The Finding That Cannot Cross

I.The First Known Thing — what may be surrendered

law · extradition, dual criminality and the rule of specialty

A state asked to hand over a person it holds does not retry the case. It asks a narrower question, and the oldest answer to it is a list. Article 27 of the Jay Treaty, concluded in 1794, made two offences extraditable between Britain and the United States: murder and forgery. For a century and a half treaties grew by enumeration, each new instrument carrying a longer schedule of named crimes, and conduct outside the schedule was not surrendered however grave it was.

the list gives way to a test

Modern treaties abandoned the list for a test. An offence is extraditable if it is punishable in both states by deprivation of liberty above a threshold of seriousness, commonly more than a year, so that all felonies are in principle covered. The test is of conduct and not of nomenclature: it is enough that the particular act charged is criminal in both places, and the two states need not call it by the same name, class it in the same part of their codes, or divide it into the same number of offences.

the gap that is not a defence

But the test can fail on the law rather than on the facts. Where the requested state has no offence that the conduct answers to, there is nothing for the dual criminality test to find, and the request fails. The failure is not an acquittal and not a finding of innocence; the conduct has not been examined at all. The United States met this for years with its own inventions — racketeering, continuing criminal enterprise, and in some partners the inchoate offences of attempt and conspiracy, which several legal systems do not punish as the Americans do — and with the fiscal offence exception, under which tax and customs crimes were long excluded by treaties that otherwise reached everything serious. Each gap was closed, where it was closed, by writing a provision for it, which is to say by extending one state's taxonomy until it overlapped the other's.

and a restriction on what crosses

What does cross arrives under a restriction. By the rule of specialty, traced in United States law to the Supreme Court's decision in Rauscher in 1886, a person surrendered may be tried only for the offence for which surrender was granted, and not for other offences the receiving state would otherwise have charged, until a reasonable opportunity to leave has passed. The rule does not bar further counts of the same offence, nor crimes committed after the surrender. Its effect is that the category under which a person was transferred governs what may afterwards be done with him, so that the receiving state's attention is bounded by the requesting state's original characterisation and not by everything it later learns.

what the records can and cannot show

One consequence is seldom remarked. Because a request must characterise the conduct before it is made, conduct that answers to no category in the other state's law is ordinarily never requested. It does not appear as a refusal; it appears nowhere. Any study of how often extradition succeeds is therefore computed over requests, and requests are drawn from the intersection of the two codes. The rate tells one how good the overlap is at delivering what falls inside it, and nothing at all about how much lies outside.

II.The Second Known Thing — the finding and the word for it

drug discovery · toxicity and safety pharmacology

Before a compound is given to a person, it is given to animals, and what is seen there is written down in the vocabulary of the pathologist: a histopathological change at a named site, graded, with its incidence and the exposure at which it appeared. The purpose of the record is to decide what shall be watched for in the clinic. The watching is written in a different vocabulary, the Medical Dictionary for Regulatory Activities, which is organised around what a patient reports and a physician observes and is the language in which every clinical adverse event, every protocol's safety monitoring and every post-marketing report is coded.

the measured width of the overlap

The two vocabularies were built for different purposes, and the width of their overlap has now been measured. The eTRANSAFE consortium's mapping, published in 2025 and released as an open tool, routed three preclinical vocabularies — a histopathology terminology, a mouse anatomy ontology, and the code lists of the standard for exchanging nonclinical data — to clinical terms through an intermediate medical ontology. Thirty-four thousand preclinical concepts were mapped to that intermediary. But of the preclinical concepts themselves, between 36 and 64 per cent had no mapping to a clinical term at all, and in the other direction between 62 and 89 per cent of clinical terms had no preclinical counterpart. The reasons the authors give are not failures of effort. Preclinical work records morphology; clinical work records symptoms. Cellular changes have no clinical equivalent to be matched to. Headache, dizziness and back pain cannot be mapped the other way because an animal cannot report them.

how the field measures its own translation

The field's measure of whether animal studies predict human harm is concordance, and the two canonical studies share a feature. Olson and colleagues, in 2000, took 150 compounds with 221 human toxicities identified during clinical development and asked whether the same target organ had been hit in animals, reporting a true positive concordance of 71 per cent for rodent and non-rodent together, 63 per cent for non-rodents and 43 per cent for rodents alone, high for haematological and gastrointestinal findings and low for cutaneous and hypersensitivity reactions. The later analysis of 3,290 approved products and more than 1.6 million reported events treated each animal event as a diagnostic test for the corresponding human event, computing sensitivity, specificity and predictive values by organ class, with central nervous, musculoskeletal and respiratory observations translating poorly. To make the comparison, the animal events were coded into the clinical dictionary.

what is therefore not in the figure

Both designs require a pair. Olson's starts from a human toxicity and looks back for its animal counterpart; the larger one needs each animal event expressed as a clinical term before it can serve as a test for one. An animal finding whose category has no clinical term is in neither study's denominator. It is not scored as a false positive, nor as a missed prediction, nor as a species-specific finding properly discharged. It is absent, along with a third to two thirds of what the preclinical vocabulary can say. The mapping paper reports the unmapped residue as a limitation of curation and states that it does not examine whether unmapped categories relate to human harms that were missed.

III.The Bridge

no paper on extradition cites a paper on translational safety; each hall moves findings between two systems that name things differently, and neither has read the other's rule for it

The halls do not meet. The literature of extradition is public international and criminal law; the literature of preclinical to clinical translation is regulatory toxicology and biomedical informatics. Neither cites the other and no review spans them; the only places the two vocabularies co-occur concern the extradition of traffickers, which is a collision of the word drug and not of any idea. Four neighbours must be set aside. This is not the concordance question, which asks how well the pairs agree and is computed, necessarily, over pairs. It is not the missing biomarker problem, which concerns whether a category that exists in both vocabularies can be measured in a living patient; the deficit here is that the category does not exist in one of them, and the two overlap in a way the test below must separate. It is not the long argument about whether animal models predict human biology, which is about the organism and not the dictionary. And it is not the known unreliability of clinical coding, which concerns two coders disagreeing inside one taxonomy rather than two taxonomies failing to meet.

X: a finding crosses only on the intersection of two taxonomies, and crosses bearing a restriction to the category it crossed under

Call it surrenderable reach. Where a finding made in one system must be acted on in another, and the two systems built their categories for their own purposes, the transfer is confined to the intersection of those categories, and whatever crosses is licensed only for the category under which it crossed. Three things follow, and they are not the same thing said three ways. The first is that the quantity which governs the transfer is the unsurrenderable fraction — the share of findings whose category has no counterpart — and it is invisible to every measure of agreement, because agreement can only be computed on what crossed; a concordance figure is conditioned on surrenderability and reports the quality of the intersection, never the completeness of the transfer, so improving it and widening it are different projects and the first can be pursued indefinitely without touching the second. The second is that a finding which fails the crossing is not refuted but unaddressed: the silence is categorial rather than empirical, it is produced before any evidence is weighed, and it is indistinguishable, in the record, from a finding that was never made. The third is specialty: because what crossed is licensed for its own category alone, the watching afterwards is as narrow as the term that carried it rather than as wide as the mechanism behind it, so a hazard adjacent in biology but distant in vocabulary goes unwatched even where the crossing succeeded.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The lawyer has no reason to ask it. Dual criminality is studied as a safeguard for the person surrendered, the unrequested residue is a matter for treaty negotiation rather than for measurement, and nobody computes the fraction of foreign conduct that answers to no domestic offence because no state has an interest in the figure. The toxicologist has no reason to ask it either. The unmapped residue is a curation backlog, to be reduced by better ontologies, and the measure of translation is concordance, which cannot see it. Held together they give a quantity that is already computable from documents in the public domain, and a prediction about where human harm appears late.

the surrenderable reach hypothesis

For a cohort of approved products, is the unsurrenderable fraction of each one's nonclinical safety package — the share of its reported findings whose category has no counterpart in the clinical vocabulary the trials were monitored in — a predictor of later safety action in categories that were never monitored, independent of the severity, the incidence and the organ distribution of those findings; and where the crossing did succeed, is the monitoring that followed drawn as narrowly as the index term rather than as widely as the mechanism, so that post-approval signals fall more often in categories adjacent in biology to a crossed finding than in the crossed category itself?

why it would matter

Because it would give a safety package a second number beside its margins: not how severe its findings were but how much of what it found could be carried into the clinic at all, computed before the first human dose and comparable across programmes. Because it would tell the difference, which no present measure makes, between a hazard that was looked for and not found and a hazard for which no question was available to ask, and the two warrant different conduct — the first a margin, the second a purpose-built instrument or an honest declaration that the organ is unwatched. And because it would say whether the unmapped residue in the vocabularies is an administrative inconvenience, as it is now treated, or a register of where human harm will be discovered last.

V.How It Could Be Proven — or Broken

a candidate must be breakable, or it is merely poetry

An intercognix candidate earns nothing until it names its own test. This one needs no new experiment and no proprietary data. The nonclinical findings are in the pharmacology and toxicology reviews published with approval packages, the monitoring is in the protocols and the clinical review, the mapping tool is open source, and the later safety actions are in the label history.

the honest status

Note the narrowness. The claim is about a vocabulary and not about an organism. It does not say that animal studies predict human harm badly, nor well; it says that the published figures for how well they do are conditioned on a crossing whose failures are not counted, and it offers a number for those failures. It says nothing about compounds that never reached man, nothing about efficacy translation, and nothing about which of the two vocabularies ought to be changed. It does not assert that any particular unmonitored organ harbours harm. The whole of the claim is that the share of a safety package that cannot be expressed in the language the clinic watches in is a quantity, that it is computable today for every approved product, and that it is at present reported only as a footnote about curation.