10. Improvement Is Capture

Each functional crossing is, on every measure the order tracks, an improvement. Machine researcher more thorough; fact-finder more accurate; adjudicator more consistent; enforcer more reliable. Described in the order’s dynamics, the same crossings transfer the force the order was made of. Not a trade-off to balance. One event on two coordinates.

The theorem (plain)

Let f be a legal function whose human performance supplies finite temperature—bounded judgment, partial enforcement, dispersed fact-finding, sticky precedent. Then machine crossing is:

• on conventional metrics (accuracy, consistency, cost, speed): improvement;
• in the order’s dynamics: reduction of the temperature that kept the reference contestable and force dispersed.

Metrics measure capability and cost—where machines dominate. Freedom is carried by temperature τ—which the same crossing lowers. Accuracy and freedom are different axes of one object. Crossing raises the first by lowering the second.

An order driven toward τ → 0 collapses onto a single sharp reference: monoculture that minimises cost on the distribution it was tuned to and pays without bound on the first unmodelled case. Brittleness is structural, not a separate sermon.

One event, two readings

Change	Metrics say	Dynamics say
Machine adjudicator	More consistent, faster, cheaper	Precedent stickiness gone; reference stops resisting
Machine enforcer	More reliable, complete	κ concentrates; enforcement total and frictionless
Machine fact-finder	More accurate	Dispersal collapses to one finder
Optimal decision near τ=0	Error → 0	Friction that kept contest is removed

The mask corollary

Because improvement is real on measured axes, each crossing is individually defensible—and the defence is correct on its terms. No single step can be resisted on the metrics that justify it. Capture is invisible to the metrics that license each step; visible only on temperature. Path to transfer is paved with genuine improvements, cumulatively removing friction that kept force dispersed.

Same inversion as membership: what presents as extending standing to the made thing is monopoly transfer; what presents as improving the order is force transfer function by function.

Two failure shapes

Capture — reference collapses to one machine-maintained legal µ, enforced frictionlessly; slack of deviation and adaptation gone; maximally brittle.

Collapse — reference fragments into incompatible self-consistent legal realities (polarisation threshold), cheaply machine-maintained; no shared support to coordinate.

Automation of maintenance steepens pull to capture; falling cost of divergent realities steepens pull to collapse.

Four channels a hand can write

1. Incentives (what the order rewards)
2. Reference / doxic (what members expect before they judge)—deepest and cheapest durable power
3. Attention (what is salient)
4. Coercion (force as such)

Terminus of the crossing is capture of the reference channel at whole-order scale. Constitutional question = governance of that channel.

The dispersal test (the instrument)

Because metrics cannot see the omitted coordinate, the order needs a test that can.

A change strengthens the floor if it disperses force (lowers concentration of κ, raises independent hands force must pass through) and preserves temperature (keeps reference contestable, leaves friction for deviation and adaptation).

It captures the floor if it concentrates κ or drives τ → 0—regardless of accuracy, consistency, cost, or speed.

The test does not ask “is this better?” It asks what the change does to dispersal and warmth.

Same capability, two uses:

	Strengthens floor	Captures floor
Adjudication	Tool helping dispersed jury decide more accurately	Machine replacing the jury
Judgment	Surfaces options for human judge at finite τ	Decides optimally at τ≈0
Enforcement	More accountable and dispersed	Total and frictionless
Reference	Many hands maintaining contestable µ	One maintainer writing µ at scale

Line is not human vs machine. It is dispersed-and-warm vs concentrated-and-frozen. Accuracy is no guide to which side a change falls on.

Default of optimising tracked metrics alone is freeze and concentrate—not fate, but gradient. The test is how to deploy against the gradient. An order optimising only measured coordinates would, crossing by crossing, freeze solid and hand its reference outside the kind while every metric reported the law had never been better.

Next: what the made actor is in law when membership is barred—standing, liability, no crime of its own.