Adjudicated findings in United States federal courts that filed material contained fabricated or misrepresented authority produced with machine assistance. Represented professional parties only. Every confirmed entry resolves to a docket you can open.
| Case | Court | Decided | Party | Tool named | Penalty | Status | Docket |
|---|
It claims that each confirmed entry corresponds to a real federal docket. The docket number is printed and links to the public record. That is the whole point: a claim about verification that cannot itself be verified is the thing this project exists to object to.
It does not claim to be complete. Incidents enter the record after an order issues, so recent months are always undercounted. It does not report a rate, because there is no denominator — the number of filings that could have produced a finding is unknown.
34 entries sit on 17 dockets. These are not duplicate rows. Each pair carries two distinct decision dates, in several cases more than a year apart, and often different tools and different penalties — Coomer v. Lindell at $6,000 in July 2025 and $5,000 in May 2026; Bunce at $2,500 then $5,000. The unit of this register is an adjudicated finding, not a case. A court that sanctions twice has found twice. Repeated docket numbers are marked 2 findings.
Seven entries could not be matched in either public source. They are listed as candidates for paid retrieval rather than dropped, because an exclusion you can see is worth more than a percentage you cannot check.
Most entries are confirmed against a bulk docket archive by caption and court. A handful are confirmed through the live search index. A third route — primary document — takes the docket number off the face of an opinion held as a file and resolves it, and it exists because the first route failed. A fourth — agency alias — exists because of Rule 25(d): when a government official is automatically substituted as a party, the compilation names the department and the docket names the officeholder, and no amount of caption similarity bridges that. Those entries are found by searching the private party and recognizing the substitution.
The residual is read again rather than left to sit. Four entries were resolved on 17 September by reading the unresolved set one row at a time — among them the Sixth Circuit sanctions decision this register is most often asked about, which had been sitting under review while the paper that relies on it was being written.
James F. v. Commissioner of Social Security was confirmed by caption match at a similarity score of 0.992 to a docket opened in 2021. The opinion itself gives 1:26‑cv‑10137, filed in 2026. The caption was never a name: it is the anonymization template every Social Security claimant in that district receives, and a near‑perfect score against a template is evidence of nothing. The build now refuses to confirm any redaction‑convention caption by match, and refuses to publish any confirmation whose decision date precedes the filing date of the docket it was matched to. Both checks run on every build and both must pass.
Nine matters before federal agencies and boards — the GAO, the Armed Services Board of Contract Appeals, and the USPTO — are excluded by policy and appear nowhere in these counts. Self-represented litigants are also excluded: they carry no professional liability policy, and courts apply a different remedial practice to them, resolving such cases by warning rather than penalty. A population mixing the two understates the represented-professional distribution by construction.
Party values are carried from the source and normalized for filtering only; the original strings, including their spellings, are preserved in the published dataset.