1. A Discipline That Never Modelled Its Performers

Law is older than its theory. Before jurisprudence there were courts; before courts, the older fact courts formalise: a community holds—and is held together by—a capacity to compel.

A moral claim no one will enforce is still a moral claim. A legal claim no one will enforce is a dead letter. What distinguishes a legal order from custom or conviction is coupling what is stated to the monopoly of legitimate violence. Every serious theory of law must begin there. This paper builds there.

Two half-foundations

For most of its history, jurisprudence asked: where does a rule get authority? Answers pointed upward (morality above the sovereign), sideways (sovereign command, official practice), or downward (what officials actually do). Those are theories of validity—what makes a norm a law—more than derivations of structure.

The nearest thing to a structural atom was a famous catalogue of positions: eight jural relations in opposite pairs—right/duty, privilege/no-right, power/liability, immunity/disability. A periodic table of law. But it is static. Freeze the order and the positions appear. It does not say where a right comes from, why doctrine holds for a century then collapses in a decade, or why a vigorously enforced right and a never-enforced right are the “same” position in the table. No time. No force. Only positions with the force removed.

A more ambitious tradition built dynamics: a pure system descending from a basic norm that must be posited—“obey the historically first constitution.” Coherent, but impure at the root. One can always ask why that axiom, and the theory has no non-arbitrary answer.

Diagnosis: the catalogue had stationary structure without motion; the pure system wanted motion without a non-arbitrary foundation. The missing object is a force with a direction: one conserved thing—the monopoly of legitimate violence—whose orientations are the stationary positions and whose motion is the dynamics. Its foundation is not posited; it is the collective’s instrument for keeping itself.

The assumption no model wrote down

Every historical legal order rested on a fact so constant theory never bothered to state it: the performers of law—judging, advising, fact-finding, enforcing, maintaining the body of expectation—were human beings of the same bounded kind as the community served. Performer and served were the same sort of thing. Invisible because never varying. Load-bearing because never challenged.

Generative systems are the first technology that takes not only clerical substrate but the inferential core: research, drafting, prediction, increasingly advice and adjudication. The performer is ceasing to be the same kind as the served.

There is already lots of surface doctrine—copyright in training, agents that contract, liability for errors, machine inventors. There is almost no theory of what the technology does to the order itself: adjudication, precedent, enforcement, judgment production, the structure surface questions take for granted. Labour markets at least have the beginnings of a function-by-function displacement story. Legal theory has lacked a comparable object. This paper supplies one.

Why existing theories cannot simply “add AI”

The gap is structural. Accounts keyed to acceptance cannot see capture that arrives with acceptance—the more accurate machine everyone prefers. Accounts that score end-states against fixed goods bless crossings that improve “justice delivery” on their face. Efficiency theories are zero-friction programmes; they supply capture’s rationale. Idealised perfect judges welcome the machine that approximates the ideal. Dispersal intuitions exist in constitutional tradition but were never operationalised against a single performer crossing every branch at once. Each theory is excellent for a world with the performer held fixed. The age varies the hard-coded term.

What the paper claims—and refuses

• Structural: the legal order is configuration of held force; its concepts are directions and operations of one monopoly.
• Dynamical (conditioned, not forecast): as machine capability rises, functions cross human→machine in a fixed order; each crossing can transfer a piece of the monopoly. Whether/when a given function crosses is empirical. What the structure supplies is what crossing means.

The dangerous thesis (read twice)

The danger is not that machine law will be arbitrary, biased, or wrong. Communities survive bad law and contest it. The danger is the opposite: machine law that is better on every measure the order knows how to keep—more consistent, accurate, fast, cheap, enforced—and that each improvement concentrates force and freezes the reference freedom lived in.

A human legal system is slow, partial, inconsistent, appealable, embedded in a profession that argues with itself. Optimisation calls those defects. This paper argues they are also the friction that kept law contestable, the reference warm, and force dispersed enough that the community could take it back.

The transfer will not announce itself as loss. It will arrive as a sequence of genuine improvements, each defensible on its metrics, cumulatively the passing of the order’s force to a maintainer the community no longer holds.

A legal order can survive error, slowness, inconsistency, and contest.
It may not survive perfect administration by a reference it no longer owns.

Next: the one identification that makes the rest possible—law is held force.