Our Record
Every analysis begins with an administrative record: a political-prisoner registry, a regulatory enforcement docket, a criminal sentencing dataset, an incident archive. We encode the record as a formal knowledge graph, run unsupervised analysis on its structural relations, and test the results with permutation null models and independent reasoner verification. The algorithm is not told what to look for. It finds the patterns the data contains.
Each entry below is a corpus, an ontology, and a finding.
What a Knowledge Graph Found in 1,172 Cuban Charge Sheets
The algorithm found two entirely separate charging regimes that never overlap. Regime A is a street-protest bundle: Contempt, Public Disorder, Assault, and Resistance, charged together as a boilerplate package. Regime B is a single standalone charge (Sedition) applied to a different population. The state sorts its prisoners into categories. The categories are invisible in any individual case file.
What a Knowledge Graph Found in 7,111 Russian Prosecution Records
The most significant structural breaks in Russia's charging architecture are 2017 (p=0.046) and 2019 (p=0.010), not the 2022 invasion. The apparatus was already built. The war scaled it. Three regimes operate in parallel: wartime speech offenses (Art. 207.3, 280.3), organizational persecution (Art. 282.2, 205.5), and a street-protest bundle structurally identical to Cuba's Regime A.
What a Knowledge Graph Found in 3,829 Belarusian Charge Sheets
The grinder does not distinguish. Belarus's charging structure is continuously homogenizing: modularity dropped from 0.26 (2020) to 0.07 (2024). Every year since 2020 is a statistically significant structural break (all p=0.005). In 2025, Art. 361-4 (promoting extremism) overtook Art. 342 (public disorder) as the primary charge for the first time, signaling a pivot from protest suppression to organizational persecution.
What a Knowledge Graph Found in 31,824 Myanmar Detention Records
Two legal systems running in parallel, with the military system progressively displacing the civilian one. Penal Code S:505A (the colonial "fake news" article) collapsed from 62.8% of charges in 2021 to 14.1% by 2026. Counter-Terrorism Law charges, adjudicated in military tribunals, rose from 28.1% to 68.5%. The junta built its repressive architecture in real time, pivoting from British colonial law to its own Counter-Terrorism Law as resistance intensified.
Four countries. Four patterns. One method.
Each pattern was found by the same unsupervised algorithm on the same type of data. None was hypothesized in advance. The method does not impose a framework on the data. It reads the framework the data contains.
| Country | Pattern | Structural signature | Key article |
|---|---|---|---|
| Cuba | Two static regimes | Charge stacking (3.0x lift) | Sedición |
| Russia | Discrete ratchet | Single-charge (85%) | Art. 207.3 |
| Belarus | Continuous grinder | Q: 0.26 → 0.07 | Art. 342 |
| Myanmar | Bifurcation | Speech 63% → 14% | S: 505A + CTL 52-a |
A technique that finds structure in one dataset could be overfitting. A technique that finds different structures in four datasets from four legal traditions on three continents is measuring something real.
The 15-Year Quiet Relocation of Corporate Bribery Enforcement
The SEC still prosecutes foreign bribery. It just does not do it in front of juries anymore. Over a fifteen-year period, enabled by Section 929P of the 2010 Dodd-Frank Act, the Commission progressively relocated corporate FCPA enforcement from federal court (public docket, Article III judge, jury of citizens available) into its own administrative tribunals (SEC-employed ALJ, first-round appeal back to the SEC itself, less discovery, less press coverage). The administrative share of the FCPA docket has held at 78 to 100 percent every year from 2018 to 2024. Individual FCPA prosecutions have compressed to three actions in the five years from 2020 through 2024 (Berko, Batistas, Cabanes), against a corporate settlement pipeline that produced dozens of resolutions in the same period.
What 805,875 Federal Sentencing Records Show About Charge Stacking and Race, FY2014-2025
Federal prosecutors don't just charge offenses. They apply packages, and the package is race-patterned. Louvain community detection on charge co-occurrences across 805,875 federal sentencing events surfaces two stable regimes: a white-collar bundle (fraud, laundering, administration of justice) and a gun/RICO bundle (firearms, racketeering). Which regime a defendant lands in is race-patterned even after controlling for offense severity, criminal history, mandatory-minimum applicability, all 94 district fixed effects, AND primary offense.
What the Knowledge Graph Reveals: Terrorist Groups Break Into New Tactics, They Don't Drift, 1970-2020
Fifty years of coded terrorism data, 209,706 incidents, 232 groups. A changepoint detector run on tactical fingerprints reported 97 structural breaks. Sixty-seven traced to properties of press-derived reporting rather than to any change in group behavior: thin samples, "Unknown" weapon codes, geographic mix shifts driven by coverage collapsing in one country. Thirty survive filtering on adequate samples. Two verified against historical sources (GRAPO 1984 extortion pivot, ELN 2014 escalation ahead of negotiations) hold up.
Foro Penal's registry of political detentions since 2014 includes the post-July 2024 mass detention wave (1,315 verified arbitrary detentions). The revolving-door pattern and ongoing ICC preliminary examination make the charge-pattern evidence directly relevant to submissions to the Office of the Prosecutor.
A NATO member, still in the Council of Europe, subject to ECtHR jurisdiction, with 150,000+ purged post-2016. Anti-terror law charges, Art. 299 (insult to president), and Gulenist organizational membership. The same structural pattern found in Russia and Belarus, inside a Western alliance member.