docket entries, then verified by consensus: a second, independent model re-derives each case's outcome facts from the same docket record, claims must be corroborated by the docket text itself, and win/loss calls are cross-checked against independent codings (including Just Security's litigation tracker). Cases where the sources agree are marked verified; disagreements are resolved by a human or excluded. We publish coverage alongside every outcome figure, for example "208 of 1,334 cases coded," rather than implying the numbers describe the full universe.

What we deliberately do not publish yet

TRO and PI grant rates. Our coding currently captures grants but not denials. A rate computed from grants alone would always read 100%, which is a construction error, not a statistic. The column ships when denial tracking does.
Any figure whose source rows we cannot produce on request.

DOJ attorney workforce data

What OPM FOIA is. The Office of Personnel Management periodically releases its Enterprise Human Resources Integration (EHRI) dataset under FOIA. The Q1 2017 release covers individual federal employees from 1973 through early 2017. Each record contains an agency code, pay plan, series, and hire date. We filter to agency code "DJ" (Department of Justice), series "0905" (General Attorney), to isolate DOJ attorneys. The dataset is not continuously updated; it is a snapshot through the release date.

File and line traceability. As a concrete example: a separation event in fiscal year 2003 (1,270 recorded) traces to individual EHRI employee records with separation_date falling in that year, each identified by an anonymous employee ID, pay plan GS, series 0905, and agency DJ. The OPM release filename and record structure are documented in the DOJ attorney roster spec (auditor/SPEC_DOJ_ATTORNEY_ROSTER_20260718.md) for any future verification.

The 11/12 FedScope cross-check. OPM's FedScope public tool publishes monthly employment snapshots by agency and occupation. We cross-check our attorney counts against FedScope's GS-0905 DOJ figures. The two series track closely but not identically because FedScope includes non-permanent and excepted-service appointments that may not appear in the EHRI dataset. Where they diverge materially we note it.

What "no longer appearing in tracked litigation" means. For attorneys who entered DOJ service after the Q1 2017 FOIA cutoff, no hire-date record exists in the EHRI data. Their presence in the people table comes from appearing as counsel on a docket entry in a case we track, starting January 20, 2025. "First appearance" in that context is a lower bound, not a hire date; the true hire date could be years earlier.

Caveat on headcount figures. The current-headcount series uses OPM FedScope monthly employment data (attorneys, DOJ, employment status). The September snapshot each year is used as the annual figure because OPM historically publishes September as its official end-of-fiscal-year count. The 2026 figure uses the most recent available month. Seasonal variation between months is roughly 1–3% and does not affect year-over-year comparisons materially.

Complete vs. partial years in the flows chart. The annual hires and departures chart marks 1982 and 2014 as partial years (shown with lighter bars and an asterisk). The 1982 figure covers only part of the fiscal year in the EHRI data as delivered; the 2014 figure is truncated by the Q1 2017 release covering through early 2017 but with incomplete reporting for the most recent year at the time of extraction. All intervening years (1983–2013) are complete annual counts.

Corrections

If a number is wrong, we say so in the next Scorecard and correct the archive copy. Paid subscribers can question any specific figure through the monthly subscriber form; selected questions and answers run in the Scorecard's Ask the Docket section.

Funding

This work is funded by readers through Substack subscriptions and one-time support. Subscription revenue pays for PACER fees, document archiving, and database infrastructure. No political organization, party, or campaign funds any part of it.