Skip to content

Rapport Cadiet to Article 33: unions, SupraLegem, and the judge‑profiling line

Posted on:April 15, 2019

Rapport Cadiet to Article 33: unions, SupraLegem, and the judge‑profiling line

In 2016, two of us - Anthony Sypniewski and I - hacked together SupraLegem. It wasn’t a start‑up or a media stunt. It was a side tool for tax work: find decisions pointing in one specific direction, enrich each case with metadata (court, formation), and add ML‑assisted tags like who wins, topics, etc. so queries can be very fine. We weren’t trying to “score” judges; the point was method (big data, still on a specific question everything can be checked manually).

Then the Cadiet report happened. The Ministry of Justice ran an official mission - le rapport Cadiet - with auditions of government members, judge unions, courts, CNIL, legal publishers, researchers, university law professors - the whole ecosystem - before writing the rules for opening decisions. It’s all on the ministry’s site, with the hearing notes attached.

And right there, in black and white, SupraLegem shows up as the concrete example that made the abstract risk painfully tangible. In the SJA’s (administrative judges’ union) hearing, they start with: “Un site, baptisé supra legem, a classé les juges sur la base des décisions en droit des étrangers” - and they explain why per‑judge readings (“pour/contre les étrangers”), recusal spikes, and correlations with outside signals (social media, etc.) are a real problem. This isn’t me retelling; it’s the ministry’s minutes.

What the unions actually said (and how SupraLegem figured in)

What the mission wrote down as the real risk

When the mission synthesised all that, it called out the core danger plainly: profilage des magistrats and scoring - i.e., per‑judge analytics that enable targeting, forum shopping, and personalised campaigns; and a justice system weakened if people think outcomes hinge on “who” rather than law. That’s the paragraph where the ministry also footnotes SupraLegem and my Open Law testimony to underline that publishing judges’ names brings no analytic gain for the kind of work we were doing, but it does light up the misuse risk.

Where the law landed (and why it matches the unions’ concern)

Article 33 of the 2019 reform kept the open‑data program moving and drew a bright, enforceable line: you cannot reuse magistrates’ or clerks’ identity data to evaluate, analyze, compare, or predict their professional practices - the very definition of per‑judge profiling. That rule is now in COJ L.111‑13 for the judicial order and CJA L.10 for the administrative order. In short: open decisions, yes; judge‑profiling, no. (Légifrance)

What I was trying to do (and what I learned)

Back at the audition, I said what our code already assumed: names don’t improve the model; they change the ethics. The job was to let people find and replicate reasoning paths (tax, in our case), not to build dashboards of individuals. We built SupraLegem for directional search and metadata/ML enrichment, not to turn judges into features. The ministry’s file ends up telling the same story the unions told: a small, concrete project made the profiling risk undeniable - and that’s the piece the law later carved out.


Minimal references: the mission report and auditions (Ministry of Justice) and the profiling ban (COJ L.111‑13 / CJA L.10) are the backbone for the points above. See the ministry’s page for the Cadiet report and the hearings; and Legifrance for the codified ban. (Cadiet report, Légifrance)