A report from the record · Cuban political prisoners

The rock smelled like him.

In Cuba this is called evidence. A military court accepted it, and Miguel Alfonso Jiménez Marrero is serving the second year of a twelve-year sentence for glass that the state valued at 4.33 euros. He is one of dozens of Cubans convicted on scent-print evidence produced by a single police dog, applied to a rock the security services collected at a time and place the case file does not disclose.¹ He is also one of 1,172 people whose charge sheets we read. Those charge sheets do not read the way you would expect from a criminal system that adjusts to what happened. They read like two different documents produced by two different states, with no overlap between them. The difference is not incidental, not statistical noise, and not the natural variation of criminal law. It is a system for classifying people, operated without disclosure by a state with no independent judiciary, and it meets the definition of arbitrary detention under international human rights law. This report is the proof.

¹ Prisoners Defenders, 8 nuevos presos políticos en Cuba elevan la cifra total a 1.158 a final de mayo (12 June 2025), section "4 nuevos presos políticos civiles en Cuba condenados por Tribunales Militares." Available at prisonersdefenders.org. Jiménez Marrero, 51, was tried by the Sala de lo Penal del Tribunal Militar Territorial Central, in violation of the requirement under UDHR Article 10 that criminal charges against civilians be heard by an independent tribunal. The sentence records only that "one white-colored rock has been seized and attached to the case," without linking the rock either to the alleged material damage or to the accused. The same "huella de olor" technique has been applied against dozens of 11J protesters per the Prisoners Defenders report cited. ↩

Analysis

E. Brattin

Source data

Prisoners Defenders (Cuba)

Method

Formal ontology + permutation testing

Records analyzed

1,172 individuals · 2,264 charges
(May 2026 snapshot)

Background · How the system works

The state that files the charges is the state that writes the law, appoints the judge, and runs the prison.

Cuba is a one-party state. The Communist Party is written into the constitution as the "superior driving force" of the society and the state. There is no independent judiciary. Judges are elected and removable by the National Assembly, which is Party-controlled. The Fiscalía General (Prosecutor's Office) reports up through the same chain. There is no meaningful opposition press, no independent bar association, and no legal framework for public protest that has not been pre-authorized. When a prosecutor files a charge for a political offense, none of the institutions that would push back in a rule-of-law system are structurally able to do so.

The 2022 Penal Code, which replaced the 1987 code, widened the range of expression the state can criminalize. The statutes central to this analysis are written to be broad. "Public Disorder" (Article 272) covers any gathering that "disturbs public order," a term the code does not define. "Contempt" (Article 185) criminalizes offending a public official in the exercise of their duties, including online speech. "Sedition" (Article 120) covers acts intended to subvert the constitutional order and carries seven to fifteen years. "Enemy Propaganda" and "Propaganda against the Constitutional Order" cover the production or distribution of material critical of the government. These articles are broad enough that a Facebook post, a slogan shouted at a march, a cardboard sign, a livestream, or an act of standing still in a public square can each independently support a charge, and often several at once.

The practical result is what Cuban defense lawyers and families have described for years as the expediente armado, the "built file." Prosecutors do not need to demonstrate that a specific act corresponds to a specific charge. They need to file a charge that a Party-supervised judge will accept. In street cases the file is built from the boilerplate bundle. In cases the state wants to make an example of, the file is built around a single grave charge that carries a decade. Both are decided before trial. The trial confirms what the charging decision already established.

This is the mechanism the analysis below makes measurable. The two-regime pattern is not something Cuban prosecutors invented. It is a straightforward consequence of what happens when overbroad statutes meet an unaccountable charging apparatus: prosecutors pick from a menu, and the menu has two sections.

The finding

The menu has two sections. They never overlap.

Given the mechanism described above, we would expect prosecutors to combine charges in whatever way maximizes conviction and sentence length. What we find instead is more specific and more revealing. The charges do not combine freely. They cluster into two entirely separate groups. Cases built from the street bundle almost never include a grave charge. Cases built around a grave charge almost never include the street bundle. The state treats these as different kinds of case. And the separation is stable across the entire five-year dataset.

Regime A · The street bundle

Applied to protest and public assembly

A boilerplate combination filed against people arrested at demonstrations, marches, or anti-government expression in public space. The charges reinforce each other and compound the sentence without requiring independent evidence for each.

Desórdenes Públicos Desacato Atentado Resistencia

Contempt + Public Disorder appear together 288 times, against a chance expectation of ~94. That is 3× more than random. p < 0.001.

Regime B · The standalone charge

Applied to frame individuals as regime threats

Serious political offenses used on their own, without the street bundle. These charges carry 7 to 15 years and are levied against people the state wants to characterize not as unruly citizens but as enemies of the constitutional order.

Sedición Sabotaje Propaganda Enemiga

Sedición appears with Public Disorder only 10 times, against a chance expectation of ~57. That is 5× less than random. This separation is deliberate.

The evidence

What the numbers say when you run them against random chance.

For every pair of charges, we asked a simple question: if the state handed out charges randomly (preserving how often each one appears overall), how often would this specific pair happen together? We ran that random reshuffle 5,000 times to build a null distribution, then compared it to what actually happens. The results below show the pairs that are the strongest evidence of a deliberate bundling pattern, and (at the bottom) the pairs that reveal the opposite: charges the state keeps apart on purpose.

Charge pair	Observed together	Expected by chance	Lift	Verdict
Contempt + Public Disorder	288	94	3.0×	Bundled
Assault + Public Disorder	253	88	2.9×	Bundled
Assault + Contempt	175	63	2.8×	Bundled
Public Disorder + Sabotage	71	31	2.3×	Bundled
Public Disorder + Resistance	30	11	2.8×	Bundled
Contempt + Resistance	26	8	3.4×	Bundled
Sedition + Public Disorder	10	57	0.2×	Kept apart
Sedition + Assault	9	38	0.2×	Kept apart
Contempt + Sabotage	14	22	0.6×	Below chance

How to read this A “bundled” pair appears far more often than chance would predict, and reveals a formulaic prosecution template. A “kept apart” pair appears far less often than chance would predict, meaning the state actively avoids using those charges together. Both directions are evidence of deliberate design.

Case in point · July 2021

The single day that fills the dataset.

On July 11, 2021, a mass protest against food shortages and government policy erupted across Cuban cities. What followed is quantified in the dataset with a precision the state has never acknowledged: nearly 500 arrests in one week, concentrated in the first three days, spread across Havana, Artemisa, Matanzas, and Santiago de Cuba. These arrests account for more than a third of the entire dataset.

267

Jul 11

117

Jul 12

38

Jul 13

16

Jul 14

24

Jul 15

20

Jul 16

25

Jul 17

In the five years of data since, no comparable second wave has occurred. Later mass-arrest days are regional (Granma in March 2024, Villa Clara in November 2024, Cienfuegos in May 2024) and top out in the double digits. The July 2021 spike is not a pattern. It is a singular event that the state's charging apparatus is still processing.

Family separation

Most prisoners are held near home. The exceptions are far.

Of the 674 prisoners we can locate both by home province and prison, 85% are held in their home province. But the tail matters. Fourteen percent are moved out of their province, and seventeen are held more than 600 kilometers from home. In a country 1,250 km end to end, that is the difference between a family being able to visit and a family that cannot.

85.8%

Held in home province

The dominant pattern. Most facilities serve their local region, and most prisoners are transferred locally when transferred at all.

14.2%

Displaced from home

The exception, and where evidence of intent lives. Long-distance transfers cross the island and functionally end in-person family contact.

01

Pinar del Río

→

Guantánamo

915 km

02

La Habana

→

Guantánamo

810 km · four cases

03

Guantánamo

→

La Habana

810 km

04

Artemisa

→

Santiago de Cuba

782 km · two cases

The argument

Why this pattern satisfies the legal definition of arbitrary detention.

Everything above has shown what the pattern is and that it holds. This section addresses the legal question the pattern raises. A skeptical reader is right to demand more than a statistical result. Charges cluster in every legal system. Prosecutors specialize. Cases differ. If Sedición just happens to arise from different fact patterns than Desórdenes Públicos, no violation follows. The argument below is that this is not what is happening in Cuba, and that the pattern the data reveals meets a specific evidentiary threshold recognized in international human rights law. Five steps, each of which must hold for the argument to hold, and each of which is answerable if any part is wrong.

01

The statutes overlap by design, so charging decisions are choices, not conclusions.

Article 120 of the 2022 Cuban Penal Code (Sedición) is defined as "tumultuous action" that "disturbs the socialist order" or "impedes the fulfillment of laws or decisions of authorities." Article 272 (Desórdenes Públicos) is defined as any gathering that "disturbs public order." The elements of Article 272 are a subset of the elements of Article 120 as applied to any collective political act. Any conduct satisfying Article 120 in a public context necessarily satisfies Article 272. The state's choice to file one and not the other, or both, is a discretionary decision by the Fiscalía, not a legal necessity arising from the acts. This is a legal reading of the statutes, not a statistical claim. [Ley No. 151 de 2022, Arts. 120, 272; Gaceta Oficial No. 93 Extraordinaria, 1 Sep 2022]

02

The same underlying acts produce different regimes for different people.

The 11 July 2021 arrests are the natural experiment. More than five hundred individuals were detained within one week for participating in the same protests, in the same cities, often at the same corners. The corpus records their charges. Under a charging practice tied to the acts alleged, one would expect broadly similar charge sheets: the same set of statutes triggered by the same underlying conduct. What the record shows is that some 11J arrestees received the street bundle (Desórdenes + Desacato + Atentado + Resistencia) and others received Sedición filed alone or with Sabotaje. The two groups did the same thing. The state charged them differently. The difference therefore lies not in the conduct but in the state's assessment of the person. [Corpus §3.3 stability across 11J subset; HRW Prison or Exile (2022) documents identical conduct across defendants receiving different charge combinations]

03

The state cannot predict its own regime assignments from any lawful criterion.

If regime assignment were tied to the acts alleged, we would expect features of the incident to predict the outcome. If it were tied to a lawful person-based criterion, we would expect measurable personal characteristics (prior international documentation, IACHR precautionary-measure status, organizational membership) to predict it. Empirical analysis of 871 Regime A/B defendants finds neither. Prior international documentation is uncorrelated with Regime B assignment (Fisher OR = 0.96, p = 0.81). IACHR-protected persons are, if anything, slightly less likely to face Regime B (25.0% vs 30.5%, p = 0.14). The only statistically significant predictor is arrest during the 11 July 2021 wave (OR = 1.42, p = 0.027). Regime assignment therefore tracks the moment of arrest, not the person and not the act. Under the Working Group on Arbitrary Detention's Category III jurisprudence, discretion exercised without disclosed criteria, without reasoned judgments, and without independent review constitutes arbitrariness regardless of the direction in which the discretion falls. [HRC General Comment No. 35, CCPR/C/GC/35 (16 Dec 2014), ¶12, 17; UDHR Arts. 3, 7, 9, 10; WGAD Deliberation No. 9 (A/HRC/22/44, ¶63); analysis: selection_summary.md. Note: Cuba has signed but not ratified the ICCPR; legal arguments route through UDHR and customary international law.]

04

Cuba's institutional structure removes every check on this discretion.

Prosecutorial discretion is a normal feature of criminal systems. It is not arbitrary where it is bounded by (a) written charging guidelines, (b) disclosure to defense counsel, (c) reasoned judgments explaining charge selection, and (d) independent judicial review. Cuba has none of these. The Fiscalía General reports to the Party-controlled National Assembly (Constitution Art. 156). Judges are elected and removable by the same body (Arts. 148, 149). No independent bar association exists. The Cuban Criminal Procedure Code does not require prosecutors to disclose the basis for charging selection to defense counsel. Judgments in political cases do not typically address why one charging regime rather than another was selected. Discretion exercised without any of these constraints is not "discretion" in the sense that international human rights law recognizes. It is unconstrained power, which the Working Group has repeatedly found gives rise to arbitrary detention. [Constitución de la República de Cuba (2019), Arts. 5, 148, 149, 156; WGAD Deliberation No. 9 concerning the definition and scope of arbitrary deprivation of liberty under customary international law (A/HRC/22/44, 24 December 2012); WGAD Opinion No. 63/2021 concerning Cuba]

05

This exact configuration has been repeatedly found arbitrary by the Working Group's own Cuba jurisprudence.

The Working Group on Arbitrary Detention has issued at least nine Opinions concerning individual Cuban political detainees between 2021 and 2025, all finding the deprivation of liberty arbitrary and requesting release with compensation. Opinion No. 63/2021 concerning Maykel Castillo Pérez explicitly noted the "recurrent pattern of arbitrary detention" the Working Group had identified in Cuba and urged the Government to accept a country visit. Opinion No. 13/2024 addressed seventeen named 11J protesters at once. Cuba has implemented none of these Opinions. The submission this report supports offers the systemic-level frame within which those individual Opinions can be understood not as a series of anomalies but as evidence of the designed prosecutorial architecture documented above. The Inter-American Court's judgment in Norín Catrimán v. Chile (2014) is instructive by analogy on how the Inter-American system treats the selective application of overbroad political-offense statutes in jurisdictions with limited independent judicial review, though Cuba's status outside the American Convention means the primary vehicle for this argument is the Working Group's own developing Cuba caseload. [WGAD Opinions No. 41/2021, 63/2021, 37/2022, 52/2022, 51/2023, 13/2024, 66/2024, 46/2025, 57/2025 (all Cuba); Norín Catrimán y otros v. Chile, IACtHR Ser. C No. 279 (29 May 2014)]

Falsification testing Step 3 as originally drafted for this report predicted that persons with prior international documentation would be over-represented in Regime B. That prediction failed. Among 871 A/B defendants, documented persons appear at 28.6% Regime B versus 29.5% for undocumented persons (Fisher exact p = 0.81). Step 3 above was rewritten to reflect what the data actually shows: neither act-level nor person-level features predict regime assignment in the way any lawful discretionary system would require, and that absence of any measurable predictive criterion is itself the arbitrariness the Working Group's jurisprudence identifies. The failed hypothesis is preserved in the analysis annex; the finding survived falsification testing rather than being confirmed by it, which is the stronger form of the claim.

Independent validation (Step 09) As an independent validation of the two-regime hypothesis, Louvain community detection (Blondel et al. 2008) was applied to the charge co-occurrence graph at the default resolution parameter (1.0). The algorithm was given no prior knowledge of the analyst-defined regime partition. It independently assigned Sedición to a separate community from the four Regime A charges (Desacato, Desórdenes Públicos, Atentado, Resistencia), which were placed together in a single community. This separation was perfectly stable across 100 runs with different random initializations (100/100). Per-prisoner classification agreement between the analyst-defined regimes and the algorithmically discovered partition was 97.2% for Regime A and 98.0% for Regime B (09_community_detection.py; results archived in 09_community_detection_results.json). The two-regime structure is not an artifact of the analyst's framing; it is a property of the data that an unsupervised algorithm recovers independently.

What follows

If the argument holds, three things become possible.

The five-step argument at the top of this page is the case. Everything below it, from the numbers to the case files to the timeline, is the evidence the case rests on. Assume for the moment that the argument holds. Three consequences follow that the current human rights record on Cuba does not yet reflect.

Consequence 1 · For pending Working Group opinions

Individual cases can be argued from systemic ground.

The Working Group on Arbitrary Detention has issued opinions on individual Cuban detainees (Ferrer, Otero Alcántara, and others) finding arbitrariness on case-specific grounds. The pattern documented here lets future opinions cite systemic evidence, not just the facts of one detention. That shift matters: systemic findings survive individual releases, they generalize to future cases without requiring re-litigation, and they create pressure for statutory reform rather than ad hoc clemency.

Consequence 2 · For treaty-body reporting

The next UPR cycle has a citable pattern, not an anecdote.

At Cuba's Fourth Cycle review in 2023, states raised concerns about political detention in qualitative terms that Cuba was able to characterize as politically motivated. A quantified finding with reproducible methodology is harder to characterize away. It is not a claim about intent. It is a claim about pattern of practice, which is the evidentiary standard the treaty bodies actually apply.

Consequence 3 · For the record itself

The unwritten category is now written down, in a form the state cannot revise.

The Cuban state maintains the two-regime distinction without publishing it. It exists in the pattern of decisions prosecutors make. Until this analysis existed, that pattern was known only to the people it was being done to and to the lawyers and families who represent them. What a formal ontology does is give the pattern a name, a definition, and a citable public form. The state can no longer treat the distinction as an artifact of individual cases. It is a documented feature of the system, whether the system chooses to acknowledge it or not.

Method note · What we actually did

We took the public database of 1,172 documented political prisoners published by Prisoners Defenders and encoded every record in a formal ontology: a machine-readable schema that separates the state's own terminology from our analytical assessment. Charges, arrests, sentences, and facilities were modeled as distinct entities so that each could be counted, cross-referenced, and validated against international human rights instruments (ICCPR, UN CAT, ILO Convention 29).

The charge-bundling analysis uses a permutation test with 5,000 iterations. For each pair of charges, we compared the observed co-occurrence count to a null distribution generated by randomly redistributing charges across defendants while preserving each defendant's total charge count and each charge's overall frequency. Pairs whose observed count exceeds the 95th percentile of the null distribution (empirical p < 0.05) are reported as bundled. Pairs whose observed count falls well below the null mean are reported as kept apart. The two-regime finding holds under stricter thresholds (p < 0.001) and under alternative null models.

The ontology itself is the artifact that lets the finding stand up. Without a formal T-Box distinguishing charge instances from charge types, without SKOS separation of the state's terminology from our analytical assessment, and without SHACL validation of provenance on every claim, the same statistical result would be a chart in a blog post. With them, it is a citable evidence artifact whose every claim traces to a SPARQL query against a versioned graph.

The rock is still in the file. The file is still on the shelf. Miguel Alfonso Jiménez Marrero is still in the second year of twelve. The ontology does not open the cell. It does one smaller thing. It writes down what the state has done, in a form the state cannot revise, in a language other institutions can read. That is not a remedy. It is a precondition for one.