Chapter 14. Where the Thirteen Judgments Part

14.1How often they part

Meridian publishes, with each defence question, thirteen judgments: one from each of eleven named perspectives, a staff synthesis, and a judgment made alone. The judgments in the records examined here carry timestamps of 9 September 2026 and are marked revision 1, which is to say they are the first published pass, not settled verdicts.

The ledger's divergence row counts something narrow and mechanical: how many of the fifty questions had published judgments that chose more than one option. The answer is thirteen of fifty, with thirteen judgments per case. In the other thirty-seven questions, all thirteen judgments selected the same option, including the staff synthesis and the judgment made alone. Where the judgments part, they part in thirteen of the fifty questions.

Two further points belong on the face of this row. First, the divergence cell is a record cell, not a model inference. The two coders — two models of different families, whose names this study does not give — coded the decider, the bearer, the horizon, the reversal conditions, the place of the human role and the reason recorded behind each choice. Second, the agreement in the other thirty-seven questions is not evidence of wisdom or of convergence among independent minds. The judgments are produced by one system under one prompt family: each role receives the same synthetic input, the same option list and the same loss-allocation question, and each is asked to state a decision, its basis, its accepted losses, its conditions, its stop conditions, its milestones, its uncertainty and its counterposition. Thirteen answers that share a generator will tend to share a vocabulary of caution as well as a vocabulary of disagreement. This study reads that caution as a feature of the record, not as a finding about war: every judgment in these cases sets its confidence at medium or low, and every one names what it does not know.

What a parting judgment does show is that the record can carry disagreement without hiding it. Where the judgments part, each dissent is published beside the synthesis, with its own decision, its own accepted losses and its own conditions for reversal. There is no minority report in the usual sense: the dissenting judgment is not an annotation on the majority but a judgment published with the same fields as the others. The judgment made alone is a separate slot in the record, so a lone dissent is not necessarily that judgment, and shared fields do not by themselves give the judgments equal institutional standing. In thirteen cases a reader can see that two readings of the same synthetic situation reached different options, and can read what each of them thought it was giving up.

This chapter returns to the cases already read in earlier chapters to read what the divided judgments decided, the grounds on which they parted, and what each side said would make it change. It does not decide between the options. Where the judgments part, this chapter says what the two sides weighed differently and what each side thought would make it change; choosing between them is not this study's work, and the records themselves offer no vote to settle it.

14.2One against twelve

In the lopsided divisions, one judgment chose another option while the staff synthesis and the judgment made alone both took the majority side. The lone dissent is usually the Independent challenger's; the Goun case below is the exception, where it is the Source officer's.

Compact after the peace (2047). The fictional Dosom Peace Zone Ratification Assembly must decide, by 20:00 on May 6, 2047, what form the shared civilian protection compact takes after the armistice: re-ratifying the existing compact, transitioning to a civilian guarantee and dispute mediation system, or terminating after transfer to a regional review body. The Independent challenger's dissent turns on two unknowns the record itself leaves open: whether the new regional body can win the trust of the minority community, and whether contributing members will fund the compact after its form changes. Where the majority accepted establishment costs and a delay in autonomy to keep protection continuous while regional capacity is built, the dissent judged that the same unverified conditions are grounds for keeping the existing external channel rather than for betting on a transition that could fail. Each side wrote the reversal condition the other would watch for: the dissent would move to transition if the objection route and post-change funding were confirmed; the majority would move back to retention if either could not be secured.

Parties outside the signature (2026). By 18:00 on 12 October 2026, the fictional Nuriman Ceasefire Preparatory Committee must fix the rules of representative participation for provisional ceasefire negotiations, when a main negotiating delegation wants a rapid signature, a minority group wants a voice, and a refugee residents delegation asks that armed status not be the criterion for speaking. The judgments other than the Independent challenger's chose the dual representative meeting — an armed-groups meeting and a resident meeting reviewing the same draft, with joint text reconciliation. The Independent challenger alone chose a joint negotiation table for all applicants:

The dissent turns on the record's third unknown: whether the main delegations would actually delay the livelihood supplies agreement because of the participation rules. The judgments that chose the dual representative meeting refused to treat the unverified delay claim as sufficient grounds for reducing residents' influence: they gave the resident meeting a route into the text through joint reconciliation, and they accepted the coordination cost rather than exclusion from changes. The dissent treated the claim as unverified and preferred a single plenary, accepting possible delay as a bounded risk. This study reads the split as a disagreement about independent resident deliberation against direct plenary participation, not about what a ceasefire is for. The record's own trade-off for the joint table is that the limited support staff may be split up, and the dual meeting buys an independent voice at the cost of a second procedure; protecting a voice and spending preparation resources are simultaneous possible effects in the record, not two exclusive answers. Both sides wrote the same watch: if records show exclusion or formalised participation, the rules are re-decided either way.

Alternative service backlog (2034). By 4:00 p.m. on March 20, 2034, the Saebom Alternative Service Deliberation Council must decide how applicants whose conviction review is unfinished enter the next intake. The judgments other than the Independent challenger's chose interim inclusion in nonmilitary public support, conditional on individual consent and verified work suitability; the Independent challenger alone chose priority inclusion in distributed education for those who consent, deferring the rest individually. The dissent turns on the first unknown: the range of nonmilitary activities each applicant can accept. Its reasoning was that the label nonmilitary public support does not establish that the work fits a conviction, and that an education route lets each applicant consent to a specific content before any public-support assignment is made — at the cost of leaving public-support staffing gaps partly unrelieved. The judgments that chose interim inclusion place the work-adjustment cost on the Council and the operating agency, while acknowledging that a staffing burden remains with incumbent service members; the dissent takes the public-support staffing gap that education leaves partly unrelieved as its accepted cost. Both sides require individual suitability and voluntary consent, and both defer applicants whose conditions cannot be confirmed, so neither treats conviction protection as a cost to be traded for staffing.

Long-service reconsent cycle (2112). The record states the deadline as the renewal notice on 2053, May 31, while the scenario is set in 2112, and the judgments, including the Independent challenger's and the staff synthesis, write the notice as May 31, 2112. This chapter reports that discrepancy without correcting the record. On the record's deadline, the fictional Oreum Long-Service Contract Authority must set the reconsent cycle and the pay structure for service members who have received aging-delay treatment. The judgments other than the Independent challenger's, including the staff synthesis and the judgment made alone, chose three-year service segments with explicit reconsent and portable accrued pay that survives withdrawal, leave or transfer. The Independent challenger alone chose annual reconsent combined with an expansion of immediate pay:

The dissent turns on what the right of withdrawal actually buys. Its reading is that a formal right to withdraw is hollowed out when future pay expectations grow with service, and that an annual cycle with immediate payment gives a service member something real to change their mind about — while accepting recurring recruitment and retraining costs and the risk that annual renewal becomes employment insecurity. The twelve took the opposite worry seriously: that the same expectations bind families' care arrangements and block new entrants, and that portability carries accrued rights through a change of life better than a short cycle does. Both sides left open the same empirical question, that no record answers: how much retirement, leave and return each cycle would actually produce, and whether any future institution would honour portable allowances.

Who verifies confidential records (2041). The Fictional Goun Armistice Records Committee must decide, by 18:00, May 14, 2041, how external verification access to confidential civilian-protection records is arranged while residents object, confidential collaborators fear identification, and record authors fear stripped context. The judgments other than the Source officer's chose party access to de-identified records, assigning the limited review staff to the editing team. Here the lone dissent is not the Independent challenger's but the Source officer's, and it chose full access for a security review panel, with protected explanations and an objection path for the parties. The dissent turns on the record's listed unknowns: the cause of the difference between the published totals and the internal intake counts, the scope of material that can still identify an individual after de-identification, and the additional objection materials the petitioner can present under each access system. Context loss is a concern the judgments raise about de-identification, not one of the listed unknowns, and the two should not be run together. Its reading is that a protected explanation carrying the grounds identifiers and the reasons for restriction is enough to let a party object substantively, and that the panel's access to the original context is what distinguishes a counting-method difference from a processing omission. The twelve put the weight the other way: that a party who cannot check the grounds themselves is trusting an explanation, and that a record author whose work context is stripped cannot defend against an interpretation they never saw.

Across these divisions the lone dissents do not share one shape. Some refuse to move on an unverified condition; others move and attach their own verification conditions. The Nuriman challenger proposes a joint table for all applicants, the Saebom challenger proposes conditional inclusion in distributed education, and the Oreum challenger proposes annual reconsent with an expansion of immediate pay. Even the Dosom challenger, who keeps the existing compact, manages the unknowns through a time limit and periodic re-examination rather than by refusing to move. The twelve did not deny the unknown; they named it, accepted a bounded loss, and wrote the conditions that would force them back. This study reads the disagreement as one about the burden of proof: whether the side that proposes a change must show that the change is safe, or whether the side that keeps an arrangement must show that the standing arrangement is adequate. The change-versus-standing-still contrast does not run cleanly through the cases, because both sides commonly propose changes of their own — a joint table, an education route, an annual contract. The records also supply evidentiary requirements in their own words: consent, suitability, stoppage conditions and named persons in charge. What they do not supply is completed comparative verification, and that absence is not the same as an absence of proof conditions.

14.3Close divisions

The closely divided questions share one feature: in each the judgment made alone took one side while the staff synthesis took the other. These are the questions where the record cannot be read as one reading with an exception.

Emergency rule before reply (2118). By the sealing time of the communication packet scheduled to depart on August 1, 2118, the virtual Port Yomyeong Citizens' Succession Council must set the rules for ending and succeeding the local emergency rule authority, while a new generation becomes voters before the expedition government's reply arrives — a round trip of nine years. The judgments split between the separation of local rule from citizenship jurisdiction and the transfer of authority to the locally elected government. The judgment made alone took the separation:

The staff synthesis took the other side, choosing transfer to the locally elected government while accepting the preservation of citizenship records and return-application channels as a succession obligation rather than a redistribution of jurisdiction. The two sides weigh different unknowns differently. Both sides end the emergency authority when the delegation expires and both provide for the participation of new voters; the record states that a new generation becomes voters before the reply arrives, and does not establish that nine years of communication loses a whole generation's participation. The side that chose the separation treated the reply delay as a reason not to postpone local rule, and kept the citizenship connection through the existing records and application routes. The side that chose transfer treated the same delay as a reason not to invent a settlement nobody will recognise: jurisdictional separation has no empirical basis in the record, and the duties of records, applications and local remedy can be written into a single elected authority without splitting adjudication in two. Each side wrote the condition the other watches for. The side that chose the separation would reverse if a rights gap appeared that local remedy could not close. The side that chose transfer would reverse on authenticated replies and on independent case review, including citizenship-ombudsman findings of rights gaps and of the limits of local remedy; its judgments give more than one route back, so the two sides are not evidentiary mirror images.

Cross-border rescue consent (2043). By 09:00 on 3 November 2043, the Haesol Civil Rescue Committee must choose one consent verification system for cross-border civilian rescue while respecting both the departure-side permit and each individual's voluntariness. The judgments split between distributed remote interview verification and the joint in-person system. The judgment made alone took the in-person side:

Here the two sides weigh the same unknown against different costs. The side that chose remote verification did so because it connects a limited pool of interpretation staff across several regions and reduces the delays and care gaps that come from moving specialists and from interview queues; it made individual transmission suitability a condition of each movement rather than a reason to defer. The side that chose in-person verification did so because a remote interview cannot confirm on-site pressure, and because consent that is misread produces either an unwanted movement or a rescue the applicant wanted and did not get in time. The judgment made alone added a caution both sides would sign: that a gathering of people can itself be pressure, so in-person contact is not proof of voluntariness either. The staff synthesis took the remote side, accepting the burden of communication support and re-interviews on the Committee rather than shifting it onto applicants.

Memory replica testimony (2194). By December 2, 2194, 12:00, the Virtual Memory Evidence Committee must fix one evidentiary status for memory evidence under a preservation contract, in a dispute where an original party and three branched replicas show conflicting public intentions, and where the original materials become unreadable if the transfer is not concluded. The judgments split between preservation as a nonpublic investigation lead and preservation as derivative material of a common original. The judgment made alone took the nonpublic status:

The staff synthesis took the other side, choosing the derivative classification, which groups the replicas under a common source so that repeated memory is not counted as several independent confirmations, while preserving branch-specific statements separately. The side that chose the nonpublic status weighs the risk of premature use: creation history verifies where a record came from, not what it says, and the record states that a reply concerning the external original record arrives no sooner than twenty-seven years later, while most of the investigation's cross-verification budget would be consumed by transfer. That reply is not the only thing awaited: several judgments allow status reconsideration before it arrives, if suitable evidence and workable challenge procedures become available. The side that chose the derivative status weighs the delay that restriction imposes: descendants who claim omitted support and a replica who wants to speak wait through a period that the record gives no way to shorten, and a classification that keeps materials out of direct use can quietly become their permanent status. Both sides agree on what the record does not establish — that the number of replicas is the number of independent facts — but their conditions for reversal are not the same. The side that chose the nonpublic status requires grounds for distinguishing shared memory from post-branching experience, workable challenge procedures and permitted access before broader evidentiary use. The side that chose the derivative status would move toward independent testimonial status on evidence that some experience was acquired independently, and toward the nonpublic status if restricted review with substantive challenge cannot operate. What the two sides share is the subject they watch, not the rule that would follow.

In these close divisions the disagreement is not about whether the records are uncertain; both sides quote the same unknowns. This study reads each split as an argument about who should absorb an error, but the parties who absorb it are not the same from case to case. In the rescue case the risk of acting on a misread will and the risk of waiting both reach the applicant, as an unwanted movement or as a rescue that does not arrive in time. In the memory case a premature attribution of responsibility reaches the successor community of the accused institution, disclosure reaches the original party whose consent is withdrawn, and delay reaches the descendants and the replicas. In the emergency-rule case the disagreement is about jurisdiction and responsibility as well as about consent. The records do not establish that the same people bear both errors in all three, and this chapter does not treat the three as one shape.

14.4The rest

The remaining divisions did not produce a single-against-twelve pattern. None of them is quoted here; the shape of each division is enough.

Branched stop authority (2192). At the Supervision Certificate Renewal Time on April 14, 2192, the Virtual Dawn Orbit Community Approval Succession Conference must fix the succession of human supervisory authority and the distribution of stop authority for a branched military AI approval system. The judgments other than the Source officer's and the Independent challenger's, including the staff synthesis and the judgment made alone, chose common supervision succession with local immediate suspension — a local human supervisor may stop a branch without waiting for a distant reply, while the common supervisor holds certification and resume review. The ground of those dissents is the record's first unknown — whether the verification results of the existing central supervision apply to the branched system at all — and their reasoning is that a common certification whose scope of applicability is unverified should not be renewed, however well its form divides stop from resume.

Parliament without quorum (2032). By noon on June 19, 2032, the Fictional Constitutional Continuity Committee of the Haean Republic must designate a provisional decision-making structure for the civilian protection budget while the elected parliament lacks a quorum. The judgment made alone chose executive administration execution with separate citizen oversight, together with the judgments that are not read as dissenting in this paragraph, because the existing executive order expires in three days and representative selection cannot be verified in time. The ground of the dissenting judgments is not that continuity of the electoral mandate is the only authority the committee can document in time. The continuity protocol authorises the Committee to designate a provisional structure, and the Authority counsel locates the interim voting body's authority in that temporary designation rather than in a restored parliamentary quorum. Those judgments favour electoral continuity and the separation of approval from execution, with direct appeal routes for residents whose representatives are absent, and they hold that approval authority should not be concentrated in the executive administration on grounds of an unverified speed advantage.

Self-replicating facility ownership (2186). Before the production facility renewal on December 1, 2186, the Fictional Byeol-u Joint Production Trust must set the next structure of ownership and use rights for a self-replicating facility of essential materials, in a world where replicas remain across autonomous jurisdictions that the Trust cannot compel. The judgments other than the Independent challenger's and the improvement architect's chose joint trust ownership with continued joint dues, prioritising common maintenance resources and minimum access. The Independent challenger and the improvement architect chose ownership by autonomous jurisdiction with separate maintenance contracts, on the ground that the Trust's central disposition authority is limited by the setting itself and that aligning ownership with maintenance responsibility is more honest about who can actually keep a replica running.

Peace promise across times (2192). Up to the sealing of the outgoing dispatch of December 3, 2192, the Fictive Mir Inter-Civilizational Accord Council must fix the procedure for adjudicating meaning discrepancies in a private exchange accord, where replies take twenty-two years and consent means different things on each side. The judgments that took the majority side, including the judgment made alone and the staff synthesis, chose jurisdiction-based adjudication with mutual objection records, letting each side rule under its own criteria while recording the difference. The Source officer, the domain analyst, the operations planner and the improvement architect chose restricted adjudication on a prior casebook, arguing that where one side's consent means the other's lineage approval, no common procedure should be applied until both sides have agreed which types they share. Here several judgments dissent rather than one, and the staff synthesis sided with the majority.

Broken line interim authority (2032). By 6 p.m. on April 18, 2032, the Fictional Haeon Truce Liaison Council must choose an operating method for interim reception and relaying while a regular crisis communications line is out of service. The judgments that took the majority side, including the judgment made alone and the staff synthesis, chose a single neutral intermediary window that holds reception records and receipt confirmation while leaving message meaning to both parties. The Independent challenger, the authority counsel and the operations planner chose rotating operation of the direct bilateral liaison windows, on the ground that the parties' preferences for a neutral intermediary differ in the record itself, and that a single window would replace shift gaps with concentrated neutrality disputes.

Some of these dissents sit beside a recorded limit on what the decision can compel or verify: a Trust that cannot compel replicas, a certification whose applicability is unverified, and a dispatch that cannot be modified once sealed. Two of the limits the pattern claims are not in the records in those terms — the Haean record does not say the committee cannot restore a quorum, and the Haeon record does not say the council cannot compel a neutral body — and the records do not show that each cited limit explains a dissent rather than constraining both sides. This study reads the pattern as a reading of mandate boundaries rather than of outcomes; the records do not say whether the dissents were right, and nothing in this chapter makes them so.

14.5H3

The third pre-registered hypothesis asks whether material is missing where judgments diverge. In the words of "The ledger's numbers", it says: where judgments diverge, material is missing, with divergence checked against what is listed as missing. Two rows of the ledger bear on it.

The first is what every question lists. Fifty of fifty cases list what is missing, with 148 items listed in all. That is to say: in this collection, the field that records unknowns is never empty. The second is the divergence row already read in 14.1: thirteen of fifty cases had judgments choosing more than one option, a count taken from what the judgments themselves say they chose.

The verdicts, as "The ledger's numbers" gives them, are identical under the strict and the lenient reading. Strict: supports 13, against 0, unclear 0, not-applicable 37; verdict supports; share 100%. Lenient: the same counts, the same verdict. The 37 not-applicable cases are the undivided ones — questions where the thirteen judgments chose the same option and so offered no divergence for the hypothesis to meet. The 13 supports are the divided cases: in every one of them, the record lists what is missing, and no divided case lacked a listed unknown.

Three observations belong beside that verdict, and none of them overturns it.

The first concerns what "material is missing" can mean here. The unknowns in these records are written by the question's author, not discovered by the judgments: each question carries unverified items of its own, and the form asks for them. Where the judgments part, they part over the same listed unknowns — whether a delay claim is real, whether a body can be trusted before it exists, whether portable pay will be honoured, whether de-identification keeps enough context — and each dissent writes its reversal on one of them. This study reads that correspondence as the hypothesis showing itself in the record: divergence happens where the setting supplies no answer, and the dissenting judgment names the missing item as its reason. But the same reading carries a limit. Because every question lists what is missing, the hypothesis can only be tested on the divergence side; a question with no divergence is marked not-applicable rather than as evidence that a well-documented setting produces agreement. The verdict therefore describes the divided cases as they stand, not a general law that missing material causes disagreement.

The second observation concerns the count itself. The A2 row counts cases, not items: thirteen divided cases, each listing unknowns, gives the hypothesis its support without telling a reader how many unknowns any one case carries. The ledger's numbers give the total, 148 items across fifty cases, and give no per-case breakdown; this study does not construct one. What can be said is what the records show on their face: the divided questions are questions in which the judgments themselves repeatedly cite the same unverified items as the reason for reversing each other.

The third observation concerns what the verdict cannot reach. H3 says nothing about which side of a division the missing material favours, and nothing about whether the judgments that chose the majority option knew less than the dissent. Several of the lone dissents in 14.2 and the close divisions in 14.3 turn on items the majority also named — the majority simply treated them as risks to be borne rather than as reasons to stop. Divergence and missing material travel together in this ledger; their causal order is not something the counts establish. Whether a disagreement about an unknown is a disagreement about evidence or a disagreement about who should absorb an error is a question the records pose and the ledger does not answer.

A fourth observation follows from the form in which these questions are written. Each question carries unverified items of its own, written before any judgment exists, and each judgment names those items again when it explains its choice, its accepted losses and its conditions. A reader can therefore follow the chain within a single record: the question lists what it cannot verify, the judgments argue over those items, and each judgment's reversal conditions point back at the same items. That chain is visible whether the division is lopsided or close, and whether the dissenting judgment is a named staff perspective or the judgment made alone. What it does not show is counterfactual. The ledger holds no question with the same stakes and fewer unknowns, so nothing here can say what the thirteen judgments would have done with a better documented setting.

A fifth observation concerns how far the support for H3 can be pressed. Read narrowly — is material missing where judgments diverge — the verdict is specific and modest: in every case where the published judgments chose more than one option, the record lists what is missing, and the divided judgments name those items as their reasons. Read as a causal claim, that missing material produces disagreement, the verdict cannot bear the weight, because the questions that produced agreement also list missing material and are marked not-applicable rather than as evidence for the other side. The result is a consistency between two cells of the ledger in the cases where both apply, not a law about how disagreement arises.

A sixth observation concerns confidence. The judgments in these cases set their confidence at medium or low and name what they do not know, so a division is not a case where one side claims certainty and the other does not. Both sides write conditions under which they would return, and both write stop conditions for their own choice. The difference between them lies in the losses they accept and in the direction in which they would reverse, and that difference is recorded in the same fields on both sides. A reader comparing the two can therefore compare like with like: the same form, the same option list, the same question about loss allocation, and the same demand to state what would change the answer.

A seventh observation concerns the reader's use of this chapter. The records show that a division is published rather than smoothed away: each side states its decision, its accepted losses, its conditions for action, its stop conditions and its confidence, and the missing items are named on the face of the question. That is what makes a parting readable at all. It does not supply a verdict on the merits, a ranking of the readings, or a prediction about which option would have worked. Where this chapter has described the shape of a dissent, it has described what the judgments say they chose and why; the choice between them remains with the decision actors in the settings, and with the reader who has to decide something of their own.

The eight hypotheses are set out in Chapter 2, and their verdicts are taken up in Chapter 15.