What review secures is the common conditions of dispute, not the end of dispute. After the review structure of the Qingya River Valley operated in full for the first time, the four-day date discrepancy was revised — yet the real disagreement had not moved an inch: only three months remained before the flood season, there was only one allotment of stone and transport capacity, the committee faction insisted on committing everything to reinforcing the embankment, while the questioning faction insisted on first paying requisition compensation according to the base. The standstill criterion of Chapter Twenty-One takes effect here: the facts cited by the two sides are largely the same and the dispute continues anyway; the problem lies not in the materials but in the ordering of values — the base gives no answer, and review should not pretend to give one. What remains is not further verification but another task: under the condition that the disagreement does not disappear, to reach a time-limited agreement on the concrete terms of joint action.
This chapter treats limited agreements amid value conflict; it is not a manual of negotiating technique — what it analyzes is a structure that holds for both sides at once. The questions, in order: why the object of an agreement must be terms of action rather than value consensus; why duration and exit clauses are constituent parts rather than appendices; why the moral basis of an agreement is temporary joint action rather than the discovery that "deep down we are the same kind of people." It also confronts two failure modes directly: the agreement used by the stronger side as an instrument of delay, substituting negotiation for change; and the breakdown of an agreement blamed on the side "unwilling to compromise," whereby structural disagreement is moralized into a defect of character.
The Object and Conditions of Limited Agreements
Review confirmed the ledgers, and it also confirmed the hardness of the dispute. Both sides cite the same base and the disagreement stands where it stood: the committee says that without the embankment there is no river valley, and the accounts can be settled slowly after the flood season; the questioning faction says it has already waited two flood seasons and a third cannot be waited out. Each side places the losses of different people first — this is precisely what Chapter Twenty-One identified as beyond review's reach: to stuff value disagreement into review produces only position choices dressed in the robes of adjudication. What should take over is the agreement, but what the agreement takes over is not the disagreement itself; it is action within the disagreement.
No agreement is not neutrality. The flood season is a natural deadline: if no arrangement is made within three months, transport capacity will flow to the work sites under existing authorizations, reinforcement will proceed as usual, and compensation will remain at the end of the ledger — an outcome favored by the committee faction that realizes itself automatically without anyone cheating. Chapter Twenty-One analyzed the time dividend of the beneficiaries of the status quo; here, "wait until the disagreement is resolved before acting" is itself an action, one that maintains the existing distribution and books the costs to the absent side. The reason an agreement is needed is not harmony; it is that no agreement already has consequences.
A limited agreement is thereby defined: a time-limited commitment by both sides on the concrete terms of joint action — what is done, who contributes what, when verification occurs — rather than a reconciliation over the ordering of values. Chapter Nine already showed, in the context of the small town, that a limited agreement does not require unity of values; this chapter decomposes it into structure: every element of an agreement's design — object, duration, exit, moral basis — should be derivable from the premise that "the disagreement will not end." An agreement that requires prior mutual recognition in values in order to function is not a limited agreement; it is a ritual of submission disguised as procedure.
The Delimitation of the Agreement's Object
The terms of an agreement should satisfy the same discipline as entries in the base: write observable actions, not states and sentiments. "Complete the reinforcement of the northern section before the flood season; pay the first installment of compensation for the first thirty items within sixty days" is a term; "jointly guard the safety and justice of the river valley" is not — the former can be verified, violated, and registered; the latter can only be proclaimed. Chapter Nineteen required that entries register claims rather than conclusions; the corresponding discipline of agreements is to register actions rather than intentions. The closer the terms are to the observable, the smaller the surface of dispute at breakdown; the closer they are to intentions, the more each side can declare that the other never truly performed.
The numbers an agreement cites should already be in the base, or be registered simultaneously at signing: the total of compensation rests on the requisition entries, the work schedule on the ledgers of the work camps — the two sides should not discover at performance time that they hold different versions of the underlying figures. A term that cites unverified numbers is as fragile at performance as it was loud at signing; every such spot is a pre-embedded point of rupture. The version chain of Chapter Twenty is here the agreement's cost control — a dispute over numbers that returns to the base is far cheaper than one that returns to before the base was rebuilt.
The text of an agreement should contain no statements of motive or identity. Sentences of the kind "out of loyalty to the river-valley community" generate no verifiable obligation, yet they stockpile ammunition for later moralizing accusation: the side required to profess loyalty at signing will, if it defaults, be described as having betrayed that loyalty, though what it violated was only some specific term. Motive clauses are a reenactment, within agreements, of the declaration-based standing mechanism analyzed in Chapter Nine — folding the commitment to act into a proof of character, so that exit automatically carries the appearance of treason. A clean agreement registers only terms and conditions; motives are left to each faction's narratives.
Duration Is a Constituent, Not an Appendix
Chapter Eighteen warned that when a public standard is open to argument only at its founding, it will continue to govern later subjects after circumstances change; Chapter Nineteen turned that warning into the validity period of entries. Agreements need the same temporal structure: an agreement without a term announces that the value disagreement has already ended, whereas the whole legitimacy of the agreement derives precisely from the disagreement's persisting. The term distinguishes an agreement from surrender — the weaker side accepts provisional terms because the value argument has not been confiscated, only queued. An agreement without a term smuggles "provisional" into "permanent"; the residents of the next flood season will inherit an arrangement they never discussed.
Expiry means neither automatic renewal nor automatic lapse. An agreement should specify at signing the structure of renewal: how many days before expiry review begins, on which records of performance it proceeds, and what changed conditions — water levels, accounts, migration of population — trigger early renegotiation. Renewal is a new act of limited consent, not an inertial glide; after its third renewal, an agreement that renews by default is no longer read by anyone, only cited. Chapter Nineteen required that lapsed entries move to the archive rather than be deleted; the counterpart here is the complete archiving of an expired agreement: performance records, registers of default, reserved objections — the material of the next review.
A term is not neutral. For the downstream residents, three months is too long — the risk accumulates daily; for the requisitioned owner awaiting compensation, three months is too short — his loss continues from the first day the agreement is in force. The fairness of a term lies not in splitting the difference but in making the burdens visible: the agreement should register what each side bears during the term and make that bearing visible — paying the first installment of compensation during the construction period is the acknowledgment that delay itself has a price. An agreement that treats duration as a purely technical parameter mutes one side's waiting costs, and muted costs return in due course.
Exit Clauses of the Limited Agreement
An agreement may set a term; it cannot cancel exit. Chapter Four showed, in the context of the reading room, that exit does not mean negating all common values but only ends the corresponding relation of participation; the agreement institutionalizes this principle: any party retains the right to withdraw within the term. Withdrawal requires conditions — advance notice, completion of work sections, settlement of resources — but not the approval of the other side. An agreement that requires approval to exit is not an agreement but membership of unlimited duration: it carries into public action the accumulating-debt structure criticized in Chapter Four, where the longer one performs, the harder it is to leave.
The exit clause must distinguish two kinds of obligation. Incurred obligations do not vanish upon exit: advances already drawn, work sections already begun, payments already promised to the ferryman and the stonemason are handled under the settlement terms at withdrawal; obligations not yet incurred end with exit. This distinction marks out the cost structure of exit — it has costs, and exit is therefore not an option to be repeated free of charge; the costs are bounded, and exit is therefore not indenture. The settlement rules should be written at signing; if afterward they are interpreted only by the remaining side, an opportunity exists to load additional burdens onto the one who leaves, and whether this is biased must still be checked against the specific terms and decisions.
Exit should leave a trace: the withdrawing party, the date, the grounds invoked, the settlement record, all entering the agreement's archival layer. A community without exit records has only two narratives available — "he is still in the agreement" or "he was never reliable" — and both rewrite a conditional exit into a fact of identity. Chapter Nine showed how the friend-enemy mechanism expands a limited opposition into total character; exit registration is its counter-structure: one exit is one exit, with its reasons, conditions, and settlement, and at the next round of negotiation it is a citable record rather than an oral rumor of the kind "it was they who walked out last time."
Joint Action amid Disagreement
The moral basis of an agreement is often covertly replaced with: we were able to reach agreement because deep down we remain the same kind of people, and the agreement is merely proof of this deeper sameness. This version has a genuine attraction — it paints the terms in warm colors; and a genuine poison — it stakes the agreement's survival on a fiction: at breakdown, the collapse of the fiction cuts deeper than the loss of the terms themselves, and "so we are not the same after all" appears with the anger of the deceived. The honest version is far colder: we are not the same, we know it, and we must still act together before the flood season; the agreement records the conditions of this action, not our identity.
Chapter Four already showed that people can maintain the same condition for different reasons; the agreement uses this insight as its moral foundation. The committee's engineer maintains the compensation terms for the sake of flood control — the resistance of the requisitioned owner would delay the works, and paying the first installment is part of the logic of the project; the questioning faction's organizer maintains the reinforcement terms for the sake of restorative justice — if the embankment fails, the compensation entries sink with it. The two sides maintain the agreement for different reasons, and each may even regard the other's reasons as mercenary — this does not prevent the terms from being performed by both. An agreement that demands unified reasons begins to dissolve at the first meeting called to explain motives.
Chapter Four showed that remaining does not signify full endorsement; the corresponding discipline at the agreement layer is: signing commits only to the terms, not to the ordering of values behind the terms; performance proves only the keeping of the agreement, not persuasion. Recording the other side's performance as "they have finally admitted that repair comes first" is the most effective covert sabotage of the agreement — it forces the other side to choose between continued performance and public self-justification, and a side forced to justify itself will begin to consider exit. The agreement's text can stop this seam in advance: performance records register only the completion of terms and may not be cited as a change in the other side's position. Citation cannot be forbidden, but it can be pointed out — Chapter Twenty said that archives which declare where they lean are the more verifiable, and visible appropriation is the first attrition of it.
Majority, Minority, and the Place of Agreement
Chapter Four traced the boundary of majority rule: procedural power extends to common resources and the conditions of conduct, and cannot decide inner conviction and views of life. Within this boundary the agreement is a supplement to majority rule: majority rule handles matters that "must be decided though not agreed," while agreements handle matters that "must be acted upon though values are opposed." Majority rule crosses the disagreement by counting votes; an agreement goes around it by way of terms — the cost of the latter is slowness, the gain is that the terms carry the consent of both sides at once. Where the deliberative council could both vote and agree on the same matter, agreement should be tried first: a vote wins a decision; an agreement wins an execution.
A limited agreement does not mean that everything becomes exchangeable. The requisitioned owner's basic standing to lodge complaints, the right to initiate review, the right to use the base — these cannot serve as bargaining chips of an agreement; they are the entry tickets analyzed in Chapter Nineteen, and to place them on the negotiating table is to accomplish, through an agreement, an exclusion jointly consented to by both sides. Chapter Four showed that certain minimal boundaries cannot be canceled by majority negotiation, and the agreement, as a stronger form of consent, cannot cancel them either: it can exchange resources, sequences, and methods; it cannot exchange standing. Whether a boundary is crossed is judged not by whether the sum is fair but by the type of the thing exchanged — resources can be offset; standing cannot.
Why might the questioning faction accept an agreement that prioritizes repair rather than insist on confrontation to the end? Structurally, only one reason is legitimate: the agreement simultaneously preserves its position in the argument — the compensation entries remain in the base, the path of complaint remains in review, the value claims remain in the council's records, and the agreement has a term. What it accepts is not "repair first is right" but "act first under this flood season and these terms, with the argument continuing as before." If the agreement also demands that it withdraw its complaints, cease its questioning, and publicly affirm that safety is assured, it is not an agreement but extortion; the criterion has already been given in Chapter Nine — whether a limited concession is expanded into the surrender of total character and communal standing.
Delay, Breakdown, and Refusal
We now take up the first failure mode. Once the institution of limited agreements is established, the stronger side acquires a new weapon: the agreement itself. The side asked to change the status quo will find that an infinite distance separates entering negotiation from actual change — another round can always be held, drafts can be revised to the eleventh version, and nothing need move before the flood season. The appearance of negotiation is harmless, even virtuous; functionally it is the delay analyzed in Chapter Twenty-One amplified at the layer of agreements — delay by review consumes conclusions, while delay by negotiation consumes the term itself. The more the institution of agreement is respected, the sharper this weapon becomes.
The beneficiary is the side that benefits from the status quo. The embankment is reinforced as usual, transport capacity is committed as usual; every week the negotiation extends, it enjoys another week of the fruits of unilateral action — what it pays is only time at the table, while the distribution it seeks to defend runs on automatically. The bearers of cost are double: the requisitioned owner awaiting compensation bears the full interest of delay; the public credibility of the agreement institution bears a hidden attrition — one negotiation dragged past an entire flood season, and "agreement" becomes a synonym for "wait a little longer." This delay requires no malicious character: the negotiators may sincerely believe the next draft will be better; structural delay realizes itself through countless local sincerities.
The distinction cannot rest on inference of motive — Chapter Three warned against assigning motives directly from patterns of behavior — but on observable thresholds. Sincere negotiation has an internal tempo: the issues are listed and closed one by one, draft versions converge, and actual exchanges of material occur between rounds. Delaying negotiation shows the opposite signs: the same issue is reopened again and again, new conditions always appear just before signing, and during the talks the unilateral action of the status-quo side never slows — transport capacity still diverted, work sites still open. The most powerful single test is the acceptance of a time constraint: whether the stronger side will accept an idling clause such as "if sixty days of negotiation pass without signature, the disputed matter is submitted to review as it stood." A negotiation that refuses every time constraint is, however earnest its language, structurally delay.
Counterexamples are also real. Complex terms genuinely require long negotiation; sometimes it is precisely the weaker side that needs the longer negotiation to assemble its evidence and its organization — the analysis of the costs of proof in Chapter Seven applies here. The term should therefore itself be an object of negotiation; the test targets the refusal of every time constraint, not insistence on a reasonable term. The conditions for rebuilding common facts follow accordingly: a side whose dispute has been dragged past a critical deadline should be able to invoke the negotiation record and claim before the review structure a finding of "negotiation in place of change." The consequence of such a finding is not punishment but restoration: the disputed matter returns to review or to the vote under the conditions that obtained when the delay began, and the delaying side may not retain the fait accompli changes it manufactured during the delay.
How the Breakdown of an Agreement Is Moralized
The second failure mode is more covert. When an agreement breaks down, both sides face a contest over the authority to explain "who broke it," and the side in a dominant position holds a ready-made weapon: blaming the breakdown on the other side's "unwillingness to compromise" — rejecting the final draft, insisting on extreme conditions. The structure of this narrative has already been dissected in Chapter Nine: assigning character to positions. Structural disagreement — compensation and repair contending for the same transport capacity — is translated into a defect of character — obstinacy, extremism, inability to see the larger whole. Once the translation is complete, no subsequent discussion needs to touch structure at all: it suffices to repeat "they simply will not yield," and responsibility acquires a permanent address.
The raw material of the moralizing narrative is supplied by the ambiguities of the terms. Every agreement has unprovided cases — whether a sudden flood counts as force majeure, how requisitions outside the list are handled — and breakdown often occurs precisely in these zones. The question is then no longer who defaulted but who has the authority to interpret. If that authority falls in fact into the hands of the stronger side, then every dispute in the ambiguous zones is resolved in the way it prefers, and the moment the other side refuses, it is booked as "unwilling to compromise." The repair does not lie in tighter drafting — unprovided cases will always exist — but in agreeing in advance: disputes in the ambiguous zones are submitted to the review structure under the four conditions of Chapter Twenty-One, not filled in by the unilateral interpretation of the executing party.
The beneficiaries remain the same two kinds. First, the side that does not want to perform the agreement — moralization gives default a legitimate appearance, and "they were the ones who refused to compromise first" functions as an exoneration; second, the stance industry that trades on antagonism — "the extremist faction that will not compromise" is a perpetual-motion machine of mobilization. The bearers of cost are, besides the moralized side, the community's next chance of reaching an agreement: every breakdown that closes with "the extremists tore up the agreement" raises the threshold of trust for the next one. The deepest injury of moralization is that it fulfills itself — the side repeatedly described as unwilling to compromise will in the end find that its compromises are indeed no longer believed, and so will genuinely no longer compromise.
The hedge is not to demand restraint from both sides — appeals to restraint have no force in the face of interest — but to change the recording format of breakdown. What should be generated automatically at breakdown is not a bulletin of mutual accusation but a structural record: at which term or in which ambiguous zone the breakdown occurred, the final positions of both sides, the prior record of performance, and whether the dispute has been submitted to review. This record is isomorphic to the receipts of Chapter Twenty. When the next generation of river-valley residents opens the archive, what it finds should be "party A proposed paying thirty percent first, party B proposed settlement after the flood season, review accepted and pending," not "in such-and-such a year, the extremists tore up the agreement" — the former preserves the structure of the argument, the latter preserves only a charge that can be cited forever.
The Distinction between Refusing an Agreement and Destroying an Agreement
The counter to moralization is to draw a line the river valley has never had: refusing an agreement and destroying an agreement are two different acts with entirely different moral standing. The side that refuses an agreement says: these terms protect too little, the term is unfair, the performance records sit in your storehouse. What it refuses is a concrete proposal, not joint action as such. Chapter Nine already showed that refusing an agreement is not automatically extremism; this chapter advances it into a procedural principle: so long as the refusal simultaneously states testable conditions — which terms would be acceptable, which bottom lines cannot be conceded — the refusal is part of the negotiation and should be recorded as a negotiating position, not as evidence of character.
Destroying an agreement is another act: signed, in force, a concrete term violated within its duration. The line between destruction and refusal is therefore clear: whether the signature was given. To sign and then violate is destruction; to refuse to sign and oppose is refusal. A community that confuses the two exhibits a strange moral order: the side that tears up signed terms exculpates itself with "the other side would not compromise," while the side that declined to sign bears the charge of "destroying the agreement" — the charge is detached from the act, and only position assigns guilt. The charge of destruction is isomorphic to the charge of capture analyzed in Chapter Twenty-One: it can likewise be strategically aimed at those who never signed.
The boundary has another side as well: not all insistence is obstinacy. That the requisitioned owner insists on thirty percent of compensation paid first before work begins may be not greed but experience — the two earlier promises of "work first, compensation later" were never honored, and his stubbornness is a rational discount on the probability of default. The distinction between insistence and obstinacy looks not at the hardness of the attitude but at the structure: insistence offers testable conditions and a track of concession; obstinacy refuses every test and merely repeats the final demand. A community that describes all insistence as obstinacy ends up with two kinds of people: the fake-flexible and the genuinely obstinate — the latter produced precisely by the former narrative.
An agreement, however broad, has boundaries. The domains that do not enter the agreement — the complaints the questioning faction continues to file, the defense of the works the committee continues to make, each side's internal meetings — must remain lawful. If an agreement demands the freezing of every other conflict as a condition of signing, it covertly replaces the limited agreement with a total ceasefire; a total ceasefire requires total trust, which is precisely what the river valley lacks most. Chapter Nine showed the importance of maintaining non-agreement relations; the institutional expression at the agreement layer is: the terms explicitly list the matters covered, and outside the list the argument goes on as before, both sides proceeding under the common conditions of the base and of review — the coverage list is the written proof of the agreement's limitedness.
The Institutional Interface and Boundaries of Agreements
The agreement is not a new thing floating above the institutions of the river valley; it plugs into existing structures. Disputes over performance — whether the works are on schedule, whether compensation has arrived — are not resolved by volume but return to the review of Chapter Twenty-One: materials come from the base and the ledgers, procedure is fixed in advance, and the adjudicators keep a distance of interest from the outcome. Review cannot adjudicate values, and disputes over performance are precisely disputes of fact — was it done, was it paid — exactly review's jurisdiction. The agreement thereby completes a closed loop: value disagreement stays in the council, terms of action enter the agreement, and execution of the terms is entrusted to review — three kinds of conflict, each with its own structure, no longer contaminating one another.
The full text of the agreement, its revised versions, performance records, exits and settlements, and the structural record of breakdown are preserved under the version chain of Chapter Twenty: originals retrievable at multiple entry points, revisions with provenance, citations traceable. The use of this archive begins only after the agreement ends — the first material of the next negotiation is the previous round's performance record. The agreement archive extends Chapter Twenty's principle of "making malice leave an account" from records to commitments: the side with a record of delay will next time be required to accept harder time terms; the side with a record of performing on time will have earned lighter supervisory requirements.
Failure Boundaries and the Conditions for Rebuilding Common Facts
The signs are observable: negotiating cycles lengthen round by round while issues cease to close; bulletins of breakdown are always character narratives, while no structural record can be found in the archive; refusal and destruction have merged into a single word in public narration; the withdrawing party has never left a settlement record; the coverage list negotiates ever wider, until it swallows the disputes of the base and of review themselves. Any single sign may have another explanation; when several coexist, the word "agreement" has only rhetorical function left — it is used to name delay, to name blame, to name everything except limited joint action.
Restoration is isomorphic to the restoration of entry points in Chapter Twenty and begins with the cheapest step. First restore the structural format of the breakdown record — the record template is itself neutral, and even if only one side executes it, the other side's bulletin shows its hollowness by contrast. Then restore the distinction between refusal and destruction, pointing it out whenever public narration conflates the two — pointing out requires only citing terms and records, not power. Then reopen the audit of the coverage list, cutting the swollen agreement back to limited matters. The river valley in its period of failure still needs joint action — the flood season waits for no one — and the substitute now is not idling but a Chapter-Nineteen-style minimal arrangement: a provisional undertaking, single-item, short-term, with maximally simple terms, accompanied by an honest performance record, as the transition to rebuilding.
The boundary is thereby clear. A limited agreement does not end value conflict, nor promise that conflict can eventually be ended — it only enables parties whose values are opposed, when they must act together, to dispense with first remaking one another. Its moral basis is temporary joint action, not the discovery of deep sameness; its object is concrete terms, not value consensus; its duration and exit clauses are not technical appendices but the structural guarantee of the word "provisional." It will be used by the stronger side as a surrogate of delay and by the breakers as a stage for blame; these two failure modes have signs and can be registered, and registration is itself the beginning of repair. What the agreement can do is one limited thing: to separate "we do this one thing together" from "who we are," so that the former can hold without impersonating the latter. With this boundary held, the agreement will not swell into a total ceasefire nor shrink into the rhetoric of delay; the argument continues, and action still comes off — this is not the failure of disagreement but precisely the form order can take while disagreement remains capable of coexistence.