Juan Antonio Matiz Monroy — Lawyer, Universidad de los Andes · Specialist in Criminal Law, Universidad Sergio Arboleda · Professional licence 294.746 (C.S.J.)
At the pre-trial detention hearing almost always the same things are argued: seriousness, purposes and proportionality. That argument has to be made, but it comes after another that is almost never made. What is established, on what material and to what standard, which is the question prior to all the others.
The law compresses that standard into two words, «reasonable inference», and does not say what they mean. The courts have said what they are not, and scholarship has explained why they admit no figure. This article gathers the three sources and proposes how to review the inference without merely asking that it be named.
It is one layer among others, because the facts, the rules and doctrine weigh too. What procedure requires for the measure is set out in the article on what the prosecution must prove. Here we look only from the evidence side, which is the layer that usually goes unexamined.
1. What the law says and what it does not say
The expression appears twice in the Code of Criminal Procedure, with the same verb and a different object. Article 287 allows charges when, from what has been gathered, it «can reasonably be inferred» that the accused «is the author or a participant» in the offence. Article 308 requires inferring that he «may be the author or a participant», and adds one of three requirements; neither defines the expression.
The first thing the expression is not, though it is used that way daily: inner conviction. To «infer» is to move from what is on record to what is asserted, and «reasonable» qualifies that step, not a state of mind. The law does not say the judge must be convinced; it says the conclusion must follow from the material.
The second thing it is not, though it is sometimes demanded: the knowledge proper to a conviction. To indict, «probability of truth» is required (article 336), and to convict, «knowledge beyond all reasonable doubt» (article 381). Reasonable inference is the lowest degree of that scale, and article 295 reads it restrictively when liberty is at stake.
2. Why the law does not define it, and nor does the Court
The courts have surrounded the expression with what it is not, and that already imposes order. Since 1994 the Constitutional Court has said that well-founded grounds are «facts, factual situations» (C. Const., Sent. C-024, ene. 27/94). It added that «mere suspicion or the simple conviction» is not enough, and in 2001 it required serious, lawfully proven indicia (C. Const., Sent. C-774, jul. 25/2001).
The Inter-American Court said it for the region: suspicion must be «founded on specific facts, that is, not on mere conjecture» (Corte IDH, Caso Norín Catrimán y otros vs. Chile, sentencia de 29 de mayo de 2014, Serie C No. 279, párr. 311). It is a limit, not a definition, and the Supreme Court has not gone much further either.
The Supreme Court defined reasonable inference as an exercise of extraction (CSJ, Cas. Penal, Auto AP5910-2015, oct. 8/2015, Rad. 46526). «Given certain elements, from them a conclusion or a hypothesis can be extracted or deduced». It is reasonable «if it rests on logical parameters», opposed «to the intuitive, to the instinctive».
It is a definition of form, not of degree, and the Court itself has wavered on the degree. For charging it has said that «the degree of likelihood on which this rung rests is a simple possibility» (CSJ, Cas. Penal, Sent. sep. 16/2009, Rad. 31795). Other decisions bring the expression closer to probability, according to Andrés Díaz’s review on the Externado criminal law blog.
The Constitutional Court, for its part, upheld article 308 without fixing the point (C. Const., Sent. C-695, oct. 9/2013). That is, it did not say how much is enough either, and that silence has an explanation.
It is worth understanding why no court has filled that gap, because the reason is not laziness. A standard of proof has two parts: how a hypothesis is justified and how much is enough. The first is epistemological and can be demanded; the second is a political decision, and belongs to the legislature.
Carmen Vázquez, a legal philosopher who studies expert evidence, sums it up: epistemology gives «the how but not the how much». And the how much cannot be a figure, because the probability judges handle is logical, not mathematical. A number would only move subjectivity somewhere else, with an appearance of exactness.
From that follows the consequence that orders the rest of this article, and it is worth keeping. If the legislature did not fix the how much, review falls on the how, and it is more demanding than it looks. Daniel González Lagier, a legal philosopher, condenses it thus: the accepted hypothesis must be «the best justified» against its rivals.
To say «reasonable inference» amounts, on that reading, to requiring that the prosecution’s hypothesis be the best explanation of the data. Of the data there is, at this moment and with this material. Read that way, «reasonable» stops naming a conviction and starts naming a justification examinable from outside.
3. How it is reviewed
Every evidentiary inference has three parts, and all three can be examined separately. The datum is what is on record, the hypothesis is what is asserted, and the warrant is the rule connecting them. That rule, called a maxim of experience, is almost never stated, and so review begins by asking for it.
Once stated, the inference is put to four questions that the theory of evidence has refined. They assign no figures, because there are none; they say whether one hypothesis is better supported than another. They are the same four that are taken to the hearing, and so it is worth learning them in order.
Is the datum reliable?
A datum is worth what its origin is worth, and not all of them enter the hearing alike. A technical result from a controlled procedure is not on the same plane as the conclusion of another, earlier inference. The Supreme Court quashed a conviction because the indicium «was built upon another indicium» (CSJ, Cas. Penal, Sent. may. 12/2004, Rad. 19733).
The questions are concrete and are asked with the file in hand. How the datum was obtained, who recorded it and when, and whether it is an observation or already an interpretation. And the most practical one: whether the item is in the file or was only mentioned in submissions.
What kind of rule is it?
Here is the work that yields most, because rules are not all alike. There are scientific laws, high-probability generalisations, common-sense generalisations and prejudices in the shape of a rule. The kind of rule sets the ceiling of the conclusion, and so it is worth naming it.
«Whoever flees usually has something to hide» is common sense, and admits so many exceptions that it sustains little. «Whoever is caught with the stolen goods shortly after the theft probably took part» is a high-probability generalisation. «Whoever has that character tends to kill» is a prejudice, and the Supreme Court rejected it as a rule of personality (CSJ, Cas. Penal, Sent. may. 12/2004, Rad. 19733).
The more invasive the measure, the more demanding the kind of rule that sustains it. For deprivation of liberty, common sense and prejudice should not be enough. And a true rule may not govern the case: the question is whether it governs these facts, not merely whether it is true.
Is the hypothesis the best explanation of these data?
This is the question that turns review into a standard, with no need for any figure. The prosecution’s hypothesis must explain the data coherently, and better than the rivals compatible with the same data. If the data fit two accounts equally, what there is is a suspicion.
The quantity and variety of the data count, because each kind of datum eliminates different alternatives. Three witnesses to the same rumour are a single datum, not three converging data. The Council of State condemned a detention built on «sub-indicia», suspicions added together that «were no more than mere suspicions» (C.E., Sec. Tercera, Sent. ago. 15/2002, Exp. 13671).
For the measure it is not required to refute every alternative, as it is in order to convict. It is required that the alternative not explain the same material equally well, and that allows work without evidence of one’s own. It is enough to show that the datum admits another reading and that the application did not close it off.
The law uses the same expression for charging and for the measure, but article 295 separates them. Charging affects reputation and the measure affects liberty, which is read restrictively. The demand for justification grows with the intensity of the measure, though the method is the same.
There is also a textual datum that favours that reading. Article 287 requires inferring that the accused «is» the author; article 308, only that he «may be», despite costing liberty. That is why article 295 orders it to be read restrictively: the inference must exist and hold, not merely be mentioned.
What is missing that was easy to obtain?
The last question looks at what is not there, and it is the one asked least. The weight of a body of material depends on its completeness, not only on its direction. A hypothesis resting on few data, when more were within reach, weighs less even if the few agree.
The prosecution chooses when to apply for the measure and is not obliged to do so with the charges. If it sought deprivation of liberty with a video it had not watched or an interview it had not checked, that incompleteness is reviewable. Contesting, under article 306, includes pointing out what is missing and what it would have changed.
Looking forward: the risk is a forecast
The second requirement of article 308 looks to the future, and there the rule weighs more than the datum. A future fact is not proved; past facts are proved, together with an empirically supported rule that allows the fact to be expected. The circumstances in articles 310 to 312 are indicia, not presumptions, and the risk «is not presumed» (Corte IDH, Caso Norín Catrimán y otros vs. Chile, sentencia de 29 de mayo de 2014, Serie C No. 279, párr. 312).
The provisional legal characterisation is not decisive, by express command of the first proviso to article 308. The seriousness of the conduct is not enough on its own either (C. Const., Sent. C-1198, dic. 4/2008), and the risk needs data of its own. If the only datum for the risk is the charge itself, there is no forecast; there is repetition.
How it is put at the hearing
Article 306 opens the space, because it allows the material supporting the application to be contested. The Supreme Court requires whoever applies for the measure to present and explain the evidence (CSJ, Cas. Penal, Sent. STP7721-2019, jun. 11/2019, Rad. 104439). And it requires the judge to deny what was not «lawfully substantiated» (CSJ, Cas. Penal, Sent. SP1832-2019, may. 28/2019, Rad. 53888): naming the inference is not making it.
Where the explanation does not appear, the defence does not bring another version: it points out which question went unanswered. With what rule does the prosecution move from this datum to that conclusion, and what other hypothesis explains the same thing? What item within reach was not collected before asking for a person’s liberty?
That work is done beforehand, with the file in hand, and is not improvised in the room. It also has a technical advantage: the Court reviews inferences for false reasoning with a fixed scheme (CSJ, Cas. Penal, Auto AP1567-2026, mar. 11/2026, Rad. 67049). What the item reveals, which rule was disregarded, what the correct inference was and why the error was decisive.
You may also like: pre-trial detention: what the prosecution must prove, what to do in the first 36 hours after an arrest and how much evidence is enough to convict.
4. Conclusions
The law requires a reasonable inference for charging and for imposing pre-trial detention, but does not define it. The courts have said what it is not: not inner conviction, not mere suspicion, not the knowledge proper to a conviction. How much is enough is a decision for the legislature, and it cannot be expressed as a figure. What can be demanded is how it is justified: that the prosecution’s hypothesis be the best explanation of the data.
That review is ordered in four questions. The first, whether the datum is reliable; the second, what kind of rule connects it to the conclusion. The third, whether the hypothesis explains better than its rivals; the fourth, what was missing that was easy to obtain. The risk, for its part, is a forecast that is not presumed and needs data of its own. Article 306 allows all of this to be contested, and the judge must deny what was not substantiated.
That analysis is prepared before the hearing, with the file in hand, and is not improvised in the room. In the initial consultation we review with you the material supporting the application and the rules the prosecution uses to connect it. On that study, the firm can take on representation at preliminary hearings and the criminal defence in the stages that follow.
Frequently asked questions
What people usually ask us
What is reasonable inference in a pre-trial detention decision?
It is the level of knowledge that article 308 of Ley 906 de 2004 requires in order to assert that the accused may be the author of or a participant in the offence. The law does not define it; the Supreme Court describes it as logical, not intuitive, reasoning that extracts a hypothesis from specific elements. Interpreted with the statute, the case law and scholarship, it requires the prosecution’s hypothesis to be the best explanation of the available data, not a conviction held by the judge.
Is reasonable inference the same for charging and for imposing the measure?
The law uses the same expression in articles 287 and 308, but article 295 orders whatever deprives of liberty to be interpreted restrictively. Recent scholarship proposes demanding more justification when the measure deprives of liberty, in particular a high-probability rule of experience and data of its own for the risk. It is an interpretive proposal, not a written rule.
Why can it not be defined with a figure?
Because the probability judges use is logical or inductive, not mathematical: it allows one to say that a hypothesis is better supported than another, but not by how much. And because fixing how much is enough is a decision about which error is preferred, which belongs to the legislature. What can be reviewed are the criteria of justification: reliability of the datum, kind of rule, rival hypotheses and completeness.
How is a maxim of experience reviewed?
First by asking for it to be stated, because it is almost never said aloud. Then by asking what kind it is: scientific law, high-probability generalisation, common sense or prejudice. And finally whether it governs this case, because a true rule may not apply to the specific facts. For a measure depriving of liberty, common sense and prejudice should not be enough.
To what standard must the risk be established?
The Code does not fix it uniformly: serious and well-founded grounds for obstruction (art. 309), well-founded grounds for danger to the victim (art. 311), probability for non-appearance (art. 308, num. 3) and no formula at all for danger to the community (art. 310). In every case the risk is proved with past facts and an empirically supported rule, and is not presumed.
Can the defence attack the inference without evidence of its own?
Yes. It does not need to prove an alternative version. It is enough to show that the datum admits more than one reading and that the application did not rule out the others, that the rule is common sense, or that what was easy to obtain was left uncollected.
What happens if the judge imposes the measure without explaining the inference?
The Supreme Court requires the application to be explained and the judge to deny what was not lawfully substantiated. The decision can be appealed, and the attack must follow the same technique the Court requires for false reasoning: what the item reveals, which rule was disregarded, what the correct inference was and why the error was decisive.
Legal and bibliographic references
Legislation. CPP (L. 906/2004), arts. 7, 26, 287, 295, 306, 307 (par. 2.º), 308, 309, 310, 311, 312, 336 y 381. C.N., arts. 4, 28, 29, 93, 230 y 250. L. 153/1887, art. 4.
Constitutional Court. C. Const., Sent. C-024, ene. 27/94. M.P. Alejandro Martínez Caballero (well-founded grounds are objective, verifiable factual situations; mere suspicion or the officer’s conviction is not enough). C. Const., Sent. C-774, jul. 25/2001. M.P. Rodrigo Escobar Gil (formal and substantive requirements of detention; serious indicia «based on evidence lawfully produced»; the concept is indeterminate and its scope is for the legislature and the courts to fix; adopts the Supreme Court’s formula on the seriousness of the indicium as a review of soundness after contemplating «all the confirming and disconfirming hypotheses»). C. Const., Sent. C-1198, dic. 4/2008. M.P. Nilson Pinilla Pinilla (the seriousness and manner of the conduct are not enough on their own for danger to the community). C. Const., Sent. C-695, oct. 9/2013. M.P. Nilson Pinilla Pinilla (art. 308, num. 3 upheld; reasonable inference is neither certainty nor a judgment of responsibility; no threshold fixed).
Supreme Court of Justice. CSJ, Cas. Penal, Sent. may. 12/2004, Rad. 19733. M.P. Jorge Aníbal Gómez Gallego (review of circumstantial evidence: proven indicating fact, explicit and applicable rule of experience, contemplation of all hypotheses, prohibition of indicium upon indicium and of the rule of personality; «mere and weak probabilities» are resolved in the accused’s favour). CSJ, Cas. Penal, Sent. jul. 8/2009, Rad. 31280. M.P. Julio Enrique Socha Salamanca (charging «sits in the terrain of possibility» and the indictment requires «a degree of probability of truth»). CSJ, Cas. Penal, Sent. sep. 16/2009, Rad. 31795. MM.PP. Yesid Ramírez Bastidas y Julio Enrique Socha Salamanca («the degree of likelihood on which this rung rests is a simple possibility», on art. 287). CSJ, Cas. Penal, Sent. oct. 26/2011, Rad. 36357. M.P. Julio Enrique Socha Salamanca (the presumption of innocence is normative; the decision must «set out arguments capable of being tested»; maxims of experience are refuted empirically and argumentatively; prejudices beg the question). CSJ, Cas. Penal, Auto AP5910-2015, oct. 8/2015, Rad. 46526. M.P. Fernando Alberto Castro Caballero (origin of the definition: the inference is an exercise in which «given certain elements, from them a conclusion or a hypothesis can be extracted or deduced», reasonable «if it rests on logical parameters»; Justice and Peace, art. 18B of Ley 975). Reiterated in CSJ, Cas. Penal, Auto AP968-2020, abr. 22/2020, Rad. 56715. M.P. Gerson Chaverra Castro (whoever proposes the inference bears the burden of supplying the elements that sustain it; Justice and Peace). Reiterated in CSJ, Cas. Penal, Auto AP1046-2023, abr. 19/2023, Rad. 59628. M.P. Myriam Ávila Roldán (second instance, detention measure in Justice and Peace; upholds the refusal because knowing of the act and having been invited to take part does not entail co-perpetration: non sequitur). CSJ, Cas. Penal, Sent. STP7721-2019, jun. 11/2019, Rad. 104439. M.P. Patricia Salazar Cuéllar (duty to present and explain the evidence at the level of knowledge set by law). CSJ, Cas. Penal, Sent. SP1832-2019, may. 28/2019, Rad. 53888. M.P. Patricia Salazar Cuéllar (containment function of the judge of guarantees; duty to deny an unsubstantiated application; the judge does not complete what the prosecution did not substantiate). CSJ, Cas. Penal, Auto AP1567-2026, mar. 11/2026, Rad. 67049. M.P. Gerardo Barbosa Castillo (argumentative burden for false reasoning). On the oscillation between «simple possibility» and «probability», see Díaz, Andrés. «¿Qué es una inferencia razonable? (Segunda parte: “razonable”)». Blog de Derecho Penal y Criminología, Universidad Externado de Colombia, 2025, which attributes the «simple possibility» formula to Rad. 31280 and 31795 of 2009, Rad. 44599 of 2017 and SP095-2023 (Rad. 60133), and the approach to «probability» to SP10944-2017 (Rad. 47850) and AP5493-2024 (Rad. 66586); the last two could not be consulted in full text for this article.
Council of State. C.E., Sec. Tercera, Sent. ago. 15/2002, Exp. 13671. C.P. Alier Hernández Enríquez («sub-indicia» are suspicions that do not amount to a serious indicium, separately or together).
International sources. Corte IDH. Caso Norín Catrimán y otros (dirigentes, miembros y activista del pueblo indígena mapuche) vs. Chile. Fondo, Reparaciones y Costas. Sentencia de 29 de mayo de 2014. Serie C, No. 279, párrs. 311-312 (suspicion must be founded on specific facts; procedural risk is not presumed). CIDH, Informe sobre el uso de la prisión preventiva en las Américas, OEA/Ser.L/V/II. Doc. 46/13, 30 de diciembre de 2013, párr. 185 (the risk must be founded on objective circumstances; the mere invocation of the grounds is not enough).
Colombian scholarship. Velásquez Velásquez, Fernando. Fundamentos de derecho penal. Parte general. Tercera edición. Bogotá: Tirant lo Blanch, 2020, pp. 10-17 (dogmatics as the science of criminal law: concept, three-phase method and functions), 110-111 (procedural criminal law and its science), 125-138 (sources: statute as the only direct source; case law and scholarship as auxiliary criteria) and 175-188 (interpretation: possible literal meaning, systematic, teleological and historical elements, rules of interpretation, interpretation in conformity with the Constitution). Ramos López, Juan Carlos. Estándar de prueba de la medida de aseguramiento en Colombia. ¿Es la inferencia razonable un estándar preciso, unívoco y claro para la imposición de medidas de aseguramiento? Master’s thesis in Law, Criminal and Criminological Sciences. Bogotá: Universidad Externado de Colombia, 2024 (reasonable inference lacks a threshold; the judicial definition «does not help to define or settle the content»; proposal of three conditions, among them that the set of elements «cannot be explained from another hypothesis»; question-begging between authorship and purposes). Ortega Botina, Víctor Camilo. «El estándar de prueba de inferencia razonable: fundamento epistemológico del auto que decreta la medida de aseguramiento preventiva». Revista Criterios, vol. 30, n.º 2, 2023, pp. 83-98 (the definition omits «information on its application in legal practice»; Toulmin-Ferrer scheme for reasoning the decision). Marín Vásquez, Ramiro Alonso. «El estándar de prueba de conocimiento más allá de duda razonable». Derecho Penal y Criminología, vol. 35, n.º 99, 2014, pp. 113-138. Agencia Nacional de Defensa Jurídica del Estado. Esquema de decisión adecuada para imponer la medida cautelar de privación de la libertad. Documentos Especializados n.º 9. Bogotá: ANDJE, 2014 (principal investigator Óscar Julián Guerrero Peralta).
Scholarship on evidential reasoning. González Lagier, Daniel. Quaestio facti. Ensayos sobre prueba, causalidad y acción. Lima-Bogotá: Palestra-Temis, 2005, cap. II (structure of the evidentiary inference); and «Distinciones, estipulaciones y sospechas sobre los criterios de valoración y los estándares de prueba». En Johann Benfeld y Jorge Larroucau (eds.), La sana crítica bajo sospecha. Valparaíso: Ediciones Universitarias de Valparaíso, 2018, pp. 15-32 (criteria of soundness: quantity, variety and relevance of the data, kind of warrant, coherence and explanatory power of the hypothesis; logical probability admits no quantification; every standard requires the hypothesis to be the best justified). Ferrer Beltrán, Jordi. «Prolegómenos para una teoría sobre los estándares de prueba». En Diego Papayannis y Esteban Pereira Fredes (eds.), Filosofía del derecho privado. Madrid: Marcial Pons, 2018, pp. 401-430; Prueba sin convicción. Estándares de prueba y debido proceso. Madrid: Marcial Pons, 2021 (methodological requirements of a standard; progressive standards by procedural stage); «Los momentos de la actividad probatoria». En Jordi Ferrer Beltrán (coord.), Manual de razonamiento probatorio. México: Suprema Corte de Justicia de la Nación, 2022. Vázquez, Carmen. De la prueba científica a la prueba pericial. Madrid: Marcial Pons, 2015 (epistemology says the how, not the how much). Cohen, L. Jonathan. The Probable and the Provable. Oxford: Clarendon Press, 1977 (inductive versus mathematical probability). Haack, Susan. Evidence Matters. Science, Proof, and Truth in the Law. Cambridge: Cambridge University Press, 2014 (epistemology of evidence; degrees of proof are not mathematical probabilities). Laudan, Larry. «Por qué un estándar de prueba subjetivo y ambiguo no es un estándar». Doxa, n.º 28, 2005. Taruffo, Michele. Simplemente la verdad. El juez y la construcción de los hechos. Madrid: Marcial Pons, 2010, pp. 237-240 (the four kinds of maxims of experience: scientific laws, high-probability, common-sense and spurious generalisations). Valenzuela Saldías, Jonatan. «Hacia un estándar de prueba cautelar en materia penal: algunos apuntes para el caso de la prisión preventiva». Política Criminal, vol. 13, n.º 26, 2018, pp. 836-857; and «Sin convicciones sobre el futuro». Discusiones, n.º 28, 2022, pp. 159-177. Mora Sánchez, Jeffry José. «Predictibilidad conductual y proceso penal». Quaestio facti. Revista Internacional sobre Razonamiento Probatorio, n.º 2, 2021, pp. 53-85 (the risk as a «danger of danger»; statutory indicators as indicia and not presumptions; for imprisonment the warrant must be «at least a non-universal generalisation with a high degree of probability»). Manríquez Oyaneder, Jaime Andrés. «Prisión preventiva y error judicial probatorio». Revista de Derecho (Valdivia), vol. 33, n.º 2, 2020, pp. 275-295 (a subjective standard is no standard; demand graded by the impact on liberty).
This content is for information only and does not constitute legal advice on a specific case. The reading of reasonable inference as the best explanation of the data is the author’s interpretive proposal, built on the sources cited, and not a legal rule in force.
About the author
Juan Antonio Matiz Monroy — A graduate of the Universidad de los Andes (2017) with a specialisation in Criminal Law from the Universidad Sergio Arboleda (2021), he built a solid command of oral litigation at the Universidad del Rosario and at the Universidad de Salamanca, Spain. He sharpened his rigour in evidential reasoning with CESJUL and the Universidad Externado de Colombia. In parallel he went deeper into asset forfeiture —a field that requires understanding the State’s logic on property with the same rigour with which liberty is defended—. He is currently a candidate for the Master’s in Evidential Reasoning at the Universitat de Girona (Spain) and the Università degli Studi di Genova (Italy). Professional licence 294.746 (C.S.J.).