The Parcel Described by Its Neighbours

I.The First Known Thing — the deed that begins at a white oak

property institutions · land demarcation

For most of history a parcel of land was described by walking round it. The deed begins at a marked tree or a stone, runs a bearing and a distance to a creek, follows the creek to the corner of a neighbour's field, and follows that neighbour's line back to the start. This is metes and bounds, and it has one property that is easy to overlook: the parcel has almost no boundary of its own. Its edges are borrowed — from a tree that will die, a creek that will move, and above all from the neighbours' lines, which are themselves borrowed from their neighbours.

the grid of 1785

The Land Ordinance of 1785 did something different with the American public domain. It laid down a principal meridian and a baseline, divided the land into townships six miles square and each township into thirty-six numbered sections, and described every parcel by its position in that grid. A section corner is where the frame says it is. It does not care what the neighbour's deed says, and it does not move when the neighbour re-surveys.

a natural experiment in Ohio

Because the two systems met at a line in Ohio — the Virginia Military District kept metes and bounds while the land around it was gridded — the economists Gary Libecap and Dean Lueck could compare them on similar soil. Under the grid, land was worth more, changed hands more often, and generated far fewer boundary disputes, and the gap had not closed a century and a half later. Their explanation is the one that matters here. A relational boundary is re-litigated every time any adjoining boundary changes, and there is always some adjoining boundary changing. The cost of an unanchored definition is not paid once, at the drawing; it is paid on every neighbour's schedule, indefinitely, by an owner who decided nothing.

II.The Second Known Thing — the disease that is defined by what it is not

medicine · disease definition and indication choice

Walk to a hall with no surveyors in it. The first irreversible decision in a drug programme is the choice of indication: which patients, defined how. A disease definition is a boundary drawn on a population, and medicine draws such boundaries in two ways. Some are anchored to a positive finding — an antibody, a genotype, a tracer that lights up on a scan. Others are drawn by exclusion. Idiopathic, essential, functional, cryptogenic, not better explained by another condition: each of these is a deed that runs to the neighbours' fences and stops.

the parcel that shrank while the trial was running

In 2014 stroke neurologists defined a new indication, embolic stroke of undetermined source, as a non-lacunar infarct with no proximal arterial stenosis and no major cardioembolic cause found. It was drawn by exclusion, deliberately, so that trials could be run. Two large trials followed, testing rivaroxaban and dabigatran against aspirin; both failed to reduce recurrent stroke, and the first was stopped early for bleeding. Meanwhile a neighbouring diagnosis was acquiring a better instrument. Insertable monitors left in place for months found atrial fibrillation in roughly three in ten cryptogenic stroke patients within three years — patients who, under the exclusion rule, had never belonged in the parcel at all. The field's own post-mortem said the population was heterogeneous. It was. It was heterogeneous in exactly the way a metes-and-bounds parcel is uncertain: not because anyone drew it badly, but because its edges belonged to other people.

what medicine knows, and what it does not count

Epidemiologists know this movement well and call it diagnostic substitution. It is treated as a nuisance in incidence series, corrected for after the fact. What is not done is to treat it as a property of a definition, measurable before an indication is chosen — to ask, of a candidate disease, how many neighbours it has and how soon their instruments will improve.

III.The Bridge

no nosology paper cites a land ordinance

The two halls share no literature. Land economists do not read diagnostic criteria, and the committees that write diagnostic criteria do not read the Journal of Political Economy. But the object each is handling is the same object: a region defined as the complement of its neighbours, and the running cost of owning one.

X: demarcation by adjacency inherits the neighbours' revision schedule

Call it demarcation by adjacency. A region whose extent is defined as whatever its neighbours do not claim has no boundary of its own; its edge is a function of theirs, and so its edge moves whenever any of theirs does. The grid, and the positive biomarker, do the same work: they externalise the boundary into a fixed frame, so that disputes are with the frame, once, rather than with each neighbour, forever. Which yields the sharper half of the bridge, the half medicine has never ported: the cost of an unanchored definition is set by the number of its neighbours and the rate at which their instruments improve, and it never amortises. Libecap and Lueck's parcels were still paying in the 1990s for a survey choice made in the 1780s. An indication defined by exclusion is a parcel whose fences are owned by other specialties, and it is resurveyed on their timetable.

IV.The Question That Becomes Possible

neither discipline would ask this alone

If the surveyor's framing is right, then the heterogeneity that sinks trials in exclusion-defined indications is not a biological surprise discovered afterwards. It is a quantity available beforehand, from the definition alone: count the adjoining diagnoses, estimate for each the likelihood that it acquires a positive test within the development window, and weigh each by the share of the candidate population it would resect. Call the sum the indication's neighbour exposure.

the neighbour exposure hypothesis

Does an indication's neighbour exposure — computed from its definition and its neighbours' instruments before a trial is designed — predict which exclusion-defined indications will fail for heterogeneity, and by how much their enrolled populations drift over the life of the programme, beyond what the mere presence or absence of a biomarker in the indication itself predicts?

why it would matter

Because indication choice is made once and cannot be unmade, and today it is made without any estimate of how much of the chosen population will still be inside the boundary when the trial reads out. The same arithmetic reaches endpoint selection: a non-inferiority margin is a boundary measured from a neighbour's fence, and the slow erosion the field calls biocreep is metes-and-bounds drift under another name.

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 names two, and both can be run on records that already exist.

the honest status

Note what is deliberately narrow here. The claim says nothing about the biology of any disease; it concerns only the behaviour of a boundary. The stroke trials are one case, and one case is an anecdote until the programme test is run across many. Until then 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.