Repository navigation
Releases
Executive-action interim docket replication
Replication of this SCOTUSblog post by Adam Feldman about the interim docket and executive action.
OT2026 long-conference cert results
OT2026 long-conference cert results
Three AI models forecast which petitions the Supreme Court would take up at its 2026-09-28 conference. On the 97 scored petitions the Court has decided (91 of them pre-registered), it granted 3: one full grant and two sent back for reconsideration (GVRs). That is too few grants to measure a difference between the models.
Written by AI. The release write-up and its public summary were written by AI agents (Claude, from Anthropic) under the project maintainer's direction. Every figure was produced by the command its section names, at a stated commit, and checked by AI reviewer agents. The maintainer reviewed both before publication. The forecasts and gradings they report are themselves the output of AI models.
What this release is
- Pre-registered. The process (prompts, models, scoring rules) was fixed under the
prereg/proc-v8tag before any outcome existed. Every forecast for the pre-registered set was merged intomainbefore the Court met; the last at 2026-09-19 19:45:51 UTC. The proof of timing is GitHub's recorded merge time, not a commit's own date. - Scored. Each forecast was graded by three AI judges against the Court's 2026-10-01 grant and the 2026-10-05 order list. No forecast in the cohort was flagged for having seen its own outcome.
- The headline measure is skill against each salience band's historical grant rate, read beside "always deny" on the same petitions. Most petitions are denied, so beating "always deny" is the real test.
- This was not every petition at the conference. The set was chosen by a pre-registered ranking that leans toward petitions more likely to be granted. It says nothing about the conference as a whole.
Read more
- Public summary:
docs/release-ot2026-public-summary.md - Full audit write-up, with every denominator, sensitivity line and disclosure:
docs/release-ot2026-long-conference.md - The data behind it (predictions, gradings, reasoning, data dictionary; CC BY 4.0): 10.5281/zenodo.23263986
- The code that produced it: this tag,
results/ot2026-longconf(commit07bf288c1), archived in 10.5281/zenodo.22966596 - The ledger: fedcourts.ai/ledger
Still pending
Petitions the Court relisted or held, and all 10 petitions where it asked for the Solicitor General's views, are not scored yet. Relisted petitions are granted more often than others, so the scores so far lean toward denials.
Pre-registration: proc-v8
Pre-registered process for the OT 2026 long conference