FedCourtsAI

← All results

Release 2

Published November 16, 2026 · tag results/ot2026-interim-exec · commit cc2af47

Rebuilt from the Supreme Court’s own dockets, a published count of the government’s emergency-application wins matches exactly for Biden and comes within one for each Trump term.

What this is. On 2026-10-19 Adam Feldman’s SCOTUSblog column, “Where the Supreme Court favors Trump on the emergency docket – and where it doesn’t”, counted how often the federal executive prevailed on emergency applications about federal executive actions, from 2017-01-20 to 2026-10-12. The column gives totals, not a list of cases. This release rebuilds those totals from the Court’s own dockets, using rules and hand codes committed to a public repository, so anyone with the code can see exactly which applications were counted and how. It is a first step: the same record is the base for later analysis of the column’s question.

Unlike Release 1, this release scores no AI forecast. It checks a published analysis of what the Court has already done.

Written by AI. This page, like the audit write-up behind it, was written by an AI agent (Claude, from Anthropic) under the project maintainer’s direction, and the agent copied each figure from the audit write-up. The rules that decide which applications count were also written by AI agents, and AI raters coded the applications and labeled the policy areas. AI reviewer agents checked the figures, and the maintainer reviewed the page before it was published.

How it compares

When the government asked the Court for emergency relief, how often did it get it? Each pair is wins over applications the Court decided; withdrawn, moot and undecided applications are left out, as the column leaves them out.

The column This release
Trump, first term 25 of 33 26 of 34
Biden 10 of 15 10 of 15
Trump, second term 27 of 33 26 of 32

How applications were placed. Two AI raters coded 298 applications by hand for whether the case challenges or defends a federal action, and every application counted above rests on their codes; a rule that reads each docket’s caption sets aside the rest. The AI agent that wrote this release settled 55 of the codes, where the raters split or where it read a case differently from both, and set requests to file over the page limit aside. It did this with the column’s figures in view, so part of the match is by design. A caption rule alone, with no hand codes, gives 26 of 34, 12 of 20 and 27 of 33: it matches the column exactly for Trump’s second term, and the hand codes move that pair one off. Two Biden applications stay unplaced.

The totals of applications differ more than the rates do. This release counts 182 applications in all against the column’s 165: 49 under Trump’s first term (39 by the government) against the column’s 39 (34), 83 under Biden (16 by the government) against 81 (18), and 50 under Trump’s second term (41 by the government) against 45 (42).

By policy area, the comparison is mixed, and most areas hold only a handful of applications. Immigration in Trump’s first term matches the column, 11 of 12 decided, and in his second the government won 10 of 10 against the column’s 11 of 12. Under Biden the release gives 2 of 2 firearms applications and 3 of 3 health applications, against the column’s 2 of 2 and 4 of 4. They part on Biden’s education applications, which the column has the government losing all 3 decided and this release has it winning 2 of 5, and on Biden’s immigration applications, 1 of 3 against 2 of 4. So Biden’s total matches while some of its areas do not. The areas are hand labels, and the raters marked some of them uncertain.

Counting cases rather than applications. One policy can produce several applications, so the column also groups them into “controversies”. The column does not say how it groups them, and this release groups them more finely, so these figures differ more: for Trump’s second term it counts 30 of 37 against the column’s 23 of 30.

What this does and doesn’t show

  • It matches totals, not cases. Because the column lists no cases, a matching total could still hide different cases on each side, and the policy areas suggest some do.
  • Biden is where the hand codes matter. They set aside or re-read six of the government applications the caption rule counts under Biden, among them a bankruptcy trustee’s objection and a suit over a state law, and that closes the whole gap between 12 of 20 and 10 of 15. Four of the six were settled with the column in view; without those four the pair would be 11 of 18.
  • It is not a measure of favoritism. Which disputes reach the Court depends on who chooses to apply and on what the lower courts did. A count of wins does not adjust for that; a later release will.
  • The rules, codes and labels are AI-built. Their errors cannot be separated from disagreements with the column.

Go deeper

This page is the public summary committed at the tag above, shown without changes. Every figure in it is copied from the audit write-up or the release’s dataset export, and if this page and either one disagree, the source is right and this page is wrong.