A judge can be bribed. A judge reads the headlines. A judge can be played, and a judge gets tired. A machine suffers none of that — but does that make it fit to judge? Judgir is an open research project testing exactly that question, on real concluded cases, in public, one verifiable verdict at a time.
Follow the research Read the questionEvery legal system on earth rests on one fragile assumption: that the person in the robe is incorruptible, unemotional and immune to pressure. No human being has ever met that standard. That is not an insult to judges — most are decent people doing an impossible job. It is the reason the question deserves a serious answer: is the job better done by a machine? We think the honest answer is "partly, and sooner than you think" — and we intend to test it rather than argue it.
It only takes one envelope to poison a verdict, and history is full of envelopes. A machine has no bank account, no career, no family to threaten. In some jurisdictions this alone would justify the research.
When a ruling will be hated, the temptation is not to be right — it is to be popular. A machine does not know which defendant the public loves, and does not care.
A skilled lawyer argues to the man, not the law. Sympathy, disgust, flattery, fear. To a machine, a tearful performance carries exactly the evidence it contains — no more.
Research on real parole hearings found rulings turned harsher as hearings dragged toward lunch. Liberty, decided by hunger. The ten-thousandth case deserves the same attention as the first.
Today a judgment is a black box wearing a robe — one word comes out, and the reasoning stays behind a closed door forever. A machine judgment is a glass box: every rule applied, every fact weighed, every step from evidence to verdict laid out in full, checkable by anyone, disputable line by line. Whatever else this research finds, it will produce something no court currently offers: verdicts you can audit.
Research note 02 — What a jury must actually decide, and why you will never see their reasoning →
Research note 01 — Who else is asking this question? A sourced survey of the field →
Research is only worth the rules it binds itself to. These are ours.
Concluded cases only, complete public records only. The machine never rules on a living person whose case is before a court, and never judges half-heard evidence. That is not caution — that is rather the point.
Every verdict published in full. The entire reasoning chain, source-linked so any reader can re-verify every step against the public record — the same standard Conectir applies to a money trail.
Failures published as loudly as successes. Where the machine reasons badly, misses what a human court caught, or disagrees with a verdict history proved right — that gets published, prominently. Research that only reports its wins is marketing.
No cherry-picking. Cases are chosen for the quality and completeness of their public record and the questions they pose — not for how clever the machine will look.
Courts would not jump from human judges to machine judges in one leap — and they should not. Trust would be earned one rung at a time. This is the ladder the research is testing, rung by rung.
Clear the backlog, translate every ruling into plain language, verify every precedent a lawyer cites, and lock evidence to a tamper-proof trail.
Compare every sentence against every similar case ever decided, and show the inconsistencies in writing. Machines are already better at this than any human alive.
The shadow bench: AI rules on the same case, in parallel, from the same file — and the record of agreement and disagreement builds, year after year, into evidence.
The bench itself — starting where stakes are low and volume is high: traffic, small claims, uncontested matters. Every verdict a glass box. Human appeal, always.
You get the case file. You pass sentence. Then the reveal: what the human court gave, what the machine gives — and how far emotion pulled your own ruling. Sixty seconds, and you are a data point in the research.
A concluded case the whole world still argues about. The full public record, fed to the machine. A complete glass-box verdict, published next to what the human courts decided under the cameras and the pressure. The first real test of the research question.
Judgir rules only on concluded cases with complete public records. We do not judge half-heard evidence — that is rather the point.
Leave your email and you'll see every experiment, every verdict, every finding — including the ones that prove us wrong.