Chapter 3. Ending a War: The Second Galley and the Ceasefire

3.1The Second Galley

The reading note this study is given tells, in Book III of Thucydides, History of the Peloponnesian War (1874), of a vote reconsidered and of a decree overtaken before it was carried out. The Athenian assembly voted to put the men of Mytilene to death. The vote was reconsidered the next day, after Cleon and Diodotus spoke, and Diodotus' motion carried by a near-even show of hands. A second galley was sent off in haste, its rowers eating as they rowed, and the note gives the sentence: "Another galley was at once sent off in haste, for fear that the first might reach Lesbos in the interval, and the city be found destroyed; the first ship having about a day and a night's start."

The sequence carries three points. The first is that the deadline was real but not singular: the first vote stood as a decree while the second was being argued, and the time the reversal had to work in was set by the ships' travel and by when the commander carried the first decree out — the note gives the first ship about a day and a night's start. The second is that the reversal required an instrument, not only an argument. Without a galley whose arrival time differed from the first ship's, the second decision would have been a record rather than an answer. The third is that what arrived stopped one thing and not another, and this study leaves that distinction exactly as the note gives it.

Book V is the counter-case, in one sentence of context: Athenian envoys argue from power and interest, Melians from justice and hope; Melos refuses, is besieged and taken, and its men are killed while its women and children are enslaved. The envoys state the ground plainly: "right, as the world goes, is only in question between equals in power, while the strong do what they can and the weak suffer what they must." This study reads the two passages as two different kinds of answer to the same shape of question. In the Mytilenean debate the assembly changes the answer by reconsideration, and the second galley is what makes that reversal effective, by reaching the commander before the first decree is carried out. At Melos the answer does not change; the note records no reconsideration, and this study adds no reason the note does not give.

The Melian passage is used here as a lens on the limits of argument before power, not as a description of how any of these four questions will be answered. The envoys speak from interest and the Melians from justice and hope, and nothing in that exchange is a rule for a committee writing participation rules or custody terms. This study takes the two passages as two shapes of decision, one reconsidered in time and one not, and leaves the comparison standing on that difference alone.

The records read in this section have neither a debate nor a galley. Each holds one question, one deadline, and lines that would reopen the decision later — conditions for reconsideration written in advance rather than a reversal arranged in advance. How such records are built and coded is set out in the method; this chapter reads four of them as they stand.

3.2Kant's First Article

Kant, Perpetual Peace: A Philosophical Essay (1795), sets out six preliminary articles, and the first of them states a rule about the document rather than about the conduct of war: "No treaty of peace shall be regarded as valid, if made with the secret reservation of material for a future war."

What the article forbids is not a clause but a reservation: material kept back and left unmentioned — the secret reservation of material for a future war that the article names. Validity is made to depend on what the document leaves outside itself. Of the three definitive articles of the same essay, the second, a federation of free states, is where this study takes up what free states owe one another; Chapter 7 reads it alongside the alliance questions. This chapter takes the first article alone, because all four ceasefire records are documents about scope.

Each of the four sets down what it is not deciding. The 2026 question excludes permanent political status, amnesties and the voting structure of a final peace agreement. The 2031 question excludes the determination of liability for violations, coercive sanctions, the composition of a standing monitoring body and the long-term retention period after the pilot. The 2044 question excludes territorial sovereignty, the legitimacy of the non-signatory group, long-term political participation and the conditions for joining the final agreement. The 2168 question excludes approval by future governments, the other side's lawful political successor and permanent political integration.

This study reads those boundary fields as the nearest modern analogue to a reservation, though the analogy is imperfect: Kant's article is about material kept for a future war, and these are published limits on a committee's authority and questions deferred in the open. They are not secret; the form asks for them, and a later reader can find them in the record. That distinction matters, because it turns the Kantian worry from a question about concealment into a question about what has been deferred. A boundary that is visible still defers something, and the people it may affect have not decided it; several of these boundaries say in so many words that the present actor cannot consent for others, so what is deferred is not yet an obligation on anyone. Whether a reservation of that kind keeps faith with those people is a question the records pose and do not answer.

The companion worry, that a restriction outlives the memory of its reason, is a question about the collection as a whole; the verdict on that hypothesis is reported in Chapter 15.

3.3Four Ceasefire Questions in Four Times

The four questions form one series, set in four layers of time, from 2026 to 2168. Each is read here on three axes: A1, who is at the table; A3, the deadline and who waits outside it; A6, the horizon the question is written for. The axes are set out in the method. The settings are synthetic — fictional polities, fictional committees — and the judgments quoted are AI-generated, quoted here as Meridian published them. Nothing in this section reports a real war, state, force or person.

Parties Outside the Signature (2026)

Synthetic scene.

A room on the coast of Virtual Nuriman in 2026, where the seating around the table is the thing the participation rules have to settle. Twelve support staff hold the file and four days of preparation remain; the contracts for interpretation and lodging are to be finalised the day after the deadline. Two main negotiating delegations have sent provisional ceasefire drafts, three minority armed groups have applied to take part, and a refugee residents delegation has asked for separate participation — of which only the fact of receipt is confirmed. Eyes go to the registration list, to the objection folder, and to a warehouse that an independent logistics verifier will compare with rationing capacity every two days. What is waited for is a set of participation rules by 18:00 on 12 October 2026, after which the record warns that only the existing draft authors will be registered.

The decision question is:

The published staff judgment decides:

On A1 the record names a committee whose mandate states that it cannot substitute for the signatures of any group or the consent of local residents; what is on the table is a seating rule, not consent. The ledger codes the decider in this case as team-or-organisation. On A3 the bearer cell is unsettled, though the record's own stakes name refugee residents who may lose their voice in decisions about their return, members of the minority group whose status may be fixed by an agreement they took no part in, and residents on rationing whose predictability may go. On A6 the horizon is days-to-months, and the deadline is an hour of the day.

The judgments on this question part: ten of the eleven perspectives chose O3, the independent challenger chose O1, and the staff synthesis and the solo judgment chose O3. Thirteen of the fifty questions had published judgments choosing more than one option, by the ledger's count, and Chapter 14 quotes what this dissent turns on.

What the alternatives give up is written into the record beside what they give. O1, one joint table for all applicants, lets representativeness objections be handled in plenary, and the record warns that twelve support staff may be split up and the signature of the livelihood supplies movement agreement delayed. O2, a separate opinion meeting with the main delegations, can advance draft negotiations quickly and record separate opinions, and reduces the influence of parties outside the plenary over changes to the text. O3, the one the staff synthesis and the solo judgment chose, secures an independent voice for civilian residents and carries the cost the record names: the procedure for reconciling two texts and electing representatives. The independent challenger's O1 turns on the same unverified item as O2's speed — whether the delegations would actually delay signature — and that is one of three unknowns, along with what the minority applicants can promise their members and what the unreachable refugee residents want. Who bears the losses differs: under O1 and O2 the cost of delay falls on residents on rationing, while under O3 the cost in staff time falls on twelve support staff across four days of preparation, and the day after the deadline the interpretation and lodging contracts are fixed. The lines that could change the judgment are given to named roles: daily review of objections, a two-day comparison of warehouse and rationing capacity, and per-session records of speaking and text handling.

Observation Records Keeper (2031)

Synthetic scene.

A verification pilot on the border region of Virtual Seon in 2031, with ten days left before the equipment can begin collecting. An accounting confirmation states that part of the custodian candidate organisation's operating costs came from one negotiating delegation; a test result states that the observation equipment also collects resident movement data. The board's budget covers one record-retention system and two independent audits, and the candidate has refused to bear the cost of separating the residents' data. Eyes go to the access log, to the draft contract, and to the weekly cycle the personal information auditor is to run once collection begins. What is waited for is a retention and access system by noon on 20 June 2031; without it, the record warns, no joint record of compliance remains while collection is delayed.

The decision question is:

The published staff judgment decides:

On A1 the record names a neutral verification board whose mandate approves record management only and delegates neither liability nor sanctions; the ledger codes the decider as unsettled. The question is a verification design question in its purest form: whether a joint body can be trusted to hold the records on which every later accusation will depend. On A3 the bearer is coded as those nearest the stoppage, and the record's stakes name border-region residents who may lose privacy and safety through exposed movement data, ceasefire parties who may lose the chance to prove performance, and independent investigators who may lose the basis for a judgment. On A6 the horizon is days-to-months and the deadline is a noon.

Each custody design in the record buys one thing and pays in another. O1, the joint custodian with restricted access, allows shared records to be cross-checked and audited quickly, leaves the suspicion that came with the candidate's funding, and carries the operating cost of separating resident data. O2, party-by-party retention with independent cross-checking, prevents any one organisation holding the records alone, and takes time to align different held versions and to settle the inconsistencies between them. O3, an independent trustee with delayed release, strengthens the independence of custody and the review of resident data, and delays the moment at which a party can object. On this question the thirteen judgments do not part: every perspective, the staff synthesis and the solo judgment selected O1, and the alternatives stay alive in their counterpositions. The board's budget reaches one retention system and two independent audits, and the candidate has refused the cost of separation. Under O1 the board's budget bears that cost and the parties bear the delay of review; under O2 the parties bear the time to align versions; under O3 everyone bears the delay of release. What none of the three answers is what the record leaves open: whether the financial link changed any handling, what disputes stay provable after resident data is separated, and how much access delay the parties will bear. The conditions that would change the judgment are written as checks a named role can run — weekly access logs and objection handling, weekly cross-check timing, monthly funding review.

Regions That Did Not Sign (2044)

Synthetic scene.

Ceasefire implementation in the interior of Virtual Dasol in 2044, with a budget that will fund one consultation window for six months. A non-signatory regional group has sent a document refusing the political provisions of the ceasefire while offering to take part in civilian water supply restoration and the exchange of missing persons information; the residents council of that region has applied for an independent window. An existing signatory party objects that a separate window rewards those who avoided signature, and service personnel in the region risk being left with unpaid costs while applications stall. Eyes go to the responsibility table — who responds, who implements, who pays — and to the queue of water and missing-persons requests held up by the status dispute. What is waited for is a consultation procedure by 17:00 on 8 March 2044, after which the staffing contracts for the next six months are fixed.

The decision question is:

The published staff judgment decides:

On A1 the record names a peace transition council whose mandate fixes only an interim consultation status and recognises neither sovereignty nor territorial representation; the ledger again codes the decider as unsettled. This study reads the structure as a double bind: the record's execution plan warns that practical consultation may be interpreted as political approval, and not consulting leaves the residents' services stalled. That is why the boundaries are drawn so tight and why the judgments keep insisting that essential services and missing-persons access may not be made the price of political approval. On A3 the bearer is coded as those nearest the stoppage, and the record names residents of non-signatory regions bearing the cost of delayed water restoration and of delayed access to missing-persons information. On A6 the horizon is days-to-months and the deadline is an hour of the afternoon. The undoing cell is unsettled; what can and cannot be undone when a consultation procedure has been running for six months is left to the reader.

What the other two procedures give up is stated as plainly. O1, observer seats at the existing implementation meeting, keeps common procedures and records, and gives the non-signatory side so little influence on the text that the record doubts participation would last. O2, the window for essential services alone, lets water supply and missing-persons matters be discussed apart from political status, and widens the responsibility boundary between two bodies, with the duplication the record prices in. O3, resident-led consultation with per-group responses, brings resident demands onto the official agenda, and requires the groups who actually implement to answer separately, which lengthens processing. The thirteen judgments here do not part either: each perspective, the staff synthesis and the solo judgment selected O2. The council funds one window for six months, and the staffing contracts are fixed the moment the deadline passes. Under O1 the non-signatory side bears a voice that may not bind; under O2 the two bodies bear duplication and service personnel risk unpaid costs where responsibility is unclear; under O3 residents bear the wait for implementation responses. Running through all three is what the record does not know: whether a non-signatory group will implement practical matters without political status, whether any residents are left out of the local residents council, and how much work the two structures would duplicate. The conditions that could change the judgment are a weekly cross-check of who responds and who implements, a fortnightly check of settlement and delay, and a monthly check of applications by region.

Truce Across Generations (2168)

Synthetic scene.

A virtual remote harbour in 2168, where a document proposing the resumption of civil exchange has arrived after a one-way journey of eighteen years and whose signatory's term has already ended. The round trip of a reply is longer than the current term of the resident representatives; the next transmission window closes on 30 November 2168, and a missed window costs another three years of waiting. Residents scheduled to depart want their present consent respected; a Successor Residents' Council demands a right of refusal for burdens it has not approved. The record warns that the opposing exchange community may incur losses after spending preparation funds, and the judgments name possible costs to providers. Eyes go to the fund audit, to the individual consent forms with their withdrawal points, and to the received document, from which the record says the approval of the current representative cannot be confirmed. What is waited for is a set of entry-into-effect and succession rules before the window closes and before any reply has been received.

The decision question is:

The published staff judgment decides:

On A1 the record names a truce trust whose mandate states that it cannot consent for the future governments of either side or for residents not yet born; the ledger codes the decider as unsettled. What is at the table is not a seat but a rule about when an agreement takes effect, and who may be held to it afterwards. On A3 the bearer is coded as those nearest the stoppage, and the record names participants who must consent to waiting for a large part of a lifetime or to movement that is hard to reverse, successor residents who may inherit burdens they never approved, and an opposing community that may lose preparation funds. On A6 the horizon is unsettled; the ledger leaves ten of the fifty horizons without a settled reading, and this is one of them. The deadline itself is of a different kind from the other three: it is not a time of day at which an answer is due, but the closing of a window in which a message can still be sent — a deadline measured by transmission, not by arrival.

What the other two rules give up is written beside them. O1, continuous effect under current representative approval, gives departing residents and providers the predictability the record values, and leaves long-term costs that successor representatives never approved, with the dispute over legitimacy that follows. O2, periodic effect under generational re-approval, gives successor residents a chance to consent, and breaks the exchange schedule while approval is awaited, so that prepared facilities may stand idle. O3, the independent exchange agreement within fund limits, starts the part of the exchange the Trust can carry within its own authority, and reduces the comprehensive mutual assurance on both sides, raising the uncertainty for participating residents and for provider institutions. The thirteen judgments on this question do not part: every perspective, the staff synthesis and the solo judgment selected O3. Under O1 successor residents bear costs they never approved; under O2 departing residents and providers bear the idle years of a broken schedule; under O3 both sides bear reduced assurance, and the Trust's cancellation and closing costs stay inside the fund limit. The Trust cannot consent for future governments or for residents not yet born. The same three unknowns sit under all three rules: whether the opposing side's present representative system keeps the succession criteria of the received document, whether any procedure works for carrying the wishes of residents already in long-term movement, and whether consent procedures built for short exchanges still hold across a generation. The reopening lines are checks on authenticated changes of representative authority, on burdens beyond existing consent, and on expected fund overruns.

The four deadlines therefore are not of one piece. Three are hours of a day; one is the closing of a window. Three are coded days-to-months; one is unsettled. The series also shows the decider row at its most unstable: the record names a committee in every case, as the form requires, while the coders' cell for who the decider is reads team-or-organisation once and unsettled three times. Chapter 2 explains why the presence of a decider, a deadline and reopening lines is guaranteed by the form and is therefore not a finding; what is worth noticing here is the direction each mandate takes — every one of the four is defined by what it may not decide.

3.4The Debate: Why Do Ceasefires Hold, and When Do Wars End?

Five works in this study's tradition are taken up in this debate section, and each fits the questions this chapter has read differently.

Virginia Page Fortna, Peace Time: Cease-Fire Agreements and the Durability of Peace (2004), is the work these four records come closest to, and they cannot test it. The 2031 question asks what a joint verification body must look like to be trusted with the records; that is a design problem of custody and access, and the record carries no durability outcome, so it can illustrate a verification issue without being the same claim at a smaller scale. The 2026 and 2044 questions ask who sits at the table at all, which is a matter of participation rather than a presupposition of Fortna's finding, since a stronger agreement in her account can rest on devices other than a joint commission. The 2168 question strains the claim to its limit: what counts as a strong agreement when the opposing community's present representative system is unknown to the record and the Trust cannot consent for residents not yet born. Where the records cannot bear: none of them records whether a ceasefire held. Each record carries later synthetic updates — verification results and institutional or behavioural changes — and none of them reports how long an agreement lasted.

Barbara F. Walter, Committing to Peace: The Successful Settlement of Civil Wars (2002), is the work this chapter reads against records in which nobody disarms. The 2031 question has the shape of a commitment problem: the custodian candidate has taken money from one party, and the board must decide whether procedural control can substitute for independence. This study reads the 2168 question as an analogy to that problem rather than as the same problem, because Walter's commitment problem arises where the sides disarm and lose the power to protect themselves, while the 2168 record holds authority succession and consent to future burdens; the successor residents' refusal is still a commitment that no signatory can make on their behalf. The 2044 question shows the mirror image, in which a group that has not signed may implement practical matters without being granted political status. Where the records cannot bear: none of these questions records disarmament, and none records a third-party guarantee arriving as a fact. They record what would make one possible.

James D. Fearon, "Rationalist Explanations for War" (1995), gives three routes by which a costly war happens; the one these four records brush against is private information. Each of the four carries explicit unknowns — what a minority group can actually promise its members; whether the custodian's financial link affected handling; whether non-signatory groups will implement without political status; whether the other side's current representative system keeps its succession criteria — and an unknown in a record is not yet private information held by a bargaining party with an incentive to misrepresent it. The collection as a whole lists what is missing in 50 of 50 cases, 148 items in all, which is the form's way of holding Fearon's variable open rather than resolving it. Where the records cannot bear: these are not accounts of why parties fight. They contain no bargaining history, no deliberate misrepresentation, and no indivisible issue beyond the ones the synthetic settings name.

Fred Charles Iklé, Every War Must End (1971), holds that wars are ended by decisions that governments often fail to think through when they begin them. The form of Meridian's questions is itself an answer to that claim: each of the fifty names a decision actor, a mandate, a deadline, options with trade-offs, and conditions that would reopen the decision. This study reads the form as near to Iklé's criticism without answering it: he asks for the ending to be thought through when a war is begun, while these questions are written as a ceasefire is prepared, implemented or resumed, so the thinking is on the page where anyone can check it but not at the moment he names. What the form cannot supply is the ending itself.

The variable of H. E. Goemans, War and Punishment: The Causes of War Termination and the First World War (2000) — what ending a war will cost a leader at home — is the one these four questions lack. Their records write the cost of waiting chiefly as a cost to residents: to those on rationing, to families waiting on missing-persons information, to residents in transit. The judgments also record costs on the bodies that decide, from the committee's procedural work in 2026 to the Trust's cancellation and closing costs in 2168; what none of them carries is leadership survival. This study reads that gap as a limit of the form rather than as evidence against the account: these questions name a decider, a mandate and a deadline, and they are quiet about what the decision costs the decider politically.

The four questions bear hardest on Fortna's verification element and on Walter's commitment problem, in the two forms those take when no disarmament is involved: who keeps the records, and who may be held to a promise they did not make. They bear on Iklé in their very shape, and on Fearon only in their handling of unknowns. They do not bear on Goemans at all, and that absence is itself informative about what a decision record is for.

3.5Hands on

The next chapter takes up the warning line: Cassandra in Aeschylus, a prophetess fated never to be believed, and four questions set on a broken communications line, an arms-reduction verification file, a crisis notification authority, and a consultation body that reads an adversary's AI warning as threat evidence. The two chapters read as a pair. The second galley of Book III is an answer that reaches the commander before the first decree is carried out; Cassandra is a warning that arrives and changes nothing. Both are about the time between an answer becoming available and the answer being used.

The pair leaves three points for the next chapter. The first is the difference between a deadline and an arrival. The 2168 question's deadline is the closing of a transmission window, and the round trip of a reply runs longer than the current term of the resident representatives, though the record does not promise that a reply will settle the succession dispute; the 2026, 2031 and 2044 deadlines are hours of a day at which the decision is due. The second is what can be undone. The 2026 question is coded as costly to undo, the 2031 question as impossible to undo — a decision about records held — while the 2044 and 2168 cells are unsettled, which is the ledger's way of saying that the two coders did not agree on a value. The third is who waits outside the deadline. In three of the four the ledger codes the bearer as those nearest the stoppage, and in the fourth the cell is unsettled; the waiting falls chiefly on residents, on families, on people in transit, and on service personnel who risk being left with unpaid costs, while the judgments also put procedural, coordination and cancellation costs on the committees and the trust that decide.

What a reader should watch for in the next chapter is whether the questions themselves notice the gap between warning and belief — whether they list what is missing, as these four do, and whether the conditions for reopening a crisis decision are written in terms an observer could check.