0.1The question
Every volume of "The Time of Questions and Answers" asks, in its own domain, what must be decided now and what would make the decision change. This volume reads war and its edges. The domain here is not the conduct of fighting; it is the decisions that surround fighting and follow it: an alarm to be raised or withdrawn before residents decide whether to go home, an evacuation budget to be divided before vehicle reservations lapse and care contracts fall through, a ceasefire's participation rules to be set while some of those affected have no seat at the table, a register of the missing to be kept for the families who will need it for decades, an alliance compact that a later generation may refuse to honour. The series reads every case on the same axes: who decides, on what grounds and with what missing, by when and who waits and pays meanwhile, whether the decision can be undone, what would make it change, and when it was judged. This volume's question, in those terms, is the series question in this domain: in war and at its edges, what must be decided now, and what would make the decision change?
Under that question lies a narrower one about the records themselves. In Meridian's defence questions and their judgments, how is the time between a question and its answer written down? The research question decomposes into the parts the series names: who decides by when, on what grounds and with what the record leaves out; who waits and pays while the question is open; what can be undone once an answer is given; what would reopen the decision; and where the judgments attached to a question part from one another. The question is asked of fifty synthetic settings in fictional polities, each carrying thirteen AI-generated judgments published alongside it. Every setting is a document, not an event. The study's evidence is what those documents write and what the judgments say; it is not testimony about anything that happened.
Four sub-questions carry the research question into the chapters. The first asks who decides and on what grounds, and what the record lists as missing from the decision. Chapters 9 and 14 treat this. The second asks about the deadline, the cost of waiting while the question is open, and who bears that cost. Chapter 10 treats this. The third asks what can be undone once an answer is given, what would reopen the decision, and whether a reader outside the room could observe the reopening condition. Chapters 11 and 13 treat this. The fourth asks whether the same series reads differently in the four layers of time in which Meridian sets it. Chapter 12 treats this. The sub-questions are the series axes turned into an inquiry, and the conclusion answers each of them with the ledger's counts.
The theme that holds the sub-questions together is time. Some of the questions this study reads sit close to their deadline: a crisis alert whose errors have been found, to be corrected or stood behind before nine in the morning; a water-purification-parts procurement, to be concluded by three in the afternoon; a ceasefire participation rule, to be fixed by six in the evening. Others sit at a distance of decades from the promise they regulate: an alliance compact whose exit settlement will be inherited by people who did not vote on it; a memory-evidence regime for a dispute whose original materials are on the point of being lost. What connects an alert corrected in hours to a compact honoured across generations is the same form. Every question in the collection carries a decision actor, a mandate, a deadline, options with trade-offs, a list of what is unknown, and conditions for reopening. The form treats both kinds of time alike. Whether it can hold them alike is one of the things this study puts to the records.
The plan of the study names a specific thinness in what the records write: who bears the waiting, and whether a reopening condition could be observed by anyone outside the room. Those are the two questions the ledger's counts speak to most directly, and the study returns to them when it lays its findings beside the older theories of war and its ending, of warning and surprise, of civilian control and machines, of justice after war and promises across generations. This study reads the classic and modern works as lenses over the records, not as verdicts on them. Where a lens fits a setting badly, the chapter that uses it says so.
The study's argument concerns the records and the ideas it reads beside them. It reports what each option in each case gains and risks, who bears the cost and when, and what the record says would reopen the decision. Where the thirteen judgments attached to a question part from one another, the study records the division and does not choose a winner. The chapters that follow describe how this is done and what follows from it.
0.2Three contributions
Three things are added, and each is narrower than the work it is set beside.
The first is a reading of three classic texts about war as texts about the time of a decision. Thucydides' second galley, in the Mytilenean debate of Book III of the History of the Peloponnesian War, is an answer that reaches a commander before a decree is carried out: a vote reversed overnight after two speakers, and a second ship sent to overtake the first, its rowers eating as they rowed. Hobbes' tract of time, in Chapter XIII of Leviathan, is his saying that war lies not in battle only but in a tract of time in which the will to contend is known — the same stretch of time that the fifty questions in this study are written to hold open. Kant's first preliminary article in Perpetual Peace addresses the document of peace itself: a treaty made with a secret reservation of material for a future war is not valid. This study reads those three as writings about the interval between a question and its answer, and sets them beside the modern literatures on ending wars, on warning and surprise, on civilian control and military organisations, and on what is owed after war. The reading is narrower than the works themselves: it attends to the temporal structure of the decisions they describe, not to the whole of each argument. Chapter 1 sets out the four literatures and where this reading places itself among them; Chapters 3 through 8 read the classic and modern works beside the cases.
The second contribution is a decision ledger of fifty questions on six axes, coded by two models of different families and read strictly and leniently, with Meridian's own frame wording kept apart from what the records say about the settings. The six axes were set before coding: the decision actor, the missing material, the deadline and who waits, what can be undone, what would reopen the decision, and the horizon of the decision. Three more cells follow from what the judgments and options themselves supply: where the judgments part, the human's place in the decision, and four higher goals that the options trade against. The two coders agreed on 46 of 50 cases about the human's place and on 50 of 50 about whether the human role was specified and whether the reason was recorded, but on only 28 of 50 about the decider's kind. The strict reading counts only what both coders found stated in the record's own words; the lenient reading also counts what the setting implies. Meridian's frame — the recurring lens of title, premise policy, watch question, usage and scope that a studio reads through before it judges — is measured and reported, and none of the 798 reopening conditions in the collection carries 30 per cent or more of that frame's wording. The counts are computed from the ledger by code and can be checked by any reader. Chapter 2 sets the instrument out in full.
The third contribution is what those counts show about these records. Every one of the fifty questions names a decision actor and a deadline and writes conditions for reopening, because the form asks for them; their presence is not a finding. What the counts show is a distribution. The cost of waiting falls, in the coders' reading, on those nearest the point where things stop rather than on whoever decides: 25 of 50 cases support that on the strict reading and 42 of 50 on the lenient one, with the rest unclear and none against. Conditions for reversal are written in all 50 cases, 798 in all, and 629 of those 798 state a number, a threshold or a time; the hypothesis that few decisions write such conditions in observable words goes against in every case in both readings. Where judgments diverge, the records list material as missing: 13 cases support that and none goes against it, with 37 cases not reaching the hypothesis. The harder a decision is to undo, the further apart the decider and the bearer, on 3 of 50 cases strictly and 30 of 50 leniently. The reason for a restriction is recorded in all 50 cases, so the hypothesis that restrictions outlive the memory of their reason goes against in every case in both readings. The thirteen judgments chose more than one option in 13 of the 50 cases. Of the 150 options the collection offers, both coders agree that 82 reduce the immediate loss, and what those options give up runs most often to long-term stability (23) and ethical coherence (22), then evolvability (17) and resilience (12). These are findings about Meridian's synthetic questions, not about wars or states. Chapter 15 gives the verdicts on all eight hypotheses; the conclusion answers the research question and the four sub-questions with these counts.
0.3The records
Meridian is the RISCON family's platform for finding what must be decided and following the judgment afterwards. Its questions are documents: each is stored under a code, with a title, a pack, a domain, a decision year and an assumption level. Every question in this study's collection is marked synthetic. Its setting names a fictional polity. Its premise policy states that organisations, events and figures are settings for decision-making. Its own words say that the issue posed is not a report of an actual event. Nothing in these records reports a real state, force, war or person.
A defence question holds a setting in a fictional polity, a decision actor and mandate, a deadline, a decision question, options with trade-offs, a list of what is unknown, and conditions for reopening. The decision actor is named with a mandate: what the actor may decide and what it may not touch. The deadline is a date and sometimes a clock time, and sometimes a moment defined by an event — the sealing of a voyage contract, the close of a transmission window, the renewal of a supervision certificate. The decision question asks what the actor will determine. The options each come with trade-offs: what the option gains, what it gives up, and who bears the cost. The unknowns are the material the record lists as missing from the decision. The reopening conditions are lines saying what would make the decision change. Together these fields hold a decision about war before its answer, with its deadline, its options, and what would reopen it.
With each question Meridian publishes thirteen judgments. Eleven come from named perspectives: chief of staff, source officer, domain analyst, independent challenger, authority counsel, operations planner, synthesis secretary, decision-quality auditor, principal's proxy, outcome auditor and improvement architect. A twelfth is a staff synthesis. The thirteenth is a judgment made alone. All thirteen are AI-generated, dated and revisable; the records carry timestamps of 9 September 2026 and are marked revision 1, which is to say the first published pass. The judgments are proposals about what the fictional actor should decide. They are not decisions that any institution has taken. The study quotes them as Meridian published them, and treats their agreement and their divisions as a record of AI-generated judgment, not as a report of what a real staff would say.
The fifty questions are drawn from twelve series in Meridian's defence collection: aftermath, alliance, autonomy, ceasefire, civil, industry, infrastructure, personnel, protection, space, stability and trust. Ten of the series carry four questions each; two — civil and stability — carry five. The questions are set in four layers of time: 2026; the 2030s; the 2040s; and the twenty-second century and later. The layers are an authoring convention of the collection, not a periodisation of real conflict. A 2026 question carries a present-tense synthetic setting. A question from the twenty-second century is labelled a synthetic assumption of a distant future and of institutions and technology. The same series runs through all four layers: the ceasefire series, for instance, carries a question set in 2026 and one set in 2168, and the autonomy series carries one in 2026 and one in 2192.
The records were written in Korean. They were translated once into English for this study. Every block quote in the study is a record's own sentence, in that translation. Where a record's field name appears beside a quote, it is the name the record itself uses. The translation is the study's own; the wording of the records was not changed.
0.4How the 50 were chosen
The fifty questions are all of Meridian's defence questions published by 2026-10-05. There was no selection within that set. The collection falls into twelve series, each of which carries four questions, one in each layer of time; two of the series — civil and stability — carry a fifth question. That is the whole of the selection: the boundary is the publication date and the domain, and nothing was chosen or left out inside it.
This matters for what the counts can mean. A sample drawn to test a hypothesis might over-represent one kind of setting or under-represent another. This collection was not drawn that way. It is a platform's defence output up to a date. The counts therefore describe Meridian's defence collection on that date, and not war or defence in general. When a count says that 42 of 50 cases support a hypothesis on the lenient reading, it says that about fifty synthetic questions as Meridian published them; it says nothing about how many real decisions of that kind have been taken or what their distribution would be.
The four layers of time distribute roughly as the collection has them: twelve questions in 2026, fourteen across the 2030s, twelve across the 2040s, and twelve in the twenty-second century and later. The ledger's time rows give the per-year counts; the study reports them as they stand and does not compute new ones. Each series runs through the layers, so that the same kind of decision — a ceasefire's terms, an alliance's costs, a machine's stop authority, a registry's custodian — is written down in each layer. This is what makes the fourth sub-questions possible: the same series can be compared across four times without changing the kind of question being asked.
The study does not add cases to the collection or take any away. Where a chapter reads four or five or six of the fifty, it does so because the plan assigns that set to the chapter's subject; the whole of the collection is coded in the ledger and counted in the hypothesis tests. Chapter 2 sets out the coding; Chapters 9 through 14 read groups of cases on individual axes; Chapter 15 gives the verdicts across all fifty.
0.5The method in brief
The method is one instrument applied to fifty documents, and Chapter 2 sets it out in full. What follows is the shape of it on one page.
Every question is coded on six axes, set before coding began: the decision actor, the missing material, the deadline and who waits while the question is open, what can be undone, what would reopen the decision, and the horizon of the decision. Three more cells follow from the records themselves: where the thirteen judgments part, the human's place in the decision, and four higher goals — long-term stability, ethical coherence, evolvability and resilience — that the options trade against. Axes A1, A3 and A7 can be read from the record's own fields: the decision actor, the deadline, and the thirteen selected options.
Each cell of the ledger is one of three kinds. A record cell is a value found in the record's words. A code cell is a value the coders assign from what the record states. A model cell is a value a coder infers from the setting. Two coders, AI models of different families, read every question independently and gave a value, a basis and a reason for each cell. Their agreement is reported per cell: 46 of 50 on the human's place, 42 of 50 on who bears the waiting, 40 of 50 on the horizon and on the reversal constraint, 37 of 50 on what reversal would cost, 28 of 50 on the decider's kind. Where the coders did not both find the same value, the cell is unsettled, and unsettled cells count as unclear in the verdicts.
Every count is read twice. The strict reading counts only what both coders found stated in the record's own words. The lenient reading also counts what at least one coder found implied in the setting. Eight hypotheses are pre-registered. A hypothesis is supported when 60 per cent or more of the decided cases it applies to support it, goes against when 40 per cent or less of those cases support it, and is unclear when support falls between those marks; a hypothesis is also unclear when the cases whose values are unclear outnumber the decided ones. The share is taken over the decided cases alone. Cases that do not reach the setting are counted as not applicable, and cases whose values the coders could not settle are left unclear; both drop out of the denominator. Every verdict in the study is the ledger's, and the study keeps each verdict exactly as the ledger gives it.
Meridian's frame — a recurring lens of title, assumption level, premise policy, comparison, watch question, usage, material, published, scope, response request and relative-to — is spread across several fields of the question rather than standing in one place. The frame rule keeps it apart from the findings: the opening of Chapter 2 quotes a frame sentence at most twice in the whole book, and the ledger measures how much of any condition's wording is frame wording. None of the 798 reopening conditions in the collection reaches 30 per cent or more of the frame's wording. What the question form guarantees — a decision actor, a deadline, and reopening lines in every question — is reported as a property of the form, not as a finding about organisations.
This study recommends no option. It reports what the records contain, what two coders could agree on, and where their readings part. Chapter 2 illustrates the coding on two questions: Support before the name and Withdrawing a crisis alert.
0.6What the evidence cannot show
The evidence in this study is synthetic. Every question is a setting in a fictional polity, written to hold a decision before its answer. The settings name committees, assemblies and boards — a family recovery committee, a ceasefire preparatory committee, a generational records compensation board — that do not exist. The dates inside the settings are the deadlines the questions set; they are not the dates of real events. Nothing in these records reports a real state, a real force, a real war or a real person. A count of fifty questions says something about fifty documents, not about the world those documents are drawn from.
The judgments are AI-generated. They were produced together under a shared frame and a shared inference plan, and they are marked revision 1, which means they are the first published pass and open to revision. This explains part of why they so seldom part: 13 of the 50 questions carry judgments that chose more than one option. The agreement of the thirteen judgments is partly the generator's agreement, and the divergence row of the ledger is a count of published dissent, not a measure of how divided a real staff would be. Chapter 14 reads the divisions and Chapter 15 reports the verdicts, and both say what the source of the judgments is.
The question form itself guarantees certain fields. Every question carries a decision actor and a mandate, a deadline, and conditions for reopening, because the form asks for them. Their presence in all 50 cases is therefore not a finding about how decisions are made; it is a property of the document type. What can be a finding is the distribution within those fields: who bears the waiting, whether a reopening condition states something a reader could observe, whether a deadline is a lapse or a target, and what the options give up. The study keeps the guaranteed fields and the distributions apart, and Chapter 2 explains how.
The coders are AI models of different families. Their agreement is a measure of two models reading the same words, not of two people reading the same records. The agreement rates run from 28 of 50 on the decider's kind to 50 of 50 on whether the human role was specified and whether the reason was recorded. Where the coders settled less often, the ledger says so. The study does not claim that the coding would survive a human re-reading.
This study is itself written by AI from Meridian's records. Its readings of the classic and modern works stay inside the reading notes attached to those works in this study's plan, and where the study marks an interpretation as its own — "this study reads … as …" — the interpretation goes past the note and is this study's construction, not the author's claim. The works are lenses over the records, not verdicts on them. Where a lens fits a setting badly, the chapter says so. The bibliography lists exactly the works the study used.
What none of this evidence can show is anything about real wars, real states, real forces or real people. The study makes no claim about how any government, army or committee has decided anything. Its findings are about how a platform's synthetic decision questions are built and judged, and about what a reader can and cannot learn from reading them on the same axes. The collection is one platform's defence questions published by a single date; the counts describe that collection and nothing wider.
0.7The plan of the study
The study is a preface, an introduction, sixteen chapters in four parts and a conclusion. The preface says who made the study and how: a volume of the Meridian series, written by AI from Meridian's published defence questions and the thirteen AI-generated judgments with each, with the design made by an editorial session and the chapters drafted by a model and checked by others.
Part I is the state of the question and the method. Chapter 1 sets out four literatures on deciding in war and at its edges. The oldest concerns war as a period of time and the act of ending it: Thucydides, Hobbes and Kant among the old texts, with Clausewitz's war as a continuation of policy beside them, then Iklé and Goemans on how wars end, Fearon on why costly wars are fought at all, Fortna on why ceasefires hold and Walter on why settlements of civil wars hold or fail. The second asks why decisions about war are taken on bad information or on good information nobody believed: Wohlstetter and Betts, then Jervis and Janis, with Allison's three models of how a government decides and Schelling on commitment and coordination beside them. The third asks who directs armed force and what organisations do when the answer is ambiguous: Huntington, Janowitz, Cohen and Feaver on civilian control, Posen on doctrine and Van Evera on the belief before 1914 that the offence had the advantage, Sagan on accidents in nuclear command, Scharre and Asaro on autonomous weapons. The fourth concerns what is owed after mass violence and what a promise is worth to people who did not make it: Teitel, Minow, Hayner, Bass and Elster on justice after transition, Snyder on alliances, Jonas on future generations. The chapter marks the gap: none of the four reads a collection of decision questions on the same axes for what those questions write about the deadline, the bearer of waiting, what can be undone and what would reopen the decision. Chapter 2 is the method. It sets out the fields of a question and of a judgment, the nine axes, the three kinds of cell and the two coders, the strict and lenient readings, the eight hypotheses and the verdict rule, the frame that is kept apart from the findings, and two questions coded in full.
Part II reads the classic and modern works beside sets of cases, four to five cases per chapter, each set drawn from one series of the collection. Chapter 3 reads the second galley of Thucydides' Mytilenean debate and Kant's first preliminary article beside four ceasefire questions in four layers of time, from Parties outside the signature in 2026 to Truce across generations in 2168, and asks what the questions bear of the debate about why ceasefires hold and when wars end. Chapter 4 reads Aeschylus' Cassandra — a prophetess fated never to be believed — beside four questions on the crisis line: a broken communications line, an arms-reduction verification file, a reassurance proposal sent years before a reply can arrive, and a consultation body that reads an adversary's warning as threat evidence. Chapter 5 reads Hobbes' tract of time and Clausewitz's continuation of policy, and Sophocles' Antigone on edict and unwritten law, beside five questions of emergency rule: an emergency budget ending, a parliament that lost its quorum, returning citizens to be verified, emergency authority that ends before a reply arrives, and a military communications extension. Chapter 6 reads Scharre and Asaro on human control and machines, Sagan on safety in nuclear command, and Posen on doctrine and Van Evera on the cult of the offensive, beside four questions from the autonomy series in four layers of time: unprovable human approval, an overseer's budget, records to be kept before the original disappears, and branched stop authority. Chapter 7 reads Snyder's alliance dilemma and Jonas on future generations beside four alliance questions, from an unfilled rescue fund in 2026 to ancestors' oaths and exit rights in 2164. Chapter 8 reads the Odyssey on return and the transitional-justice works of Teitel, Minow, Hayner, Bass and Elster beside five questions of aftermath: records to open, a restored voice, an unverified video, conflicting missing-person records, and a question of return or stay.
Part III reads the whole collection axis by axis. Chapter 9 takes the decision actor and the human's place, with five questions on service, consent and verification, and hypothesis H7 on whether a human's place left to an AI or an automatic rule is empty or a name only. Chapter 10 takes the deadline and who waits, with six questions on care that cannot stop and on registries and consent, and hypothesis H1 on who bears the cost of waiting, beside Luhmann's distinction between risk and danger. Chapter 11 takes what can be undone, with six questions on contracts and production lines and on facilities that outlive their makers, beside Friedberg on the defence economy and Collingridge on corrigibility, and hypothesis H4 on how far apart the decider and the bearer sit. Chapter 12 takes the four layers of time, with five questions including one series read across all four layers and one from the far future, beside Khong on analogies and Neustadt and May on thinking in time, and hypothesis H6 on whether the longer the horizon the more conditions are set for revisiting.
Part IV tests the study's own machinery and sets its findings beside the theories. Chapter 13 returns to eight questions for their reopening conditions alone, asking which conditions a reader outside the room could observe and which leave the trigger to judgment, beside Clarke on fantasy documents, and hypothesis H2. Chapter 14 returns to thirteen questions where the judgments parted, reading one-against-twelve dissents and close divisions, and hypothesis H3 on whether divergence tracks what is listed as missing. Chapter 15 gives the verdicts on all eight hypotheses, from the ledger's counts and the pre-registered rule. Chapter 16 sets the findings beside the four literatures from Chapter 1, with Kaldor's new wars and Walzer's two sets of rules as the chapter's own comparisons, and closes with an accounting of what the study adds and what it does not.
The conclusion answers the research question and the four sub-questions one by one, with the counts from the ledger. It records the contributions and the limits. It names what further research would look like: the other books of the series and what Volume 0 will do with this ledger beside the others, human coders, and real public records. It ends with the study's own conditions for reopening, written so that a reader could observe them.