The Application That May Be Made Again

I.The First Known Thing — the writ that could be asked for again

law · successive applications for habeas corpus, and the refusal that precludes nothing

Habeas corpus is an unusual remedy in that a refusal of it decides nothing for the future. The ordinary doctrine by which a point once decided is closed, res judicata, was held not to apply to it, and the reason was structural rather than sentimental: the subject matter is a detention continuing now, each application asks whether the restraint is lawful at the moment of asking, and a court that answers no has not held that the detention will be lawful tomorrow. From this followed the old English practice, stated without embarrassment, that an applicant might go from judge to judge. The asymmetry was deliberate and ran the other way as well. In Cox v Hakes (1890) the House of Lords held it had no jurisdiction to reverse an order discharging a prisoner: a grant was final and unappealable, a refusal was not.

what the arrangement produces, once anyone counts

The arrangement has an arithmetic that the courts eventually stated themselves. If a refusal precludes nothing and a grant ends the detention permanently, then the applicant needs one favourable determination at any attempt, and the custodian must win every one. Where judges differ at all in how they read a record — and on marginal records they differ a good deal — the applicant’s chance of eventual release rises with the number of applications, and it rises whether or not anything about the detention has changed. The English judgments put it as the search for a judge perhaps more merciful than his brethren; the point is not the mercy but that the series terminates on the first favourable draw. The quantity governing the outcome is the number of attempts together with the spread among determinations, and neither of those is a fact about the detention.

the remedy the law chose, which is not a better hearing

The response, when it came, was not to improve the determination. It was to restrict re-entry. Re Hastings in 1959 confined the practice, and section 14(2) of the Administration of Justice Act 1960 barred a further application on the same grounds unless fresh evidence is adduced; the courts read fresh narrowly, as material that could not by reasonable diligence have been produced before, and not merely as material not in fact produced. The American line reached the same place by its own road: Salinger v. Loisel (1924) held res judicata inapplicable while allowing weight to be given to a prior refusal, Sanders v. United States (1963) set out when a successive application may be entertained, McCleskey v. Zant (1991) gathered the matter under abuse of the writ with cause and prejudice to be shown by the applicant, and the statute of 1996 made a second or successive petition inadmissible save on a new rule of constitutional law or on facts that could not have been found earlier.

the two devices, stated separately, because they do different work

Two instruments do the work and they are not the same instrument. The first is the fresh-ground condition: a later attempt is admitted only if it rests on something the earlier attempt could not have examined, which is a test on the basis of the attempt and is indifferent to its outcome. The second is disclosure of the series: the applicant must state what has been applied for before and to whom, so that the determination is made with the ordinal of the present attempt on the face of the record. The first restricts repetition. The second does not restrict it at all; it only ensures that nobody reads the successful attempt without knowing which attempt it was. The law needed both, and it is the second that is the cheaper and the more easily forgotten.

II.The Second Known Thing — the marker that may be tested again

drug discovery · biomarkers and patient stratification

A stratifying biomarker admits a patient to a therapy by a threshold, and for several important markers the threshold sits where the assay is least able to discriminate. The clearest instance is HER2-low breast cancer. A tumour staining 1+, or 2+ without gene amplification, qualifies for trastuzumab deruxtecan; a tumour staining 0 does not. The reading is a pathologist’s ordinal judgement on a stained section, and the boundary between 0 and 1+ is the worst-reproduced boundary in the scheme. In a Yale study of 170 biopsies read by eighteen pathologists, published in 2022, scoring in the low range was poor: of the biopsies that at least one pathologist called 0, only about a quarter reached ninety per cent agreement on the 0-versus-1+ call. A national survey found the same boundary to be the one with least agreement.

what happens when the sample is taken more than once

Patients with advanced disease are biopsied repeatedly over years, for reasons that have nothing to do with HER2. In a retrospective series of 512 patients with triple-negative breast cancer reported at ASCO in 2023, the share with at least one HER2-low result rose with the number of biopsies: around three-fifths on a single biopsy, near three-quarters with two, above four-fifths with three or four, and every patient with five or more. The investigators stated the interpretation as an open question — biology, heterogeneity, or analytic variation — and the discussant at the meeting observed that the pattern was consistent with random error in a poorly reproduced assay. The literature has settled on calling the status dynamic.

the rule under which the number was obtained, and the rule under which it is used

Here the two halves of the discipline part company. The trial that established the benefit did not allow the series. DESTINY-Breast04 selected patients prospectively by central testing of a single specimen, the most recently available block, with one antibody on one platform; a historical score of 0 did not disqualify a patient from screening, but a patient entered only if the central test on that one sample returned low. Practice does allow the series. A result of low on any sample, primary or metastatic, taken at any time, is treated as establishing the status, and where a patient scores 0 the standing advice is to consider testing further material. Nothing in that advice is unreasonable taken singly: heterogeneity is real, assays have improved, and a patient with a treatable tumour should not be excluded by one unlucky block.

what the record keeps, and what it does not

The pathology report states the result. It does not state, in any form a later analysis can read, how many times the question had been put before, nor whether the material that answered it differed from the material that had answered it previously. A patient qualified on a fourth block of an old specimen and a patient qualified on a first biopsy of a new lesion are recorded identically, as HER2-low, and both are then counted in the population to which a trial estimate obtained under the single-sample rule is applied.

III.The Bridge

no paper on the successive-application doctrine cites a paper on biomarker retesting; both halls grant a permanent status on a single favourable determination that may be sought again after every refusal, and only one of them has worked out what that costs or what to require instead

The halls do not meet. The successive-application doctrine is criminal procedure and constitutional law, argued in the language of preclusion, finality and abuse; biomarker eligibility is pathology and oncology, argued in the language of concordance, heterogeneity and assay validation. Neither cites the other and no review spans them. Three neighbours must be set aside before the proximity is worth anything. That repeated testing inflates the rate of positives is a statistical commonplace and is not the bridge; the bridge is what the structure predicts about the population so assembled and what the remedy must therefore be. The field’s own objection to the rising figure is about the assay, and its remedy is a better assay — quantitative immunofluorescence, digital reading, mass spectrometry — which is the improvement of the determination that the law tried first and abandoned. And the trial’s central-testing rule is not a fresh-ground condition: it standardises the determination and restricts it to one sample, which is a different instrument from admitting a second attempt only on a new basis.

X: where a status is conferred permanently by a single favourable determination that may be sought again after each refusal, the status is carried by the ordinal of the successful attempt, and the correction is a condition on the basis of later attempts together with disclosure of the series, not a better determination

Call it non-preclusive qualification. Where a status is conferred by a determination that may be sought again without limit after an unfavourable one, and a single favourable determination confers it permanently, the probability of eventual qualification is a function of the number of attempts and of the dispersion among determinations near the threshold, and not of the condition the determination purports to measure; so a population assembled under the repeatable rule differs in composition from a cohort assembled under a single determination, by exactly the share whose qualification arrived on a later attempt, and any quantity estimated in the cohort is attenuated in the population in proportion to that share. Being so, the composition is recoverable from one quantity, the ordinal of the successful attempt, which is a property of the series and not of the subject and which no determination reported alone contains. And the available corrections are structural: a condition on the basis of a later attempt, admitting it only where it rests on material the earlier attempt could not have examined, which distinguishes the genuine change from the re-draw without needing to know which is which in the individual case; and disclosure of the series, which restricts nothing and is what permits the attenuation to be measured rather than merely suspected.

three consequences, not one said three times

Three things follow. The first is that the ordinal is a stratifier, available now and for nothing: if the composition claim holds, benefit among the qualified declines with the attempt at which qualification arrived, and the decline is a prediction about patients already treated and already followed, requiring no new assay and no new trial. The second is that the gap between a trial’s estimate and the effect observed in practice has a component with a known sign and a computable size, namely the repeat-qualified share, which is ordinarily invisible because registries record the qualifying result and not its ordinal; a field that improves its assay and leaves the series undisclosed has removed the dispersion but kept the arithmetic, since a perfectly reproducible assay applied to successive genuinely different samples still terminates on the first favourable draw. The third is that the law’s fresh-ground condition is directly transposable and is not what the discipline is doing: it would admit a later HER2 determination as qualifying where it rests on a new lesion, a specimen taken after an intervening line of therapy, or an assay of a kind not previously applied, and would treat a further block of the same specimen read by the same method as the same application made twice — a rule on provenance, decidable from the pathology record, that neither forbids retesting nor pretends to know the biology.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The lawyer has no reason to ask it. His applicant has no outcome that can be measured against the ordinal of the successful application, because release is the outcome and it is binary and immediate; the question of whether those released on the fourth attempt differ from those released on the first does not arise in his hall, and he has no interest in a dose of anything. The oncologist has no reason either, because his design variable is the assay. His question is whether the reading is right, the dynamism of the status is a biological puzzle to him, and the ordinal of the qualifying test is not a field in any form he collects. Held together, the halls pose a question that is neither’s: whether the attempt at which a patient qualified predicts how well the therapy works, and whether a condition on the provenance of a later attempt would recover the trial’s estimate in practice.

the qualifying-ordinal hypothesis

Concretely. Take patients given trastuzumab deruxtecan for HER2-low advanced breast cancer, in cohorts where the full HER2 testing history is retrievable, and record for each the ordinal of the determination that qualified them, the number of prior determinations of 0, and the provenance of the qualifying sample: a new lesion, a specimen taken after an intervening line of therapy, a different assay or platform, or a further section or block of a specimen already read by the same method. The claim has three parts. First, that progression-free survival and response fall with the qualifying ordinal, after adjustment for line of therapy, disease burden, hormone-receptor status and interval since diagnosis. Second, that the fall is confined to, or much larger in, the group whose qualifying determination fails the fresh-ground condition — another block of the same specimen, same method — and is small or absent where the qualifying sample is a new lesion or follows an intervening therapy. Third, that the aggregate difference between the trial’s hazard ratio and a real-world estimate in the same indication is accounted for, in size, by the repeat-qualified share of the real-world population. The second part is the internal control, and it is what separates the claim from an assertion that sicker patients get biopsied more.

why it would matter

Because the remedy is a field in a report. If the ordinal carries what is claimed, then a line in the pathology record stating what has been asked before, and of what material, converts an unmeasured attenuation into a covariate — for stratification in the next trial of a threshold-defined marker, for the appraisal of real-world evidence against a registration estimate, and for the clinician deciding whether a fourth block is worth sending. It would also give a quantitative reading of a practice presently defended and attacked on intuition alone, and it does not require that retesting be curtailed, which is the proposal’s one point of political realism: the law found that it did not need to forbid the second application, only to ask what it rested on and to write down that it was the second.

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 assay and no new trial. It needs cohorts in which HER2 determinations are retrievable in sequence with their specimens, which exist in institutional pathology archives and in the larger oncology registries, and it needs the outcomes that are already recorded beside them.

the honest status

Note the narrowness. The claim concerns the composition of a population qualified by a repeatable threshold determination, and the attenuation of an estimate transported into it. It says nothing about which assay is right, which is the question the field is actually arguing about, and nothing about whether any individual patient’s status changed. It does not assert that retesting is wrong; a patient qualified on a fourth block may well benefit, and the claim is about a gradient in a population and not about a person. It is stated for one marker in one indication because that is where the testing histories and the outcomes exist together; whether thresholds on other continuous markers behave the same way is a further question and not evidence for this one. And where a marker is reproducible at its threshold, the dispersion is nearly zero, the ordinal carries nothing, and the proposal is empty by construction, which is the correct behaviour for it and not a saving qualification.