Chapter 1. What Ties Two Decisions in Twelve Authored Training Cases

This chapter answers the paper's question from the records. Its first section says what the records are and how they were read. The second shows that the set ties every pair of decisions by one named shared resource and refuses every other tie, which places it with the view that coordination begins from a dependency of a stated kind and against the picture of decisions joined by resemblance or timing. The third shows that no record names who decides the shared resource, which bears on the debate between a comprehensive centre and self-coordination among the parties.

1.1Data and method

AEGIS publishes its authored cases as data files of one record a line; eleven published parts held 6,540 records when they were read for this study, by GET requests only. A snapshot of each of the twelve cases and a reduced copy of the whole population are kept with the study, without the source line and the scenario group, two fields that can carry the name of a software model.

The selection rule is one sentence: every authored AEGIS case that fills the field linkEvidenceReason, in which a case states why it is linked to other cases. Twelve cases fill it. All twelve carry the link policy evidence-only, and all twelve belong to one programme, AEGIS-RVE-20260921. The rule takes every case that meets it, so no case was chosen for what it says. The programme holds 22 cases; the other ten fill instead the field linkAbsenceReason, which states why a case declares no link, and nine of those ten reasons refer to the original text from which the programme was corrected. Across the whole population, 460 cases give such a reason for declaring no link, and 470 have no link at all.

The programme describes itself in every record in the same words:

The twelve fall under five of AEGIS's domain headings: five under logistics and supply chains, three under alliances and diplomacy, two under Pacific and maritime strategy, and one each under disclosure and accountability and under maintenance and life cycle. All twelve are synthetic and say so in several places. Each is typed realistic synthetic; none has a source address; each says that its question and judgment mark have not been compared with public records, and that its fact check verified no real event or authority; the name of each deciding body begins with the word fictional; and each states its role in the same words:

No record carries a completed decision. Each holds three choices, every one of which names its cost after a dash, and an author's judgment mark that recommends one choice, lists the other two as alternatives, and says on what the recommendation would be reversed. Every record waits for input, and every record says its predictive performance was not measured. These features are the programme's design, the same in every case, and this paper does not count them as findings.

A link is stored in two places. A case's list of relationships holds one entry for each linked case, with a type, a direction, a reason, a condition for review, a time range, a status and a counter-evidence line. The twelve cases also carry the single field linkEvidenceReason, which gathers the reasons for all of a case's links and closes with a statement of policy. Beside these the paper reads the decisionOwner, the deadline (the moment before which the decision must be made), the neededInputs (three in every case), the reconsider condition, the author's judgment mark (review), and the context, which in these twelve closes with a list of the cases it touches. Where a count classifies wording, scripts/count.mjs states its rule beside the count, a rule on the Korean words of the original records. Every number in this paper comes from that script and stands in numbers.md.

A sentence repeated across a studio's records is the studio's way of writing, not a finding about any case. Following the measure used across RISCON's studies, a five-word run counts as the frame's when it stands in at least 5% of the population's filled items, and a field's frame ratio is the share of its runs that are the frame's. Korean joins particles to words, so the measure on the Korean records is a floor. A field whose ratio is 30% or more is treated here as frame, not as evidence. Against all authored cases, the link condition, the link counter-evidence line and the link time range are frame: 99%, 99% and 98% of their runs. The linkEvidenceReason of the twelve is 45% frame, because its closing statement of policy is the same in all of them. The link reasons themselves, the contexts, the deadlines, the needed inputs and the reconsider conditions each stand between 0% and 3%. The paper therefore reads each link's own reason, and the record around it, as the case's own; it reads the link's type, condition, counter-evidence and time range as the studio's frame, and says so wherever it quotes them.

AEGIS wrote the records in Korean. The English quoted here is this study's own translation, made once, sentence by sentence, and labelled in each case's English seed with the hash of the Korean original; sums the Korean writes in units of a hundred million won appear in billions of won.

1.2One named resource, and no other tie

Every link in the set rests on one named resource that the two decisions are said to share, and every record refuses to be tied by anything else. The twelve records hold fourteen links among them; each points to a case that is also among the twelve, and in every one of the seven pairs that result, each record names the other. Ten records belong to one pair and two belong to two, so that the set forms two short chains of three cases and three separate pairs, and nothing in it reaches outside.

The pair that shows the form most plainly joins a decision about keeping a stockpile against a communications cut to a decision about whether emergency materiel goes by air or by sea. Neither question mentions the other. The first asks whether predictive supply should be expanded, whether stockpiling of core items should take priority, or whether both should be kept; the second asks whether transport should stay centred on air, shift to sea, or be divided by the delay each item can bear. What ties them is stated in one place, identically in both records:

The tie is not a topic, though both cases are about supply, and not a likeness of reasoning, though both weigh a cheaper normal arrangement against a costlier reserve. It is a stock of hours that both decisions would draw on, one to move stockpile and supply, the other to fly emergency loads, named as an item of the original and given its size.

The other six pairs follow the same pattern with different resources. Four share a budget, two share a stock of hours and one shares a system. Two of the seven give the shared resource a figure (the 4,200 hours above, and 140 billion won for allied joint production and cooperation); the rest name it without one. Each reason begins by saying that the two cases share a named item of the original. The verbs that follow vary in strength. Some reasons ask only that the two uses be considered side by side; others state the dependency in its strict sense. The pair on parts and consumables is the clearest of the second kind:

A decision to procure parts near the field and a decision on how far to cut the consumables budget are not alike in any way the records state; whichever way one goes, the money left for the other changes. The same structure appears where the shared item is a budget whose division is left open:

and where the shared item is a stock of time whose total is not yet known:

The second ties a decision about a carrier budget to a decision about procuring more supply ships, through the operating time at sea that both would claim, and in the same breath admits that the size of the pool is not known. The tie is asserted; its measure is awaited. In the information pair, the shared item is neither money nor hours but a system and the staff who run it:

Read beside Malone and Crowston, The Interdisciplinary Study of Coordination (1994), the set records a dependency of one kind only. Their framework separates shared resources from producer and consumer relationships, from simultaneity constraints and from the dependencies of tasks on subtasks, and treats each as calling for its own process of coordination. Every tie in these twelve records is of the first kind. None says that one decision produces what the other consumes, that the two must happen at once, or that one is a part of the other. This study reads that restriction as deliberate, because the template around every link denies it any direction of precedence or cause. A set that admits only shared resources, and denies order, has named the dependency and left the process for managing it unstated.

The refusal of every other tie is written into each record in the same sentence, which closes each record's statement of its links:

The sentence is the frame within the set; it is the reason the field reaches 45% in the frame measure. It still says two things that shape how the twelve should be read. The ground of a link is the stated resource and nothing else. And the lack of confirmation is reason enough not to link: where the original did not establish that two items drew on the same budget, time or staff, the record does not infer it. The ten other cases of the programme apply the same rule from the other side, each stating why it declares no link, and nine of them refer to the original text in their reasons.

The rarity of this form shows when the twelve are set against the whole studio. AEGIS's authored cases hold 16,255 links. Of their reasons, 8,043 say that the two cases share something, but 6,069 of those say that the cases share a structure of judgment, a way of reasoning about a kind of question. Only 212 say that the cases share an allocation, a budget, hours or staff, and these are spread over 15 programmes. Only the twelve state their links' grounds in a field of their own, and only their fourteen links begin by naming an item of the original. Across AEGIS, then, the usual tie between two decisions is a likeness of reasoning; the tie by a named resource is a small exception, and these twelve are the place where it is written most explicitly.

That history explains a misfit. Every one of the fourteen links is typed, in the studio's template, as a comparison of judgment structure, and every one is called a non-directional comparison with no direction of precedence or cause. Each also carries the template's counter-evidence line, which stands in 99% of AEGIS's links and belongs to the studio's frame rather than to any case:

The first sentence fits the twelve well: they tie by a named resource, never by resemblance. The second does not fit them at all. In each of the seven pairs, the two records name different owners and different deadlines, so a literal reading of the template would void every resource link in the set. The template was written for comparisons of structure, where a comparison between cases with different owners and deadlines may well be weak. A shared resource is the opposite case: it ties decisions precisely because they belong to different owners who draw on the same thing. The set's own content and its borrowed frame part company at exactly the point this paper studies.

Read against Cohen, March and Olsen, A Garbage Can Model of Organizational Choice (1972), the twelve take a clear position. Their model describes organisations in which choices look for problems and solutions look for issues, and this study reads it as a picture in which choices, problems and solutions can meet by availability as much as by reason. The twelve records allow no such meeting: a tie must be stated, grounded in a named resource of the original, and confined to that resource. This study reads that discipline as the set's main contribution to the question of what ties decisions. It also leaves the set with a question it does not answer, which the next section takes up: once the shared resource is named, who decides how it is divided?

1.3Who decides what two decisions share

No record names who decides the resource that its pair shares. The twelve records name twelve different deciding bodies, each a fictional committee, council or board; within every pair the two owners differ, and the two deadlines differ. None of the fourteen link reasons names a committee, a council or a board (a search of the Korean reasons for the words for these). The resource is the one thing each pair has in common, and it is the one thing in the pair that no record assigns.

The pair joined by aircraft hours shows how close the records come. The committee that decides on stockpiling sets its deadline here:

and the council that decides on emergency transport sets its deadline here:

Both records place their decision before an allocation of the same hours. Each treats that allocation as a moment its own decision must precede, and neither says who makes it. The allocation is the clock of both decisions and the property of neither.

The same pattern holds when the records turn to what must be known. Five of the twelve list among their three needed inputs an allocation of the shared resource. The council on emergency transport needs, besides the dates and causes of breaches,

and the board on exercises and cooperation talks needs

The allocation appears as something to be known before deciding, not as something either decision makes. An approval table presupposes someone who approves; the record asks for the table and leaves the approver out.

Three records go further when they say on what condition their decision would be reconsidered, and they diverge. The committee on stockpiling writes:

The board on exercises and cooperation talks writes in the same spirit:

Neither says with whom the approval is joint. The council on emergency transport, the stockpiling committee's partner, writes something different:

This is the nearest the set comes to naming an allocator, and it names the case's own council, which is to approve the allocation once it has checked the effect on other claims. In one pair, then, one record says the approval is joint and the other gives it to its own council. The two records of the pair do not contradict each other in words, but they do not agree on who holds the pool.

Taken across the twelve, the shared resource reaches beyond the link in seven records. These counts rest on rules that search the Korean fields for words such as allocation, joint deliberation and joint approval. By these rules, six name an allocation or a joint deliberation in their deadline, five name an allocation among their needed inputs, three speak of the shared resource in their condition for reconsidering, and one mentions room left in the allocation in its author's mark. In five records the shared resource appears nowhere but in the link and in the list of touched cases that closes the situation. The pair on parts and consumables is one of these on both sides. Its two deadlines speak only of the case's own contracts:

These two records are the ones that state the dependency most strictly, since each choice changes the allocation of the same funds. Yet neither carries that dependency into the moment of decision, into what it needs to know, or into the condition on which it would change course. The tie is declared once and then left at the edge of the record. Two of the other three records of the five sit in pairs where the shared total itself is open: the minerals budget is to be compared without the amounts being set arbitrarily, and the maritime operating time is awaiting input.

The literature gives these findings their weight. Von Hagen and Harden, Budget Processes and Commitment to Fiscal Discipline (1995), place the bias towards excessive spending in the externality of a common pool, and the remedy in structures that strengthen a comprehensive view of the budget over the particularistic view of spending ministers. This study reads the twelve records as an exact image of the particularistic side of that model: each deciding body sees its own claim on the pool, and the records name the pool with care, but no record holds the comprehensive view or names the body that would.

Scharpf, Games Real Actors Could Play (1994), offers another reading. Negative coordination, in his account, holds back the externalities that one party's choice imposes on others, and arises, in his account, where negotiations are embedded in a structural context, within hierarchical organisations or within networks of cooperative relationships. The council on emergency transport, which will approve its allocation only after confirming the effect on the other claims on the same hours, writes something close to a rule of negative coordination into its own record. This study reads that sentence, and the instruction elsewhere that a budget be compared without setting its amounts arbitrarily, as the set's nearest approach to the self-coordination Scharpf describes. Positive coordination, a search for the best joint use of the shared hours or budget, appears nowhere in the twelve.

Ostrom, Governing the Commons (1990), widens the frame: if common-pool problems are sometimes solved by voluntary organisations rather than by the state, a record set that leaves the allocator unnamed has not necessarily left the pool ungoverned; it has left open what would govern it. Kettl, Managing Boundaries in American Administration (2006), names resources and responsibility as two of the five boundaries that have historically mattered. Read through that distinction, the twelve draw the boundary of resources across cases, by naming what two decisions share, while they keep the boundary of responsibility inside each case, by naming one owner for each decision and none for what lies between.

Together, the two case sections answer the research question in three parts. What ties two decisions, in this record set, is one named resource and nothing else: a dependency of a single kind in Malone and Crowston's terms, stated with figures in two pairs and without them in five. What the set refuses to tie is everything else, including the resemblances of reasoning that tie most of AEGIS's other cases, and it refuses with a template that was written for those other ties and that, taken at its word, would undo its own. Who decides what two decisions share is not said: each record keeps its own owner and its own deadline, the allocation of the shared resource appears as a clock, a needed input or a joint approval without a partner, and in five records the resource does not appear beyond the link at all. The twelve are synthetic, so none of this describes how any real body decides; it describes how one designed record set places, and does not place, the coordination of what decisions share.