9. The law changes under itself
Draft 1, the author session, 3 November 2026 (§9.4 re-cut from the first draft of the smallest-law chapter). Provenance tags as plan.md §2; the record is renavondata/jack at 6ebea1a.
The figure: the council
A religion is a fixed point that knows it will be asked to move. Its canon is closed; its creed is recited; and it has, nonetheless, a procedure for change — a council, called by a stated authority, whose rulings bind from their promulgation and not before, recorded so that a later council can cite them. The procedure is slow on purpose. Heresy is a rule admitted without the procedure; schism is a fork: two records from one parent, each claiming the chain. And doctrines lapse — not by repeal, but by going undefended until no one can say what they guarded. The religions that lasted are the ones whose procedure for change was itself unchangeable except by the procedure. That is this chapter: the compiled law of chapter 8 amending itself, by its own rules, forward only, with the one thing that cannot be undone gated and everything else priced.
9.1 What we have, and what it cannot do
We have a law that compiles from sentences (chapter 8) and binds (chapter 4). It was written before the windows ran, and the windows have shown it wrong: a term where a maximum was meant, a reset where a continuation was required, a language built before any case needed it. A law that cannot change is Rex's seventh failure inverted — constancy to the point of uselessness — and a law that changes by any hand is no law. The amendment path of chapter 4 §4.3 exists for this; what this chapter adds is the principle that says which changes are gated and which are priced, and the record of the path being used.
9.2 Only the irreversible gets a gate — Arrow and Fisher (1974)
Result (theorem of decision under uncertainty). When information arrives over time and an act cannot be undone, the expected value of acting now, computed correctly, is less than the naive expected value by a non-negative quantity — the "quasi-option value", their term (Arrow and Fisher 1974, p. 315). Henry (1974) proved the same effect independently. The converse matters as much, and it is the book's inference rather than theirs: an act that can be undone at no cost carries no option value in waiting, so a gate before it costs delay and buys nothing. Arrow and Fisher price irreversibility rather than forbid it; the hard gate is this court's policy, drawn where the price cannot be bounded.
What it forces. Hard gates stand only before effects that cannot be undone; everything reversible is priced and scored, never blocked. For the operator this fixes his reserved acts exactly: the acts that change who may act or what the law is.
Mechanism. The master key's verbs are the irreversible ones — genesis, key, amend, push, decision [seqs 1, 33] — and the last key holding key cannot be revoked, because losing it is irreversible [CHARTER A2.5; kernel/check.py line 86]. The record itself is append-only, so admission is the one irreversible effect the kernel gates [LAW.md §2.4]. On the platform, branch protection and a ruleset on every repository's default branch forbid force-push and deletion, admins included [REPORT.md §8.1 criterion 4, as transcript; the 23 October ruleset in claude/acts-weekly.md; tools/enforcement.json, row irreversible]. The reversible is priced: refusing a proposal "costs the proposer nothing but resubmission" [CHARTER A2.3]; the counted-code cap is a resource constant raised by one push at a window's opening and "enters no score" [BRIEF-C2a §Scope, the cap]; a window that fails is not patched but followed by a new one with the wealth carried [CHARTER §6; A4.1]. The charter's fourth maxim is this result stated as policy [CHARTER §2.4].
Misapplied if. A hard gate stands before a reversible act, or none before an irreversible one. The predecessor managed both at once: force-push and deletion of every branch but master, and of all tags, were ungated for a week while bookkeeping on a status file blocked merges [CHARTER §1, F5]. On jack the test is the enforcement map's irreversible row and the count of the sovereign's acts: in the clerk's window he entered a push raising the cap, a key declaration, two revocations and the promotion — every one an act of the reserved kind — and no repair [seqs 197, 201, 216, 217, 337; status.py --subject C1b … --acts transcripts/evidence/acts-20261101.csv @6b800e14].
9.3 The path
Amendment is three acts and a wait [CHARTER A2.7; SPEC §10.12–13]. A decision of kind transmute names the amendment — Suber's (1990) move: an entrenched rule is first made mutable, as a separate act. The amend entry follows, carrying the hash of CHARTER.md with the text appended and a height above its own seq; it is in force from that height and never before, and no push is lawful in the wait, so the law changes first and the code waits for it. The kernel records the promulgated charter and the schedule and refuses a reused id or a height not above the seq [kernel/check.py lines 71, 94–95]; the judge refuses charter_mismatch, amend_untransmuted and push_during_amendment [SPEC §10.2, §10.12–13]. Each amendment prices itself in its own text and enters a forecast about whether it will ever bite [CHARTER A3.4, A4.5, A5.3; seqs 60, 121]. The procedure for changing the law is in the trusted base and changes only by the procedure.
9.4 Five cases
A constitution that has never been amended has not been tested. jack's has been amended five times, and the five are the strongest evidence this chapter has, because none was planned, each was forced by something the record had shown, and each went through the procedure the text prescribed for it. They are listed here as cases, with the witness that forced each, what it changed, and the entries.
A1 — C0 is the record (14 October). Before the log existed. The first build of C0 carried the law language — stratified Datalog rules as data, witnesses, lapse, 658 kernel lines [REPORT.md §1, as transcript] — and counsel's review found it premature: no precedent needed a rule, and the charter's own §3 said primary rules accrete from cases. The sovereign cut C0 to the record only, moved the language to C1, and set the kernel's bound at 450 [CHARTER A1.1, A1.3, A1.5]. The amendment is in the charter text that genesis hashes [seq 0]; its witness was a kernel two fifths larger than it needed to be, and its recorded lesson is that "every layer is reviewed by a party that did not write it" [A1.7].
A2 — the opening of C1 (18–19 October). The first amendment entered by the amendment path: decision transmute A2 [seq 3], amend A2 with activate_at 5 [seq 4], the installing push at seq 5. Its witnesses were two cold reads of C1's brief and counsel's research of the same day [CHARTER A2, preamble]. It settled ten things, of which four are structural: the law language enters only when a first case arises, and if none has by C2's exit the thesis that three secondary rules suffice is on trial [A2.1]; ts does not decrease, in the kernel, binding every entry [A2.2]; the sovereign's acts are entries — a change to the trusted base is lawful only when a push names its digest, and veto, promotion, budget, grant and transmutation are decision entries [A2.6]; and amendment is two acts and a wait [A2.7]. Its recorded lesson is the one every later amendment obeys: "questions of what the charter allows are put to Guy as rulings and recorded here", and "whatever counsel and builders are forbidden is what the agent of C3 is forbidden" [A2.10].
A3 — the ledger before the clerk (24–25 October). transmute A3 [seq 28], the installing push [seq 29], amend A3 active from 31 [seq 30]. The witness was an argument rather than a defect — "the clerk is better built against a real obligation than an invented one, and an overdue forecast is the first such obligation the record can hold" — put to the sovereign as options and ruled on [CHARTER A3.1; transcripts/RULING-20261025-ledger-first.md]. It reordered the build, split C1 into two pre-stated windows, put the first maxim into the kernel as self_judged, and, being the third amendment in eleven days, applied §12's guard on procedure growth to itself: the C1a report was to count object entries against procedure entries and stop the layer above 5 : 1 [A3.4]. That count is now a line status.py prints every window.
A4 — §6 corrected (26–27 October). transmute A4 [seq 66], amend A4 active from 68 [seq 67]. The witness was a defect in the charter's own mathematics, found by counsel's literature check: §6 as written restarted the e-process at each new window, "which is k draws at one bound" [CHARTER A4, preamble; transcripts/LITERATURE-20261026.md item 2]. The amendment was cut three times by two cold reads of two model families to what they left standing [transcripts/REVIEW-A4.md, -v2.md, -v3.md], and what remained is the statistics chapter 6 states: the wealth is never reset; p₀ is committed, never estimated from the window it judges; the bet is a function of the prefix with no further constant; comparisons are bets on a bounded difference. The rules the earlier versions tried to add went to a brief, to return "by amendment only after one of them has refused or decided an act". A4 entered a forecast about itself — p 0.5 that it is cited or refuses an act by 26 January 2027 [seq 60] — which is to say the amendment bet on its own necessity and will be scored.
A5 — a window is a maximum (30 October). transmute A5 [seq 136], amend A5 active from 138 [seq 137]. The witness was the record: every pre-stated criterion of C1a held on the prefix of 29 October, and A3's text — "runs fourteen days" — held the window open to 8 November with nothing left for it to measure; the sovereign's budget entry had named fourteen days as a maximum, and the text had made it a term [CHARTER A5, preamble; seq 50]. The amendment made a window end at the earlier of its stated end and the sovereign's promote, which may be entered only once every criterion holds on the current prefix, and made the stated end the resolves_by of the window's promotion forecast [A5.1–2]. It was the fifth amendment in seventeen days, it recorded the procedure ratio at 2.1 : 1 that morning, and it carried a forecast [seq 121]. The next day it was used: decision promote C1a closed that window on day six of fourteen [seq 188], and the day after that decision promote C1b closed the next on day two [seq 337].
9.5 Lapse, and the earth belonging to the living
The charter has a rule the record has not yet used, and it is this chapter's other half. A rule all of whose violating witnesses the rest of the law already rejects is a lapse candidate; it lapses at the end of a notice period unless some proposer enters a counter-witness that it alone rejects [CHARTER §3, Change]. Jefferson's letter to Madison of 6 September 1789 — the earth belongs to the living, and no generation's law binds the next without re-election — is the charter's citation for it, and the charter's gloss is the practical one: "since the test and the defence are both entries, lapse costs Guy nothing". The rule waits for primary rules to exist (chapter 4 §4.6), and the book notes it here because it is the convergence criterion chapter 11 needs: a law that can shed what it no longer needs is one whose growth can stop.
Two patterns across the five cases are worth stating because a referee will ask. The amendments move in one direction: each removes something — a rule language, a term, a reset, a design in anticipation — and the one that adds a kernel rule, A3, adds the first maxim. And each prices itself and bets on itself, so that the constitution's own changes are entries on the ledger it governs; the procedure count that guards against the court becoming its own object is applied to the amendments too.
9.6 What this chapter leaves open
The law can change under itself, and has. The thing that changes it is still a person and his counsel, through a passage with sessions in it. The question that chapter 0 set has one part left: the system changing its own programs — not its law but the free edge — on its own initiative, after we have stopped watching. Everything built so far says what that change may not touch. Chapter 10 builds the lock.
9.7 What would show this chapter wrong
An amend without its transmute, or a push in the wait, on the log: the judge's grounds make either a red main [SPEC §10.12–13], and the author checked the four pairs [seqs 3–4, 28–30, 66–67, 136–137]. A hard gate found before a reversible act — a brief that blocks resubmission, say — or a reversible one found before an irreversible act, which the enforcement map's irreversible row and the sovereign's act count would show. An amendment that does not price itself or bet on itself: the pattern of §9.5 is a claim about all five, and a sixth that breaks it would be a precedent the book must report. And the lapse rule, when it exists, failing to fire on a redundant rule — which would mean the law can grow but not shrink, and chapter 11's convergence claim fails with it.