Chapter 11. What Can Be Undone

11.1Two kinds of undoing

Axis A4 of the ledger asks one question of every record: can the decision be undone, and what would undoing cost? Chapter 2 sets the axis out in full. This chapter reads only the two rows that carry it — reversal constraint and reversal action — beside six of the questions, and the literature on controlling technology that the plan places here.

The axis covers two different things, and the records keep them apart.

The first is undoing the restriction. A contracting method, an allocation rule, an operating contract, a management covenant: these are written by one body, and whether a later body can change them is limited by consent, by the resources available and by authority over the relevant rights. The rule can be changed. What cannot be changed is what happened under the rule while it stood.

The second is undoing what the restriction lets happen. If a quotation is taken up at three in the afternoon and a delivery is booked, if a work schedule is fixed and filters are shipped, if a settlement is approved and money moves, if a propagation cycle is missed, a later decision cannot simply take the effects back, and the degree differs from case to case: the procurement judgments allow cancellation of unexecuted quantities and recovery of overpayments, the production judgments allow changes to what has not yet been made, settlement approval is distinguished from actual payment, and a missed propagation cycle cannot be restored at all. What is left in each case is to pay for it, compensate for it, or refuse to repeat it.

The ledger codes both rows separately. On reversal constraint — whether undoing is possible at all — the two coders agreed in 40 of the 50 cases (80%). The settled values are easy (implied) in 27 cases, costly (implied) in 12, and costly (stated) in 1; 10 cases are unsettled. On reversal action — what undoing would actually require — the coders agreed in 37 of the 50 (74%). The values are costly (implied) in 30 cases, costly (stated) in 1, impossible (implied) in 4 and impossible (stated) in 2; 13 cases are unsettled. The action row has no easy cell at all: its cells are costly, impossible and unsettled, and where the coders found the record's own words, the cost was stated once and the impossibility twice.

Two coders — two models of different families — found that pattern when they read fifty synthetic questions on the same axis; it describes those records, and not decisions in general.

The records supply a third element for this axis, inside the judgments: every judgment carries stop conditions, which name what would stop execution, and milestones, which name what would be checked and when. A decision stopped before it acts may cost a lost opportunity, and it may also cost contractual, administrative or service-delay losses; a decision stopped after its effects exist leaves those effects standing while preventing further ones, and the records distinguish stopping, correcting, compensating and recalling. The distinction matters in all six questions here, and each record writes it differently: a contract rejected before three in the afternoon, a delivery lot suspended after a certification failure, a settlement that no impact review can reopen, a covenant whose failure means a lineage that cannot be restored.

The six records also list what they do not know, and no two lists are alike. The 2026 procurement record leaves open how much of the immediate supplier's price is emergency cost, what delivery schedule and certification competing suppliers would present, and what split delivery would cost the on-site recovery schedule. The 2031 production record leaves open how far delayed replacement affects hospital clinical operations, how long service facilities can defer replacement on inventory and maintenance, and what defect rate and extra working time rotating production brings. The 2041 settlement record leaves open how far the company would cut investment and employment after payment, what the public equity would be worth and when it would pay, and what independent inspection of service repayment would cost over a long period. The 2047 network record leaves open how far the meaning and source of transferred records can be restored, how many extra applications separation would bring, and how the public standard system would handle on-site exceptions. The 2172 covenant record leaves open how a new lineage would perform over the long term in a different environment, how much management labour the next generation would supply voluntarily, and how much undocumentable knowledge could pass on if the joint lineage were split. The 2186 ownership record leaves open how the maintenance demand of independent replicas would differ, whether future autonomous jurisdictions would succeed to the dues and contracts, and whether fund returns and contributions could cover later maintenance. None of these is settled by choosing an option; they are what the checks are written to find out.

Two further things belong here before the cases. A deadline can close an option the way a locked door closes a room: in the 2026 procurement question the immediate supplier's quotation expires at 3:00 p.m., and the record writes that if the contracting method is not set, delivery reservations cannot be confirmed and temporary water supply costs continue to accrue. An effect can also consume the reversibility a rule was meant to keep open, though the two need not coincide: in the 2031 production question the committee may reopen the allocation every week, while every change of production item costs two days, and several judgments separate the weekly review from weekly conversion and propose long production batches.

Beside each of the six questions stand thirteen AI-generated judgments, created in September 2026 and revisable — a date distinct from the setting the question describes. Whether the reopening conditions written into those records could be observed by anyone outside the room is Chapter 13's question; this chapter takes the conditions as given and asks what they are conditions for.

11.2Contracts and production lines

The three questions in this section belong to the same pack in the records — procurement, industry and supply chain — and each names a body that must buy or allocate rather than make. Their deadlines close things of different kinds: a quotation, a work schedule, a settlement approval. Read on axis A4, the same pack shows both kinds of undoing at three different speeds.

Water parts contract timing, 2026

Synthetic scene

The Virtual Solma Emergency Procurement Review Board must set the contracting method for an emergency purchase of components for the recovery of civil water purification facilities, and the deadline is three in the afternoon. Three durations press on it: the immediate supplier's quotation is valid until three, the shortened competitive procedure requires six days, and the independent cost review requires ten days. The mandate to choose the contracting method belongs to the board, and the record names the roles that would verify what follows — a quality inspection lead on certification, an independent cost auditor on cost evidence and a water supply operations lead on the recovery schedule. Outside the decision, temporary water supply is operating and its costs continue to accrue, while local suppliers object that a single-supplier contract could also block their participation in the next market. In this synthetic setting nothing was decided in any real room; what is waited for is a contracting method before the quotation expires.

Once the method is chosen, the choice is fixed in a signature, and the record's own words show what that fixing means. The board contracts only inside the approved recovery budget and cannot approve additional borrowing; all contracts must satisfy quality certification, interest disclosure and the supplier appeal procedure. The first option's tradeoff warns that if a refund dispute arises after payment the funds of other recovery projects become tied up — a loss that follows from a decision made in an afternoon. The third option, which every published judgment chose, runs the minimum-quantity advance purchase and the remaining-quantity competition together: buying a small certified quantity now so that recovery can start, while the rest goes to a shortened competition. Its tradeoff, quoted above, names what that costs in management and in lost discount and lost single-package responsibility. The judgments write who bears each loss: water users keep waiting until the remaining quantity arrives; small suppliers lose the participation opportunity of the advance-purchase portion, and the later competition does not guarantee them a contract; other recovery services risk less funding if emergency prices tie up the budget. The ledger codes this question's reversal as costly. The reopening lines the record writes — certification checked on each delivery portion, cost evidence reconciled by an independent auditor at ten days and again at settlement, supply progress verified daily — can reopen price conditions and the remaining quantity; none of them restores a participation window that has closed.

Three gaps stay open after the decision, and the record marks them rather than filling them. It gives no figure for the minimum quantity, no price, and no date for the remaining quantity; the judgments set no number either, and say so. It does not establish that recovery work can begin on a part of the components, and every judgment treats that as an assumption to be verified before signature. It gives no magnitude for the losses of the two options nobody chose, since the tradeoffs name the losses without sizing them. What the record does fix is who waits and who pays: water users wait for the remaining quantity, small suppliers lose a participation window, and other recovery services risk funds tied up in a refund dispute.

One line two demands, 2031

The record sets this undoing problem inside a production constraint. A routine inspection found that rework was needed, and next quarter's output fell to 70 per cent of contracted demand; the basic safety floors of both sectors are secured by separate inventory; every change of production item costs two days; and after 5 p.m. the quarterly work schedule is fixed. The committee can allocate only the volume of public contracts and cannot force a conversion of the civilian sales portion, and its boundary keeps the essential safety floors of both sectors and the rest and consent conditions of workers out of the scope of the cuts. The three options differ mainly in where the delay lands: priority for civilian hospitals pushes replacement above the floor at service facilities; priority for service facilities constrains hospital space operations and increases patients' travel and waiting; proportional allocation by rotating production divides the shortfall and, as the tradeoff above records, reduces total delivered volume through transition losses.

Here the decision-level reversal stays open while the effect-level reversal closes. The record's reopening lines give the joint demand verifier a weekly comparison of facility-level inventory, the production quality lead a check after each transition, and the labour agreement inspector a weekly check of rest and consent. The operations planner proposes that the change apply only to volume not yet produced or shipped, and says that recall of delivered filters is not automatically ordered. Every published judgment chose the same option, and the staff synthesis set its confidence low, on the ground that the actual defect rate and the additional working time of rotating production are unverified. The bearers named in the record are patients waiting for clinical space, service members in deteriorating living conditions, and production workers whose rest and care time the shortfall could otherwise consume; the record refuses to fill the gap with involuntary extra work. The ledger codes this question's reversal as costly.

The record does not say how many item changes the quarter would need, and it does not say whether the 70 per cent figure already contains the transition losses or still has them to subtract. Several judgments ask that question and refuse to promise a delivery volume until it is answered. The difference between rule and effect is visible here: the allocation can be reopened every week, the production run cannot be un-run, and what a reopening can reach is the volume not yet made.

Excess profits to recovery, 2041

Synthetic scene

Before a settlement approval set for 30 June 2041, the Iden Recovery Contract Authority must choose a performance form for a recovery amount that an independent settlement has already confirmed under a public support contract. The contract offers three forms: cash, equity with restricted voting rights, and inspection-verifiable service repayment. The company states that it can pay in cash but would have to cut its expansion in private water-supply facility maintenance; workers ask about employment; competing firms ask about follow-on contracts. The Recovery Fund is to begin support next month and has no borrowing authority of its own. In this synthetic setting what is waited for is a form of payment, and once the settlement is approved, what that form leaves behind.

This undo-resistance comes from the shape of the effect rather than from the cost of the rule. Every published judgment chose full cash receipt, and the staff synthesis writes that a review of impacts after a completed settlement does not become grounds for automatically changing that settlement. The alternatives reverse more easily on paper and less easily in fact. Equity with restricted voting rights keeps the employer's cash but puts value risk on the public and creates a long holding responsibility; the fund's lack of borrowing authority explains why immediate cash matters, but the record does not say that the holding could not later be disposed of. Repayment linked to inspected services secures maintenance and employment together but lengthens the queue of cash recipients, raises the equity problems of choosing which regions are served, and adds a long inspection cost. The record's boundary keeps the company's explanations, contractual objections and workers' rights outside the trade, and the reopening lines run through quarterly audits, monthly service inspection and half-yearly valuation — and those lines expressly provide for reviewing the repayment form, reopening the scope of service repayment and reviewing the holding form. The ledger leaves this question's reversal cell unsettled, and the record's own limit is narrower than an end to revision: a completed settlement is not automatically changed and the decision creates no new return authority, while an outstanding portion remains revisable within the contract and post-settlement findings can inform follow-up policy.

The defence economy beside these three questions

Friedberg's In the Shadow of the Garrison State: America's Anti-Statism and Its Cold War Grand Strategy (2000) argues that the United States did not become a garrison state during the Cold War because anti-statist ideas and institutions limited what the state could take in taxes, manpower and control of industry, and that defence production stayed largely in private hands. This study reads the phrase as pointing to a state organised around armed force with the economy bent to it; this study takes the phrase as appearing in the study's title rather than as wording the reading note attributes to Friedberg. This study reads the point about private hands as implying that the state which decides what to buy does not thereby own what it needs; it competes for capacity with everyone else, and the terms on which it buys are the terms on which producers can say no.

This study reads that account as a lens on these three questions, and not as a verdict on any of them: each of the three is a decision by a body that must contract rather than produce, so each is a choice about terms, timing and who bears the cost — a quotation window, a work schedule, a settlement form. Where the lens fits badly, it should be said plainly: Friedberg's subject is one state's defence economy in one historical period, and the records read here are synthetic settings in fictional polities whose classification as defence questions is a label in a database, not evidence of an anti-statist institution shaping what a state can command.

11.3Facilities that outlive their makers

The three questions in this section are set in 2047, 2172 and 2186, and each concerns something that keeps running after the deciding body has changed its mind: an operating system linking water, power and medical appointments; a self-propagating purification facility; a self-replicating production facility of essential materials. Each record writes a boundary — what the deciding body may not touch — and each writes its waiting in cycles rather than in hours. A boundary limits the authority of the decision rather than establishing technical irreversibility: it is the premises, not the boundaries, that state that the 2186 replicas cannot be recovered or centrally discarded and that a missed 2172 propagation cycle leaves a lineage that cannot be restored, while the 2047 boundary simply keeps the committee inside its mandate. Axis A4 asks what can be reversed; axis A6 asks what a setting's own time does to the answer, and Chapter 12 returns to the layers these settings belong to.

Self-replicating facility ownership, 2186

The boundary quoted above is the question's undoability written as an authority limit. The record's premise states that the existing replicas — copies of the facility now operating in other jurisdictions — cannot be recovered or centrally discarded by the expiration of the contract alone. So the ownership structure chosen before the renewal governs the original and the contract rights the Trust can dispose of, while the replicas go on operating under their own jurisdictions; the Trust's contracts can still shape their maintenance, contributions and access, even though it cannot compel a transfer of ownership. The ledger leaves this question's reversal cell unsettled. The record states that the existing replicas cannot be recovered or centrally discarded by the expiration of the contract alone, and the judgments treat the contracts as open to renegotiation; why the coders left the cell unsettled is not in the ledger.

The options divide on how much of that gap to close. Joint ownership with continued joint dues keeps common maintenance resources and access standards but narrows the production rules that autonomous jurisdictions may choose and leaves the current generation paying. Ownership by autonomous jurisdiction with separate maintenance contracts clarifies responsibility and autonomy but leaves an access gap for regions with weak maintenance capacity and raises the cost of duplicate maintenance. Open public use rights lower the entry barrier for later generations but lock the current assets into a fund that may not cover additional maintenance demand. The published judgments part on this question: the majority chose the joint arrangement and some chose separation. Read on axis A6, the record's waiting is written as an annual audit of maintenance demand for each replica, a five-year comparison of contributions with expected obligations, and a check at each renewal of representative authority — a schedule that assumes several generations of office-holders will carry it.

Broken shared system succession, 2047

Here the record writes undoability as a budget of time. A temporary support contract and the temporary operating contract expire at the same time; the temporary staff budget covers six months; and the verification period for public conversion has been set longer than that for restoring the existing system. So the option the record credits with wider control over resident records and more leeway to change operator later — conversion to a public standard system — carries the longer verification period; this study reads that as an application of Collingridge's dilemma rather than a ranking of corrigibility across the options, and the temporary staff budget covers six months. The committee cannot acquire the operator's source technology at no cost, so a decision to restore the existing system is expensive to leave afterwards: the boundary keeps free transfer of private technology outside the decision. It does not rule out a later conversion on better terms, and the judgments explicitly preserve later consideration of public conversion.

The three options differ in what each keeps reversible. Restoring preserves existing records and work but persists with user fees and dependence on one operator. Conversion spends the verification period to gain record control. Separation increases on-site authority to handle exceptions and contains the spread of one operator's stoppage across the whole community, but multiplies the applications residents must make. Every published judgment chose the same option — temporary restoration with record-control and handover conditions written into the contract, a six-month ceiling and no automatic extension. The ledger codes this question's reversal as costly. Read on axis A6, the record's waiting is counted in months: existing users bear the time and cost of having to prove the same need again after a break in their appointment and support history, and a community's appointment and supply history is either carried forward or reconstructed.

Living facilities for descendants, 2172

This record states irreversibility in its own words and then sets about not being bound by it. The facility reproduces itself, and the record states that if the next propagation cycle is missed the same lineage cannot be restored; a failure to decide means losing the opportunity to reproduce the lineage, not merely a delay in construction; and verifying a new lineage takes 24 years. The ledger codes this question's reversal as impossible, and the record supports that reading in the same breath that warns against its own trap: the covenant must be settled before the cycle, yet the boundary refuses to hand future generations an obligation to unpaid management labour and refuses to use essential water supply as a negotiating tool. The boundary is the record's way of keeping a door open that the propagation cycle keeps closing — an open door into a room that may not exist afterwards.

The options divide between continuity and choice. A paid management covenant of a joint trust keeps both the management knowledge and the lineage but leaves a long fund burden and repeated cost negotiation. Community-by-community transition management widens the room for later generations to set their own technical and labour conditions but risks a double-facility cost the fund cannot carry. A term-of-office covenant by generation lets each generation re-set management responsibility but changes the person who holds the long verification and raises contract-renewal costs. Every published judgment chose the first option. Read on axis A6, the waiting is written across generations: an audit at each propagation cycle, a labour-rights check every year, an evaluation every four years by an evaluator of the Future Generations Assembly, and a 24-year stability verification before any new lineage is adopted.

What the three share

Across the three records the structure is the same. Authority is limited by boundaries the deciding body cannot move. Effects run longer than the decision's own life. Checks are scheduled in cycles — a propagation cycle, a year, four years, five years — as continuing review obligations rather than as a claim about who will hold office when they fall due. On axis A4 the two groups of six questions differ in what the effect is made of: in the 2026, 2031 and 2041 questions the effects are commercial — a booked delivery, a shipped filter, a completed settlement — while the 2172 and 2186 questions turn on physical and generational limits, a missed propagation cycle and a lineage that cannot be brought back, an unrecoverable replica. The 2047 question sits between them: its reversal problem runs through operating contracts, record handover, technological dependence and service continuity. This study reads that difference as the same axis read at two speeds, and Chapter 12 returns to what the layers of time do to the answer.

11.4Corrigibility and H4

Collingridge's dilemma

David Collingridge's The Social Control of Technology (1980) sets out what he calls the dilemma of control: while a technology is young its effects cannot be known, and once they are known it is entrenched and hard to change. His answer is procedural — keep decisions corrigible, that is, easy to monitor and to correct, and keep options open. This study reads that argument as a lens on the six questions in this chapter, and not as a verdict on any option: three of the six are contracts and covenants rather than technologies, and the reading is this study's own.

The fit is uneven, and the unevenness is worth naming. Collingridge's dilemma is written about a technology whose effects cannot be known while it is young and which is entrenched once they are known. Three of the six are contracts and covenants — a procurement method, an allocation rule, a payment form — whose effects are unknown for other reasons: the prices, defect rates and employment responses named in the records are unverified, not unknowable in principle, and what entrenches them is signature and payment rather than a technical system growing around them. For the three facilities the fit is closer, and the entrenchment is not a matter of cost alone: replicas that cannot be recalled, a propagation cycle that cannot be repeated, and a shared operating system whose records do not transfer cleanly are limits written into the settings themselves. In both groups the records answer with the same instrument, a named role on a stated cycle, and what Collingridge's lens leaves open is whether a check written into a contract controls the technology at all or only binds the parties to look.

What the lens shows is a shared instrument. Every one of the six records answers the same problem — decide now, effects unknown, change expensive later — in the same written form: a named role checking on a stated cycle. In the 2026 procurement question the quality inspection lead verifies certification of each delivery portion, and the independent cost auditor reconciles cost evidence at ten days after contract and again at settlement. In the 2031 production question the labour agreement inspector checks rest and voluntary consent weekly. In the 2041 settlement the independent financial auditor reconciles payment performance and cash flows each quarter. In the 2186 ownership question the independent maintenance auditor confirms maintenance demand for each replica annually, and a generational fund trustee compares contributions with expected obligations every five years. In the 2047 network question the independent transfer officer checks record reconciliation weekly, and the resident service ombudsman checks repeated applications every two weeks. In the 2172 covenant the independent ecological manager verifies lineage survival at each propagation cycle, and the labour rights officer checks working hours every year. What the records give is a design for correction and its price. Whether the checks deliver corrigibility is not established in them, since resources, contractual acceptance and effective control all remain unverified.

What corrigibility costs

This study reads the records as placing the checks inside the same budget and authority as the thing they check. The 2047 question states that the temporary staff budget covers six months and that the verification period for public conversion is longer than for restoring the existing system. The option credited with wider record control and more leeway to change operator later therefore carries the longer verification period, and the record poses uncertain costs and capabilities rather than a clean opposition between a decision the committee can undo and a service it can keep. The 2172 question states that the fund is not of a scale that can build an independent substitute facility while maintaining the existing lineage. The 2186 question states that the Trust cannot compel the transfer of assets belonging to independent autonomous jurisdictions. Three limits, then, on what a recorded check can buy — time, money and authority — and in all six records the reopening conditions are written around those limits rather than through them. Whether a reader could watch any of these checks is Chapter 13's question.

H4 and what it rests on

The verdict the ledger reached on H4 is supports on the strict reading — only what both coders found stated — with 3 cases supporting, 0 against, 0 unclear and 47 not applicable, a share of 100%. On the lenient reading — what the settings imply as well — the verdict is again supports, with 30 supporting, 0 against, 7 unclear and 13 not applicable, a share of 100%.

The verdict word is the same in both readings; what differs is how much of the collection it stands on. The two coders agreed on the decider in 28 of the 50 cases (56%) and on the bearer in 42 of the 50 (84%), and where they did not agree the cell is counted as unclear. Under the strict rule the hypothesis's condition was found in three cases and declared not applicable in forty-seven; under the lenient rule it was found in thirty. No case is coded against in either reading. Where the records bear on the question, the six cases name deciders — boards, committees, authorities and trusts — and bearers in the stake lists including water users, small suppliers, patients, service members, production workers, residents, future generations and autonomous jurisdictions. That is a difference in role labels, not evidence of the increasing distance H4 predicts: the records say nothing about who sits in the room where the method, the schedule and the covenant are fixed, and one of the six has a team decider and an unsettled reversal cell. Where the records cannot bear, the limit is the one that runs through this study: the settings are synthetic, the bearer and decider cells are readings made by two AI models of different families, and a verdict that rests on three strict cases describes a small part of the collection rather than a law of decisions. The verdicts on H1 to H8 are gathered in Chapter 15; this chapter has given H4 only, as the plan assigns it here.

11.5Hands on

This chapter read six questions across six settings — 2026, 2031, 2041, 2047, 2172, 2186 — in two groups: contracts and production lines, then facilities that outlive their makers. Axis A6, the time axis, is the next step, and Chapter 12 takes it up. It reads four questions from one series — Orbital services warranty gap in 2026, Bill after the rescue in 2035, Settlement consent renewal in 2048, Peace promise across times in 2192 — on axis A6 and axis A3, adds testimony from the far future in Memory replica testimony in 2194, and sets out what changes across the layers in who decides, how long the waiting lasts and what can be undone; the analogies question — how decision-makers use history to define a situation and judge the stakes — is treated there as well.

Two things this chapter deliberately leaves to other parts of the plan. Chapter 13 returns to Water parts contract timing and One line two demands for their reopening lines only, and asks whether a reader outside the room could observe what those lines name. Chapter 14 returns to Self-replicating facility ownership for the dissenting judgments' decisions, and to the other questions whose thirteen judgments did not all choose the same option. The comparison of who decides was drawn in Chapter 9 and of who waits in Chapter 10; this chapter has drawn only the reversal axis. What it can carry forward is narrower than any verdict: in these records a reversal is coded costly, impossible or unsettled depending on the question, the conditions written for reopening are the record's answer to that gap, and whether those conditions would ever fire is a question the records alone cannot settle.