Sentencia. Ponente. Materia. Análisis.

Audit any Spanish judge or court for bias — in minutes.

Pick a tribunal, a ponente, a topic, and a date range. JurisLens pulls the sentences from CENDOJ, extracts what matters, runs the statistics, and gives you a peer-reviewable report — with every claim cited back to the original ruling.

Free tier · No card · Academic plan free with .edu

Built on
CENDOJ · Centro de Documentación JudicialConsejo General del Poder JudicialArt. 105(b) Constitución Española

The problem

Judicial bias hides in plain sight — in 2,200-sentence haystacks.

Researchers have already shown that the gender of a Spanish judge can shift sentence length by months, and that custody win probabilities skew by who's on the bench.

But every one of those studies took a PhD, a year, and a research grant. Nobody else can verify the result. Nobody else can ask the same question about their court.

JurisLens turns a year of work into a five-minute query.

How it works

Four steps from a question to a citable report.

01

Frame the query

Court, ponente, materia, date range. Pick the axis of bias you want to test — defendant gender, plaintiff gender, judge gender, nationality.

02

Collect the corpus

We pull matching sentences from CENDOJ, cached and rate-limited. Every PDF is stored so the study is fully reproducible.

03

Extract & classify

Regex pulls structured metadata. An LLM classifies parties, charges, and language framings — with citations to the source paragraph.

04

Statistics, not vibes

Chi-square, t-tests, logistic regression — all deterministic, in code. p-values, effect sizes, confidence intervals. No LLM-invented numbers.

What you get

A verdict, not a vibes-check.

No significant bias

"No statistically significant disparity detected across 412 sentences (p = 0.18)."

Underpowered

"Tendency observed but sample too small. Need ~280 more rulings to reach significance."

Significant disparity

"Female defendants receive sentences 3.5 months longer for comparable charges (p = 0.002, n = 412, Cohen's d = 0.41)."

Reproducible

Every report stores the exact query, source hash, and code version. Anyone can rerun and get identical numbers.

Auditable AI

LLM classifications cite the source paragraph. Statistics never touch the LLM — they run in deterministic code.

Legally sound

Targeted, non-commercial research access. Respectful rate limits. Built within CENDOJ's terms of use.

Ask the question. Get the answer.

Free tier includes 2 studies a month, up to 50 sentences each. No card required.