About DocketDrift
DocketDrift is a navigator for the public record — specifically, state appellate court opinions. We collate published opinions from official sources, normalize them into a structured archive, and make patterns in judicial decisions easier to find and read together.
Does DocketDrift hallucinate cases?
No. Generative AI legal tools — Lexis+ AI, Westlaw AI-Assisted Research, CoCounsel, Harvey, and others built on large language models — have been documented to fabricate case citations at rates of 17–33% even with retrieval-augmentation, per a 2024 Stanford HAI study (“Hallucination-Free? Assessing the Reliability of Leading AI Legal Research Tools”). The canonical public failure was Mata v. Avianca (S.D.N.Y. 2023), where an attorney was sanctioned after ChatGPT invented six fake case citations for his brief, complete with fake parties, fake judges, and fake quotations.
That entire failure mode is structurally impossible on DocketDrift: we do not generate text. See How DocketDrift differs from AI legal tools →
Who DocketDrift is for
DocketDrift is built for appellate attorneys, journalists, academic researchers, and self-represented litigants. State appellate research has historically been dominated by paid databases. DocketDrift treats the public record as what it is: public.
Methodology
All source material in DocketDrift comes from public records. Opinions are ingested from official court releases and the open Free Law Project / CourtListener archive. Where structured fields (case number, disposition, author) can be extracted reliably from the published text, they're populated automatically. Where they can't, they're left empty for human review.
Patterns and case-pairs that the system surfaces are tools for human review, never assertions of judicial inconsistency. Two cases citing the same statute with opposite outcomes are usually legitimately different — different facts, different procedural posture, different sub-issues. The "did this judge contradict themselves" question is a question for a human reader; we just make the pairs easier to find.
Data sources
- CourtListener (Free Law Project) — historical backfill and the standing source for ongoing ingestion. CourtListener typically lags real-time by a few weeks; for Minnesota that backfill reaches the founding-era 1851 opinions of the territorial supreme court forward.
- Direct upload — new opinions are uploaded same-day on Minnesota's Monday/Wednesday release schedules, ahead of CourtListener's ingestion. State subdomains refresh weekly via the scheduled ingest job.
Editorial review
Every opinion in the corpus is currently being reviewed and tagged by a human editor. Until that pass is complete, individual records carry a status indicator showing whether the opinion is processed (machine-parsed, awaiting review), flagged (queued for re-review), or human-reviewed. The intent is that machine extraction does the bulk-work part (case number, disposition, panel composition, statute citations) while the editorial pass curates the doctrinal tags, catches parser misses, and flags edge cases. Both signals stay public; you can see at a glance which records have been read by a human.
Status
Beta · Minnesota Beta · New Hampshire Beta · Arizona
- Minnesota — full appellate corpus loaded: 60,000+ opinions spanning 1851 to current, refreshed weekly as new opinions are released. Judge dossiers, outcome categorization, statute citation graph, semantic search, and tag-suggestion review pipeline are all live. Editorial review of the corpus is in progress.
- New Hampshire — full NH Supreme Court corpus loaded: 20,000+ opinions. Byline-extracted judicial panel graph populated. Semantic search and tag-suggestion pipelines running. Editorial review in progress.
- Arizona — full Arizona Supreme Court + Court of Appeals corpus loaded: 38,000+ opinions. Byline extraction live for the Court of Appeals; Supreme Court byline format pending a follow-up parser pass. Editorial review in progress.
Don't see your state? Request it →
Open source
DocketDrift is open source on GitHub at OnionMadder/docketdrift. The parser rules, ingestion logic, and infrastructure are all readable, auditable, and forkable.
Contact
Notice an error in an opinion record, a judge bio, or a search result? Send a note to hello@docketdrift.com.
Privacy
We don't log search queries, track users, or save research history. Full privacy statement →