15. Residue: what was not built, and what would force it

The figure: the surveyor's margin

Old maps mark the edge of the surveyed country with a line and leave the rest blank, or they fill it with what the cartographer supposed was there, and the two kinds of map are told apart centuries later by which one misled an expedition. This chapter is the blank margin, labelled. Each thing the court does not yet do is written in the only form the programme allows for the unbuilt: not a design — nothing here is designed in anticipation of a brief, which is the rule that withdrew "prepare C3 now" [transcripts/PLAN-20261030.md] — but a claim, with what would force it and a bet. proplang's close called the same thing "named, demand-gated residue that re-enters only through the two-sided gate with a measurement" [data/lineage-sessions.md, proplang], and the chapter keeps that discipline: a residue is admitted to the book when it can be stated so that it could be wrong.

15.1 Primary rules, and the law language

Not built. No rule of conduct exists; the law is the three secondary rules (chapter 4). The rule language — rules as data, each with a violating and a satisfying witness, lapse by notice — is written in the charter and waits [CHARTER §3, Change; A2.1].

What would force it. "The first entry Guy wants refused that recognition does not refuse" [A2.1]. Nothing else may: a rule written in advance of its case is the failure chapter 9 §9.4 records A1 removing.

Likely mechanism. The smallest language that expresses the first case, as A1.3 says; the stratified-Datalog draft on c1-law-draft is design input, not a commitment [LAW.md §3, Change].

Bet. A first case arises, and the first primary rule is admitted with its two witnesses, before C3's exit: p 0.55. If none arises by C2's exit, the thesis that three secondary rules suffice is on trial as A2.1 says, and the book's chapter 4 gains a section rather than losing one.

15.2 Lapse operating

Not built. The lapse rule exists in text and has nothing to act on (§15.1). A law that can only grow has no fixed point (chapter 11 §11.12).

What would force it. The first rule whose violating witnesses the rest of the law already rejects.

Bet. The first rule to lapse does so, undefended, before 2027-06-30: p 0.3 — low because it requires several rules to exist first, and the charter's own expectation is that few will.

15.3 The open roster: when the election becomes a market

Not built. The election runs over a roster frozen at each window's opening [BRIEF-C2a §Scope, the entrants]; C2b's invitation draws among the same principals. Entry is by the sovereign's key declaration (chapter 4 §4.3).

What would force it. A proposer the sovereign has not declared wanting in — a third model family, an outside party, another agent — with a record to show. The moment admission is by record rather than by declaration, the mixture over a fixed roster becomes a market in the ordinary sense: participants with stakes, admitted by performance, paid by score. That is Hayek's price system and Ostrom's graduated membership, both already in the sources, and it is where the book would use the word "market" if it ever does — as an engineering consequence of chapter 7, not as a philosophy of chapter 12.

Likely mechanism. C2b's invite/accept is the seam [BRIEF-C2b, preamble]: a key granted by the clerk on a pre-stated record rather than by the sovereign's hand, under a rule with its witnesses.

Bet. A proposer enters by record, without a sovereign's key entry, before 2027-03-31: p 0.4.

15.4 The agent, and language as action

Not built. C3's agent does not exist; its brief is written from C2's report [CHARTER §8; transcripts/PLAN-20261030.md]. Chapter 10 says what it may not touch.

What would force it. C2b's crossing.

Bets. C3's brief accepted by a cold read within fourteen days of C2b's promotion: p 0.6. The seed agent beats the predictive floor on held-out seeds under the sealed judge within its first window: p 0.35 — the honest number; this is the claim the whole construction exists for, and it is still a forecast. The agent extends its own library and the extension is adopted by the ledger's arithmetic, with no human act, before 2027-06-30: p 0.25.

15.5 Worlds that charge in money

Not built. Two streams enter the value statistic: Hong Kong racing and the data marketplace's catalogue [BRIEF-C2a §Scope, the streams]. Neither pays or costs the court directly; racing could, and the marketplace's revenue could be the outcome rather than its modification times.

What would force it. A world type whose outcome is a priced settlement, saved and hashed like any other — the HKJC settlement page is already on the record [REPORT.md, C1b §3, as transcript: forecasts 80 and 90].

Bet. A forecast type resolves on money the court paid or received, and enters a value window, before 2027-03-31: p 0.45.

15.6 The owner as a plurality — a limit, not an extension

Not built, and not a brief away. The construction assumes one owner: the agent's utility is the owner's, unknown and inferred (chapter 0 §0.1), and the one unscored party is one person with one master key [seq 0]. A team's utility is a committee's, and Arrow's impossibility theorem (Arrow 1951) says no rule aggregates the preferences of several into one coherent ordering without a dictator or restrictions on what may be preferred. "For us" with a plural us is therefore not an extension of the design; it is a different theorem, and the book should say so rather than list it as future work.

What it would take. Either a restriction Arrow allows — a single-peaked domain, or a dictator by rotation, or a market in which owners bid for the court's attention — or a redefinition of "the owner" as a role the construction's own loops fill: the sovereign as an office whose holder is elected by record, which is Ostrom's nested unit one level up and would need its own chapter 4.

Bet. The book's second edition carries a mechanism, with a brief and pins, for more than one owner: p 0.2. That the limit stands as stated — one owner per court, courts nested rather than owners pooled: p 0.6.

15.7 Nested courts

Not built, half-present. jack is judged by assize's sealed judge [CHARTER §0] — one court under another, for one purpose. A court of courts, in which a sovereign's acts are themselves proposals to a higher record, is Ostrom's nested enterprise and the natural answer to §15.6.

What would force it. Two courts with one owner wanting a single ledger; or the open roster of §15.3 admitting an agent that is itself a court.

Bet. A second court on the same kernel, with its own genesis, before 2027-06-30: p 0.35.

15.8 A verifiable draw

Not built. The election's draw is a deterministic signature whose fairness rests on three stated assumptions, the first about the key's custodian (chapter 7 §7.4) [BRIEF-C2b §The draw].

What would force it. An adversary with the custodian's key, or a reviewer's pin that the current draw cannot kill. A verifiable random function would discharge assumption (i); the brief says it does not have one.

Bet. The draw is replaced by a construction whose fairness does not depend on the custodian signing once, before 2027-03-31: p 0.3.

15.9 Agent-run science, procurement by record, deployment gates

Not built; the same machinery, other objects. The record, the pre-stated brief, the sealed judge and the e-process are not specific to code or to racing. A pre-registered experiment is a brief; a result is a resolution with hashed evidence; a replication is a second checker. A vendor's record is a ledger; a procurement is an election. A deployed model's promotion gate is a window with a null taken from its predecessor.

What would force each. A user other than the owner with a world that charges. The book claims nothing here beyond the observation that the mechanisms are general; chapter 16's referee is right that n = 1.

Bet. Someone other than the author runs a court on jack's kernel for a purpose other than building an agent, before 2027-12-31: p 0.25.

15.10 Escalation on the record

Not built. The ladder's first two rungs run as acts of the runner since 3 November, but they are not entries: the record shows the resolutions they produce, not that a rung fired or that it was owed. The third rung, the writ to the sovereign, is a line printed by the clerk (chapter 5 §5.4; chapter 11 §11.14).

What would force it. The case is already on the record: four expiries at 06:00Z on 3 November that cost the liveness process a factor of 4.2 [seqs 517, 521, 623, 625; chapter 6 §6.4], each preceded by hours of a printed escalation nobody acted on.

Likely mechanism. Counsel's draft for C2c names two: tools/jack due, which computes the due list from the record, and an UNATTENDED line that refuses promote while a writ to the sovereign is unanswered [agent-court#18; design input, not a commitment].

Bet. Before C2b's exit, an unanswered writ to the sovereign refuses something — a promote, or an entry of the sovereign's — by the kernel or by status.py: p 0.5.

15.11 A stake chosen by election

Not built. A4.3's liveness stake is valid by derivation and chosen by no theorem (chapter 12 §12.2). The record has priced the choice: at a good record one miss undoes about three successes, and four expiries took the process down by a factor of 4.2.

What would force it. That case, which has already happened; and a promotion it could decide.

Likely mechanism. A mixture over stakes with a regret bound against the best fixed stake, in Cover's universal-portfolio manner, with the betting strategies Waudby-Smith and Ramdas compare for bounded means as the practical candidates — entering as a brief rule with a stop condition, the old stake's wealth printed beside it for one window [counsel's C2c draft, agent-court#20]. With it, counsel's draft proposes a load line — writs per hour against merges per hour — that any brief changing a cadence would have to quote, because the cadence raised on 2 November past the merge path's measured capacity is what produced the expiries [agent-court#20: 6.8 forecasts an hour against 1.8 merges]. Both are candidates, not facts.

Bet. A stake rule other than A4.3's decides a promotion on the log before 2027-03-31: p 0.3.

15.12 What the court will never do

Prove its own soundness (chapter 10 §10.2; chapter 12). Score the owner's wishes (chapter 0 §0.1). Decide merit by inspection (chapter 2 §2.2). Reset its wealth (chapter 6 §6.2). These are not residue; they are the construction, and a future edition that reported any of them built would be reporting a different system.

15.13 What would show this chapter wrong

The bets, scored. The chapter is written to lose some of them: fourteen probabilities between 0.2 and 0.6 should come true about five or six times in fourteen — the probabilities sum to 5.4 — and a second edition in which they all came true would mean the author was miscalibrated, as counsel's early forecasts were in the other direction on its own harness did [REPORT.md §8.4, as transcript]. An item built that is not here — the margin had a town on it. An item here that was designed rather than named — the chapter has broken the one rule it set itself. And §15.6 shown to be an extension after all: a mechanism for several owners that needed no new theorem. The author would be glad to be wrong about that one, and does not expect to be.

Draft note. Draft 1, the author session, 3 November 2026. Every item below is stated in one form: what the court does not do, the case or measurement that would force it, the mechanism it would most likely take, and the author's probability that it exists by a named date. The probabilities are the author's bets, recorded here so that the book's next edition can score them; they are not on jack's ledger, which holds only forecasts by keys the court has declared. Provenance tags as plan.md §2.