Chapter 2 named exit as one pole of a load-bearing axis: the check on any polity is that its people can leave it. But exit as it exists today is brutally expensive — sell the house, quit the job, pull the kids from school, move. The price filters out almost everyone, which means bad local governance faces almost no selection pressure from departure. This design cheapens exit at the margin where it's cheapest to honor: the border.
Households at a boundary may switch sides. If the polity next door governs in a way you prefer, you can re-register your household into it — no moving truck required. Borders stop being walls and start being membranes: polities that govern well accrete edge-households, polities that govern badly bleed them, and the map itself becomes a slow, continuous referendum. This is the selection operator of chapter 2's search — the step that decides which variants survive — acting on jurisdictions.
All of that presumes a thing with an edge — and this proposal has leaned on the word neighborhood without ever saying what one is. The research underneath it is more specific: the base unit is a cell of fifty to a hundred and fifty people, sized against Dunbar's number — the rough ceiling on how many relationships one person can actually keep track of. Under that ceiling a group can hold itself to account socially, because everyone knows everyone; over it, accountability has to be manufactured out of procedure. The cell is what the lottery draws from and what elects into the tier above, and it is small enough that whoever carries your weight upward is someone you can stop in the street.
What a cell is allowed to form on matters more than how big it is. Geography is one option and not the only one: cells can also form on shared work, shared interest, shared identity, or random assignment with an opt-out — and one person belongs to several at once. That is what makes the map concentric and overlapping rather than a partition of the ground into exclusive plots. Elinor Ostrom's polycentricity — many centers of authority working the same territory at different scales, rather than one hierarchy owning all of it — is the studied version of this shape, and it beats the alternative wherever the alternative is silos.
But not everything detaches from place. A water main, a park, a road serve a location and cannot be re-sorted by affinity. So the honest picture is two layers running at once: place-bound portfolios stay territorial and keep real borders, while everything else runs on overlapping membership. Most of what is genuinely hard in this chapter lives where those two layers touch.
A size cap with no rule for what happens at the cap is decoration. The mechanics here are a cell's, taken literally rather than as a figure of speech.
Too big, and it splits. When a unit takes on enough sub-units to cross its ceiling, division is compulsory rather than optional — the whole point of a ceiling is that nothing grows its way out of accountability. The split runs as a draft: the unit names two conveners, and they take turns claiming sub-units until none are left. A draft is chosen over a line on a map for the same reason two children divide a cake by having one cut and the other choose — each drafter is picking the half they will then have to govern, so neither can stack their own side without inheriting whatever they left behind. It also still works where there is no line to draw, which is the affinity case above.
Too small, and it dissolves. A unit that falls under viability — too few members to fill its own lottery, staff its own offices, or carry its own portfolio — does not limp on as a rotten borough, a seat with almost no voters that someone nonetheless controls. It is dissolved, and its members drafted by the neighboring units in the same alternating way. This is apoptosis — the programmed cell death that stops a body carrying tissue which has quit working — and it is the instinct chapter 8 already applies to policy: expiry is the default, and persistence is the thing that has to be earned. Chapter 9 holds the involuntary cousin of the same move, where neighbors absorb a local regime on proven spillover.
Two bounds keep the geometry honest. A unit has a maximum radius as well as a maximum headcount, scaled to its tier — a base cell whose members cannot physically convene is not a cell — and members keep the ordinary right to change units between cycles, which is the household border-switch from the top of this chapter, generalized past the border.
Endogenous borders — lines the system redraws itself rather than inheriting — raise the obvious specter: gerrymandering, the fine art of drawing lines so the rulers choose the ruled. The first answer is that most of the time nothing is drawn at all: jurisdiction follows engagement. A proposal starts at the smallest scale that could own it, and if people three neighborhoods over start spending points on it, that spread is itself the evidence that the issue is wider than its host, and it is offered upward; a policy that holds its support in some counties and loses it in others demotes back to the counties that still want it. Three tests decide where something belongs, and they are published in advance: can this scale perceive the problem, can it act on it, and are the effects contained inside it? Where a line genuinely does have to move, tax-flow apportionment — who gets the revenue a redrawn line moves — and cross-boundary enmeshment, how tangled two communities already are across it, are explicit, rarely-pulled levers rather than smoke-filled-room variables.
Wherever an optimizer is used at all, algorithmic must never be mistaken for neutral: "the algorithm decides" always means "whoever specifies the objective function decides." Compactness, contiguity, population balance — each choice of objective is a political choice wearing a lab coat. So the objective function itself lives in the meta-rules layer (chapter 12), owned by the broadest tier, on the slowest review cadence in the system.
Which exposes the question this design owes an answer to: where do the goals come from at all? There is no second machine that computes what a society should want. The goal surface is the question set of chapter 3 — written by citizens, continuously re-solicited, weighted by the engagement each question earns — and it is deliberately not a formula, because any named metric becomes a target the moment it binds anything. What the boundary machinery needs from that surface is two specific competencies, worth stating as requirements rather than assumed into existence. It has to be able to read fairness: the felt questions must be sensitive enough to catch a redrawn map buying one group's comfort with another group's sense of exclusion, which is precisely the failure this chapter closes on. And it has to be able to speak to jurisdiction: the same instrument has to report whether an issue is being handled at a scale the people under it accept, because that reading is what the routing rules consume.
Two honest warnings, kept attached to this mechanism permanently. Successful polities will grow — that is the point — but dynamic sorting is also a homogenization engine, and the empirical record says sorting sharpens segregation, not just policy fit. People do not only sort on governance quality; they sort on who their neighbors are. A membrane-bordered map must be watched for exactly this failure, and the watching belongs to the perception layer: if sorted-in comfort is purchased with society-wide sensed unfairness, the signal (chapter 3) is designed to say so. Second, cheap exit can quietly substitute for voice — why fight to fix your polity when you can hop its border? The design's answer is to keep both levers priced honestly: exit cheap enough to discipline governments, engagement cheap enough (points, delegation, open agenda-setting) that fixing remains a live alternative to fleeing.