
CourtLab.
Look beyond
the raw rate.
The interactive companion to my Northwestern honors thesis. The decisions, the comparisons, and the full method live in the application.
Legal Researcher · Sep 2025 – Mar 2026 · Departmental Honors & Best Thesis Prize of the Year
This seal identifies the court whose decisions this research studies. It is used for identification only. This is academic research, not affiliated with, endorsed by, or approved by the Seventh Circuit, the federal judiciary, or the United States.
Richard ZhuEmpirical legal research / Sep 2025 – Mar 2026
Every decision,
linked to its opinion.
When appellate outcomes vary across judges, how much reflects the judges, and how much reflects the cases they were given? My Northwestern honors thesis began with that question. CourtLab lets you explore the decisions behind it.
Browse the decisions, open each opinion, and see how the comparisons are built.
CourtLab covers Seventh Circuit criminal decisions from 2020 to 2025. Each record links back to its opinion, and the application sets out the method, the adjustments, and the caveats behind every comparison.
The comparisons are observational: they describe differences across panels after adjustment, not any judge’s views, and they do not predict how a judge will rule.Keep the evidence
within reach.
Begin with decisions
Seventh Circuit criminal decisions from 2020 to 2025, extracted from the public opinions with a schema-bound LLM pipeline and validated against hand-audited fields. Each record links back to its opinion.
Respect the panel
Each decision counts once, and its panel judges share it equally.
Compare with context
A Bayesian model accounts for posture, offense, and whether a decision is precedential before any two judges are placed side by side.
Keep uncertainty visible
Every estimate carries an interval, and the application lets readers change the comparison and watch how much it moves.
Modeled rates describe panel-associated differences after adjustment. They do not identify individual votes or causal effects.