Chapters 3 to 8 argued four literatures on deciding in war and at its edges: how wars end and ceasefires hold; how warnings are believed or missed; who governs when power is concentrated and when it is handed to machines; what is owed to the generation that signs and to the generation that inherits. This chapter sets the ledger's counts beside those arguments. It asks, in each case, what the records of Meridian's fifty defence questions can bear on a position in the tradition, what they leave untouched, and where a tradition fits the shape of these questions badly. It closes with an accounting of what this study has added and what no book built from these records could add. The counts come only from the ledger of the fifty questions; the descriptions of the works stay within the reading notes attached to them in this study, and this study's own readings are marked as readings.
16.1Ending wars and keeping ceasefires
Chapter 3 set out the debate the ceasefire questions were read beside.
The ledger sits beside this debate with counts that answer a different question from any of the works. Every one of the fifty questions carries a deadline, and every one lists what is missing from the room — 148 items in all. Hypothesis H2, that decisions rarely write down what would reverse them and that the written conditions are unobservable, goes against these records on both readings: 0 supports and 50 against.
Where the records bear on the theory of ceasefire durability, they bear structurally. Fortna's (2004) and Walter's (2002) mechanisms — verification, joint commissions, third-party guarantees, power-sharing — are mechanisms the reading notes associate with more durable ceasefires and settlements; the questions Chapter 3 read treat the composition of the table, the custody of observation records and the status of regions that did not sign as the substance of the ending rather than as a detail after it. That is a finding about how these records frame an ending: as an agreement about who is present, what will be watched and what will be recorded. It is not evidence that such agreements hold, because a question posed before an answer cannot show an outcome.
Where the records cannot bear is equally plain. Goemans's (2000) expectation of punishment at home and Fearon's (1995) bargaining dynamics belong to real wars between real governments, with the domestic and diplomatic history that comes with them. These records pose questions in fictional polities before any of it happens. Iklé's (1971) point — that wars are ended by decisions — is the one his note and these records share, and the sharing is thinner than it looks: the presence of a decider and a deadline in all fifty questions is guaranteed by Meridian's question form, as Chapter 2 sets out, so their presence is not a finding about endings. What the ledger supplies for this collection is a coded horizon of waiting. The coders agreed on the horizon of waiting in 40 of the 50 cases, most often days to months — 13 cases implied and 9 stated on that value. The agreed cases run past days to months as well: 8 implied and 5 stated at years, 3 implied and 1 stated at decades, and 1 implied at centuries or more.
16.2Warning and belief
This study quotes one line of Aeschylus's Agamemnon (1920), in Gilbert Murray's translation: "What if no man believe me? 'Tis all one."
The ledger beside the warning debate is unusually uniform. All fifty questions list what is missing, and both models recorded a reason for every decision in all fifty cases, as they also recorded the human's role in all fifty. The records write their unknowns down rather than leaving them implicit; listing an unknown is not the same as telling signal from noise or acting on a warning in time. The one hypothesis that speaks directly to this is H3 — where judgments diverge, material is missing — and it supports in 13 cases with none against and 37 not applicable, with material listed as missing in each case where judgments diverge. That is a finding about the room's information, coded from the records, and it does not touch belief.
The debate Chapter 4 left open — whether surprise is a failure of information or a failure of belief — cannot be settled by these records, and the reason is worth stating exactly. A record that lists its unknowns is silent on whether anyone in the room accepted them. Jervis's account describes decision-makers fitting new information to what they already believe and misreading others' intentions, while Janis's account describes cohesive policy groups seeking agreement over a realistic appraisal of alternatives; Allison's three models include the rational actor, organisational process and governmental politics, and only the organisational process one concerns how an organisation processes what it receives; the ledger can show that unknowns were recorded and that reasons were recorded, and cannot show whether those reasons persuaded anyone. Schelling's focal points fit the wrong direction here: the records contain common objects — deadlines, publication units, certification times, sealing moments — which is to say they contain objects that two parties might converge on without negotiation, but a question cannot show that a party reached one. His deterring and compelling register fits badly as well, since no question in the collection poses a threat to another party as such; the questions pose administrative choices inside one room.
This study reads Cassandra as a figure that fits these records closely, and reads her as a warning about the records themselves: an answer already present in the room and unable to make itself count is not the same thing as an answer that was never written down. The ledger's 798 conditions and 148 missing items suggest a collection full of written awareness. Cassandra's note says nothing about how such awareness behaves.
16.3Civilian control and machines
Chapters 5 and 6 set out two debates that meet at the question of who holds a decision and who can take it back. Hobbes, in Chapter XIII of Leviathan, holds that without a common power men are in a condition of war, and that war lies not in battle only but in a tract of time in which the will to contend is sufficiently known. The passages allowed in this study are "For WARRE, consisteth not in Battell onely, or the act of fighting; but in a tract of time, wherein the Will to contend by Battell is sufficiently known: and therefore the notion of Time, is to be considered in the nature of Warre; as it is in the nature of Weather." and "All other time is PEACE."
Clausewitz, in On War, Book I, Chapter 1, holds that war is a continuation of policy by other means, and Book I, Chapter 7 names friction, which makes the simplest thing difficult in war, while Book I, Chapter 3 holds that much of what action in war rests on is hidden in uncertainty. Sophocles' Antigone gives the tension between an edict and unwritten law in Antigone's answer to Creon, which in Francis Storr's translation runs: "Nor did I deem that thou, a mortal man, Could'st by a breath annul and override The immutable unwritten laws of Heaven."
Samuel P. Huntington's The Soldier and the State argues for objective civilian control through a professional officer corps that stays out of politics. Morris Janowitz's The Professional Soldier argues that the profession was coming closer to civilian society in its skills and outlook, and looks to integration with society and to a constabulary conception of force against Huntington's separation. Eliot A. Cohen's Supreme Command finds that against the "normal theory" — civilians set the aims and leave operations to soldiers — wartime leaders questioned and intervened in military matters, in what he calls an unequal dialogue. Peter D. Feaver's Armed Servants treats civil-military relations as a principal-agent problem: civilian principals monitor, and military agents may work or shirk. Paul Scharre's Army of None sets the human in, on or out of the loop against the pressures of speed and accountability. Peter Asaro, in On Banning Autonomous Weapon Systems, argues that lethal decisions require human judgment that a machine cannot supply, and calls for a ban on autonomous weapon systems and a requirement of meaningful human control over the use of force. Scott D. Sagan's The Limits of Safety sets normal-accident theory and high-reliability theory against the record of nuclear command, in which near-accidents occurred despite safeguards. Barry R. Posen's The Sources of Military Doctrine compares organisation theory and balance-of-power theory, and finds that military organisations left alone tend to prefer offensive, stable doctrines while civilians intervene to change doctrine when the international threat is high. Stephen Van Evera's The Cult of the Offensive and the Origins of the First World War holds that belief in the advantage of the offence made mobilisation hasty and made a first move seem urgent.
The ledger's contribution beside these debates is one clear negative and one clear positive. Hypothesis H7 — that the more a decision is left to a machine or an automatic rule, the emptier or more nominal the human's place — is not applicable in all fifty cases, with 0 supports, 0 against and 50 not applicable. No question in this collection leaves a decision to a machine; each one names a human decision actor and a mandate. The formulation H7 borrowed from the wider debate therefore has nothing in this collection to bite on. What the records show instead is that the human's place is written down, thoroughly: the human role was specified in all 50 cases, both coders agreeing that the role is stated in the record, and the place cell agreed in 46 of 50, with 33 of those where the value was stated and 13 implied, and 4 unsettled.
The weakest cell in the whole ledger is the identity of the decider: the two models of different families agreed in 28 of 50 cases, with 22 unsettled. Every question names a decision actor, and the coders still could not agree in 22 cases on what kind of entity that actor was — a team or organisation, an institution or a state, a multilateral body. This study reads that disagreement as a finding about naming rather than about authority: the records name a decider and leave its level of abstraction open, which matters for the Huntington-versus-Janowitz debate, because the separation or integration of military and civilian roles is a question about institutions, and an institution that cannot be typed from the record cannot be placed on either side of that argument.
Where the records bear on the autonomy debate, they bear on documentation rather than on firing. The autonomy questions Chapter 6 read concern what can be proven later — records that cannot prove human approval, the budget of the oversight body, preservation of originals before the system is replaced, and the succession of stop authority across branched approval systems. This study reads Asaro's meaningful human control and Feaver's monitoring as both turning on something of this kind: a record that a human decided and a route by which the decision can be stopped and reviewed. Sagan's near-accidents fit the shape of that worry — safeguards that exist on paper while events outstrip them — though these are generated settings and not the nuclear command record Sagan studied, so the fit is one of form, not evidence. Scharre's speed problem sits further along the sequence than these records reach: they ask what can be shown and who can halt, not what a human can decide in seconds.
Posen and Van Evera fit badly, and the mismatch should be said rather than smoothed over. Neither question — which doctrine a military adopts, whether an offence advantage is believed — appears anywhere in the collection. The doctrine question operates at the level of how an organisation builds its plans; these records operate at the level of who funds an oversight body, whose approval is on file, and when a certificate is renewed. A book about doctrine would read these records as the paperwork of doctrine; this study cannot do more than note that the paperwork is what the records contain.
16.4Generations and justice
Chapters 7 and 8 set out the traditions on promises that outlive their makers and on what comes after the fighting stops. Glenn H. Snyder's Alliance Politics names the alliance security dilemma: the fear of abandonment against the fear of entrapment. Kant's three definitive articles in Perpetual Peace propose a republican civil constitution, a federation of free states, and a cosmopolitan right limited to universal hospitality. Hans Jonas's The Imperative of Responsibility holds that technological power reaches so far into the future that ethics must take responsibility for future generations, and proposes giving the bad prognosis priority over the good where the stakes are irreversible. The Odyssey gives return as a matter of recognition and of what has changed meanwhile. Ruti G. Teitel's Transitional Justice names the forms justice takes after a political transition: criminal, historical, reparatory, administrative and constitutional. Martha Minow's Between Vengeance and Forgiveness sets trials, truth commissions and reparations out as responses to mass violence. Priscilla B. Hayner's Unspeakable Truths compares truth commissions on what they can establish and acknowledge, how they differ from trials, and the choices they face about naming perpetrators, reparations and access to records. Gary Jonathan Bass's Stay the Hand of Vengeance holds that liberal states support war crimes tribunals because their leaders hold legalist beliefs, but that such support is limited, since leaders are reluctant to risk their own soldiers to bring the accused to trial. Jon Elster's Closing the Books compares transitional justice from Athens in 411 and 403 BC to the twentieth century, on who is tried, purged or compensated and on the emotions, interests, reasons and constraints that shape those decisions.
Two further works belong in this chapter as well. Michael Walzer's Just and Unjust Wars holds that the rules on going to war and the rules on fighting it are judged apart, and offers "supreme emergency" as a narrow, contested exception. Mary Kaldor's New and Old Wars describes wars of identity politics, decentralised violence and a globalised war economy, and her own note records that the distinction she draws is contested.
Hypothesis H4 — the harder a decision is to undo, the further apart are those who decide and those who bear the cost — supports on both readings, but the readings differ sharply in how much of the collection they reach: 3 supports and 47 not applicable on the strict reading, 30 supports, 7 unclear and 13 not applicable on the lenient one. The two reversal cells, what could constrain reversing a decision and what reversing it would involve, were agreed by both models in 40 of 50 and 37 of 50 cases respectively, so the cost of undoing is coded in most of the collection, and the hypothesis's claim about irreversibility follows from that only in the cases where both cells settled. Hypothesis H5 — that a restriction outlives the memory of its reason — goes against these records on both readings, 0 supports and 50 against, and the ledger's reason row explains why: both coders recorded a reason in all 50 cases. This study reads that verdict as a verdict about records at the moment of writing rather than about time: a reason stated in 2026 or in 2164 is a reason present in the record, and the records are silent on whether anyone reads it after the decider is gone.
Where the alliance questions bear on Snyder, they bear on the trade-offs his dilemma names rather than on the fear his account describes. The records write the two pressures of abandonment and entrapment into their options — the autonomy of members against the continuity of protection on the ground; the persistence of an external guarantee against the restoration of regional autonomy — and a question can do that without any party feeling the dilemma. Kant's federation sits beside the far-future alliance questions as an ideal of voluntary association between free states, and the records reach for the exit question in its hardest form: whether a compact signed by an ancestor binds descendants who have not voted. Jonas's responsibility to future generations is literally posed there, and the ledger's H6, which asks whether longer horizons change what is decided now, is to be read across the books in Volume 0 rather than settled here; the strict reading codes 9 cases as days to months, 5 as years and 1 as decades, while the lenient reading codes 22 as days to months, 13 as years, 4 as decades and 1 as centuries or more.
The justice questions Chapter 8 read are administrative in a way that matches Teitel's forms closely: access regimes for handed-over records, evidence criteria where no witness survives, registration standards for returnees. Hayner's choices about naming, reparations and records are precisely what those questions decide at the level of procedure. Bass's limit appears nowhere in this collection, and that absence matters: no question asks whether to send soldiers to arrest an accused, so his account of legalist belief and its ceiling has no purchase here. Walzer's two sets of rules bear only partially, and this study reads the mismatch as a matter of where the questions sit rather than of his argument: the fifty questions mostly occupy the aftermath and the administrative edges of violence, where the separation between the rules of going to war and the rules of fighting does not map onto the question being asked. Kaldor's characterisation — decentralised violence, identity politics, a globalised war economy — describes the administrative world these settings occupy, with registries, missing-person records, cross-border rescue and returnees at the centre; this study reads the settings as occupying that world and cannot treat them as evidence for or against a distinction her own note marks as contested.
16.5What the study adds and what it does not
Five things are added, and each is narrower than the theory it was set beside.
The first is a ledger of fifty questions on the same axes, read twice. All fifty are synthetic records of Meridian's defence questions published by 2026-10-05, twelve in 2026, four in 2031, and the rest spread across the years the ledger's time table lists, the last in 2194. Every case was coded by two models of different families and read both strictly and leniently, so that a reader can check where the strict and lenient readings part. The agreement of those two coders is itself reported: on the decider cell in 28 of 50 cases, on the bearer in 42, on the horizon in 40, on reversal constraint in 40, on reversal action in 37, on place in 46, and on the human's role and the recorded reason in all 50. Where the coders part, the ledger shows it, and this study does not resolve it.
The second is a set of verdicts on eight pre-registered hypotheses, taken from the ledger and not adjusted here. H1 supports: the cost of waiting falls on those nearest the stoppage, with 25 strict and 42 lenient supports and no case against. H2 and H5 go against the records on both readings, 0 supports and 50 against, because these questions write conditions for revisiting and record reasons in volume. H3 supports in all 13 cases where judgments diverge. H4 supports on both readings but reaches only 3 cases on the strict reading, where 47 are not applicable, against 30 supports on the lenient one. H6 is to be read across the books in Volume 0. H7 is not applicable anywhere in this collection. H8 is unclear: of the 150 options across the fifty questions, 82 reduce the immediate loss on both coders' reading, and what those options give up — long-term stability 23, ethical coherence 22, evolvability 17, resilience 12 — does not run consistently enough to settle the hypothesis one way.
The third is a finding about observability that no theory in Chapter 1 supplies. The ledger counted 798 conditions for revisiting across the fifty questions, and 629 of them, 79 per cent, name something a reader could watch: a number, a date, an event or a role. The field that holds them divides unevenly. In the judgments' stop conditions, 602 of 650 are observable. In the questions' own reopening lines, 27 of 148 are. This study reads that split as the most precise thing the ledger shows: the observable reopening conditions in this collection come mainly from the generated judgments rather than from the question as it was posed, which is why the presence of a reopening line in all fifty questions is a fact of Meridian's form while the observability of those lines is a fact worth reporting.
The fourth is a finding about naming and distance. The human's place is stated in all 50 records, and the place cell was agreed in 46 of them; the decider cell was agreed in 28, with 22 unsettled, in a collection where every record names a decision actor. H1's bearer row was agreed in 42 of 50, and where it settled, the dominant value was those nearest the point where things stop, at 22 stated and 15 implied, against 2 stated and 2 implied cases where the bearer is a third party or the vulnerable, and one where the bearer is future generations. The cost of waiting therefore has a place in the record even when the typing of the decider does not settle.
The fifth is the comparison itself: four literatures that study cases, states, institutions and ideas, set beside a collection of questions that studies how decisions are written down. The traditions explain why wars end and ceasefires hold, why warnings are missed, who governs and who can stop, and what is owed after violence; the records cannot test any of that, because they pose questions before answers in fictional polities. What the records have that the traditions usually lack is the decision record in its own shape: a named actor, a mandate, a deadline, options with trade-offs, items listed as missing, and conditions for revisiting. Whether that record's presence changes anything is not something this collection can show.
What the study does not add follows from the same accounting. It adds no new account of war and no evidence about real states, forces or people; the settings are synthetic and were generated, and the judgments that accompany them are AI-generated, dated and revisable, which is also why their agreement is partly the generator's rather than the rooms'. The coders are themselves two models of different families, so the counts describe how such models read these records. The collection is one platform's defence questions published by a fixed date, not a sample of defence decision-making. The four layers of time are a design of the series rather than an observation about history, and the readings of the works in this study stay inside the reading notes attached to them, so where a note is short, this study says less rather than inventing a position. Nothing in this study answers which option is right in any question it has read; it shows what each option gains and risks, who bears the cost and when, and what the record itself says would reopen the decision. The findings are about Meridian's records and about this study's method; they are not claims about wars, states or the people in them.
The conclusion that follows takes the research question and the four sub-questions one by one, states the contributions and the limits in the same terms used here, sets out what would change the answers in the form of conditions a reader could observe, and points to Volume 0, where this ledger is to be read beside the ledgers built on other platforms and, in later work, on real public records and with human coders.