The Harm That Changed Hands

I.The First Known Thing — the three questions a court asks before it listens

constitutional law · standing, the doctrine of who may sue

Before a federal court in the United States will decide whether a law was broken, it asks a prior question about the person asking: whether they have standing. The doctrine was set out in its modern form in Lujan v. Defenders of Wildlife in 1992 and has three elements, each of which the plaintiff must establish. There must be an injury in fact — concrete, particularised, actual or imminent rather than conjectural. The injury must be fairly traceable to the conduct complained of. And it must be likely, as opposed to merely speculative, that the injury will be redressed by a favourable decision. A plaintiff who fails any one of the three is not heard, whatever the merits, and the court does not reach the question of whether the defendant did anything wrong.

two causal links, running in opposite directions

The second and third elements are both about causation, and the doctrine is careful to keep them apart. In Allen v. Wright in 1984 the Court put the distinction in one sentence: traceability examines the causal connection between the assertedly unlawful conduct and the alleged injury, while redressability examines the causal connection between the alleged injury and the judicial relief requested. The first link runs backwards in time, from the harm to the act that produced it. The second runs forwards, from the remedy the plaintiff has asked for to the harm as it now stands. Courts have often remarked that the two usually go together, and that a plaintiff who has established one has usually established the other. But they are not the same finding, and the cases in which they come apart are the ones that define the doctrine.

where the second link fails: the injury has passed to other hands

They come apart, characteristically, when the injury is by now sustained by parties who are not before the court. In Warth v. Seldin in 1975, residents of Rochester who could not find affordable housing sued a neighbouring town over its zoning; the Court held that their inability to find housing depended on the independent decisions of builders who were not defendants, and that striking the zoning would not, on the record, produce a house. In Simon v. Eastern Kentucky Welfare Rights Organization in 1976, indigent patients challenged a tax ruling that let charitable hospitals reduce their free care; the Court held it purely speculative whether reversing the ruling would cause any hospital to treat them, because the hospitals were free to decide either way and were not in the room. In both, the plaintiffs had a real injury and a plausible story about where it came from. What they could not show was that acting on the defendant would change it, because the injury was held in place by actors the order would not bind.

three features of the third element that are easy to miss

Three features of redressability matter for what follows. It is assessed against the relief actually requested, not against relief in the abstract: the same plaintiff with the same injury may have standing to seek one remedy and not another, and a complaint that asks for too much can fail where a narrower one would succeed. It carries its own standard of proof — likely, not speculative — which is a claim about the world after the order and is not discharged by any strength of the finding about the world before it. And it is a threshold, decided at the outset from the pleadings, before the expense of trial. A court that reaches the merits and finds the defendant liable, and only then discovers that the order changes nothing for the plaintiff, has done the doctrine's work in the wrong order and at the wrong price.

II.The Second Known Thing — a target with a case for causation

drug discovery · target selection and target validation

The most consequential finding about target selection in the last decade is that human genetic evidence predicts clinical success. Nelson and colleagues showed in 2015 that target–indication pairs with genetic support were about twice as likely to reach approval; King and colleagues confirmed the doubling in 2019; and Minikel and colleagues in 2024, on a larger and cleaner set, put the relative success at 2.6 and located the effect chiefly in phases II and III — the phases in which a drug is asked whether it works — rather than in phase I. The evidence takes several forms: a rare loss-of-function variant that protects, as with PCSK9; a Mendelian disease at the locus; a genome-wide association mapped to a gene; a Mendelian randomisation estimate in which naturally allocated variation in a target's activity is used as an instrument for its effect on disease. Portfolios are now sorted by it, and a target without such support must argue for itself.

what the evidence establishes: that the injury is traceable

What each of these forms establishes is the same thing, and it is worth saying exactly what. A germline variant is present from conception. The finding that carriers have more or less of a disease is a finding that the target's activity, sustained across a lifetime, bears on whether the disease arises. This is a finding about the backward link — from the injury to the conduct — and it is a strong one, because the allocation is random with respect to confounders and because the exposure precedes every stage of the disease. It is, in the court's vocabulary, traceability. It has become the standard of validation because it is the one causal finding the field can make in humans before any drug exists.

what it does not establish, and the field's word for the gap

The field is not unaware that the forward link is a different matter. Ference and colleagues showed in 2012 that lifelong genetically lower LDL cholesterol reduces coronary risk about three times as much per unit as the same reduction achieved by a statin started in adult life, and the standard gloss on Mendelian randomisation is that it proxies the lifelong consequence of perturbing a target, so that effect sizes are not to be read as forecasts of a trial. But the gap is handled as a discount on magnitude, applied uniformly and mentioned in the limitations, not as a second question with its own evidence. Minikel and colleagues wanted to split their success rates by whether a drug was disease-modifying or symptom-managing and reported that they could not find the data for enough mechanisms to try. And the sharpest cases sit in plain view. Mutations in APP and the presenilins cause Alzheimer's disease, and lowering amyloid in patients who already have it slows decline by a fraction and restores nothing. Genetically traced targets in fibrotic, degenerative and autoimmune disease have failed in phase II and III with the causation finding intact, because by the time the patient is enrolled the injury is being sustained by scar, by cell loss, by an autoreactive memory, or by a clonal population that has moved on — by parties, as it were, the drug does not bind.

the forward evidence that exists, and is not collected as such

Human evidence about the forward link does exist, in scattered forms that nobody assembles under one heading. There are revertant mosaics — patients in whom a somatic event has corrected a disease-causing mutation in a patch of tissue late in life, and in whom that patch is or is not healthy. There are diseases whose environmental cause acts on the same node as a candidate target and has been withdrawn late: eradicating Helicobacter pylori reduces subsequent gastric cancer, while stopping smoking slows the decline of lung function without recovering what has gone. There are prior drugs, on the same node, given to patients with established disease. There are genetic effects that are conditional on age, so that a variant's association is confined to onset or extends into progression. Each is a finding about what happens to the injury when the cause is withdrawn after the injury exists. None of them is scored, and a target's file records its genetic support in detail and its late-withdrawal evidence, if at all, as anecdote.

III.The Bridge

no standing case cites a genome; no target paper cites a court

The halls do not meet. Standing belongs to constitutional and administrative law and to a large literature on the separation of powers; target validation belongs to human genetics and translational pharmacology. No review spans them, and the observation that both concern causation is not the bridge, since everything concerns causation. Three neighbours must be set aside. It is not the removability of an earlier specimen, which concerned the redundancy around a node and the difference between a system built without an element and one that has had it removed; that is a question about the target's surroundings, and this is a question about the injury's. It is not the complacency limit, about targets whose variation selection has scrubbed away and which genetics therefore cannot see; here genetics sees perfectly well. And it is not the established caveat that genetic effects are lifelong and trial effects are not, which is a remark about the size of an estimate. What the halls share is a structural rule about claims that ask for an injury to be undone by acting on its cause: such a claim needs two causal links, established separately, and the second is not a consequence of the first.

X: the second link is its own finding, with its own evidence and its own standard

Call it redress separation. Where a claim asks that an injury be remedied by acting on the party that caused it, the claim requires two causal links established separately: one running backwards from the injury to the conduct, the other forwards from the remedy to the injury as it now stands. Four things follow. The forward link is not implied by the backward one, however strong, because it is a claim about a different interval — after the remedy rather than before the injury — and it fails wherever the injury is by now sustained by parties the remedy does not bind. It is assessed against the relief actually requested, so that the same cause may be redressable for one remedy and not for another. It carries its own standard of proof, likely rather than speculative, which no accumulation of backward evidence discharges. And it is a threshold question, cheapest when decided at the outset from what is already known and dearest when discovered at trial. Read into the target hall, each has a consequence the hall does not state. Human genetic support is a traceability finding and no more, and the 2.6-fold advantage it confers is the size of the first link. The trial's primary endpoint is the relief requested, and a target may have standing for a slowing endpoint and lack it for a restoration endpoint, so that redressability is a property of the target–endpoint pair rather than of the target. The evidence that bears on it is human evidence about what happens to the injury when the cause is withdrawn late — the revertant patch, the eradicated organism, the drug already given, the age-conditional variant — and this evidence is of a different kind from the genetic evidence and cannot be replaced by more of it. And because the question is a threshold one, it should be scored before the phase II trial that is at present the first place a target's redressability is tested, at the point where most of the money is spent.

IV.The Question That Becomes Possible

neither discipline would ask this alone

The target hall asks whether a target causes a disease and answers with human genetics, because that is the one causal finding available before a drug exists; when a genetically supported target then fails for efficacy in established disease, the failure is filed under the indication, the biomarker or the dose. The legal hall has no targets, but it settled long ago that a claim to undo a harm needs a second finding, that the second is assessed against the remedy asked for, and that it should be made at the threshold. Neither hall asks whether a target's redressability can be scored in advance from human late-withdrawal evidence, separately from the strength of its genetic support, and whether that score explains which genetically supported targets fail in the phases where genetic support is supposed to help most.

the redress separation hypothesis

Among target–indication pairs with human genetic support, does a redressability score — built only from human evidence that withdrawing the cause after the injury is established alters the injury's course, and scored against the trial's actual primary endpoint rather than against the disease — separate the pairs that succeed in phase II and III from those that fail, independently of the strength of the genetic support itself; and does the excess of failure fall where the injury is by then sustained by processes the target does not govern?

why it would matter

Because genetic support is now the sorting principle of portfolios, and if it measures one of two links, then a portfolio sorted on it alone is systematically long on targets whose injuries have changed hands. If redressability can be scored in advance, some phase II and III trials that would have failed for that reason are not run, or are run against an endpoint the target can move. If a target has standing for slowing but not for restoration, the endpoint should be chosen accordingly rather than discovered. And if the score can be built from evidence that already exists in scattered form, the cost of the second finding is a curation effort rather than a new experiment, which is cheap against the cost of the trials it would inform.

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; it needs a dataset the field has already built and a coding it has not yet done.

the honest status

Note the narrowness. The claim concerns genetically supported targets in indications where the injury is established at enrolment, and says nothing about prevention indications, where the backward link is close to the whole of the matter, nor about targets without genetic support, whose first link has not been made. It does not claim that genetic support is overrated, only that it measures one of two things and is being read as if it measured both. It does not claim that the late-withdrawal evidence is good; most of it is anecdotal, and the score may prove impossible to build with any reliability, which would be a finding about the evidence and not about the principle. It is worth saying what does not transfer. A court decides redressability on pleadings before any fact is found, by a standard of likelihood it never quantifies; a portfolio committee has outcome data and a base rate, and the threshold in the target hall can only ever be a probability estimated from prior cases rather than a bar cleared or not. The whole thing rests on one computation that a single group with the public datasets could make in a season: whether, among targets that genetics has already validated, a score built from what happens when the cause is withdrawn late explains which of them go on to fail. If it does not, this is an elaborate way of saying that causation is not cure, and should be said so. Until that computation is made, this remains exactly what its stamp says: a candidate. A structural echo between two halls of the library, waiting for someone with the tools of both to walk the corridor.