Chapter 1. Deadlines, Owners and Reopening in Sixteen Authored 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 lodges each protective decision in a single office and sets its deadline on the administration's own operational clock, which places it, in the crisis-management debate, firmly on the side of action close to the ground. The third shows that the set leaves the decisive unknown, in many cases, with the applicant, and writes the way back as a visible failure without a threshold, which aligns it with the incremental and adaptive view of revision while leaving the cost of waiting unnamed.

1.1Data and method

AEGIS publishes its authored cases as data files of one record a line; the eleven published parts held 6,540 records when they were read for this study. The cases studied here come from one programme within them, a set of 100 authored settings that AEGIS calls X100 and describes in each record in the same words:

The selection rule is one sentence: every case of the X100 set that AEGIS files under civilian protection or under disaster support. Eight cases fall under each heading, sixteen in all. The two headings were chosen because they are the set's protective domains, the ones in which an office acts to shield applicants from harm; the other headings of X100 (procurement, budget, alliances and the like) were left aside. The rule takes every case that meets it, so no case was chosen for what it says.

All sixteen records are synthetic, and they say so in several places. Each is typed "realistic synthetic"; each source line says the setting is not a real country, event or law; none has a source address; and each carries the same statement of its role:

The paper therefore treats the cases as authored descriptions of decisions, not as evidence of what offices do. A further property follows: no case carries a judgment. All sixteen records say that the overall judgment is not completed. Twelve wait for input; four carry a small bounded calculation, each fenced by a sentence saying what it does not show. The set frames decisions; it does not make them.

AEGIS wrote the records in Korean. This study put them into English itself, once, sentence by sentence and without additions, and each case's English seed is labelled as the study's translation and names the Korean original by its hash. Every block quote in the paper is that English, word for word.

From each record the study read the parts that answer its question: the decision owner and the actor (who), the deadline (by when), the question and the three choices (what), the needed inputs and the single unknown (what is missing), and the condition on which to reconsider together with what to observe (what would reopen it). It also read the context, the bounded calculations and the links the records draw to other cases. Every number in the paper was counted by a script from the study's own copies of the records and is listed in its numbers file; nothing was coded by a model.

Because a studio's records can repeat a template, the study measured how much of each part is template before leaning on it. The measure is the one the RISCON series uses: break a text into runs of five words, and count a run as frame when it recurs in at least 5% of the records that fill the same part, and in at least three. Across the authored records that fill each part (6,530 to 6,540 of them), the frame share was 0% for the deadline, the owner, the unknown, the condition for reconsidering and the question, and 4% for what to observe. Because Korean attaches particles to words, runs recur less in Korean and this measure is a floor; the study therefore measured again within the X100 set and within its own sixteen cases, in Korean and in English. The decision-bearing parts stayed at 0% throughout. The links between cases did not. Their counter-evidence sentences were 69% frame in Korean and 76% in English, their conditions 30% and 43%, and their time ranges 35% and 65%; the statement of source role was 100% frame. Those parts are the studio's formula, and the paper uses none of them as a finding. One more thing is uniform without being measured as frame: every condition for reconsidering in X100 is a sentence that begins with "when". Across all authored records only 2% of such conditions take that form, so the form belongs to this set's design, while the words inside it are each case's own. Finally, the presence of a reopening condition, of three choices and of three needed inputs is guaranteed by the template in all sixteen records, and the paper treats none of these as a discovery. What the paper reads is what the conditions, deadlines, owners and unknowns say.

1.2Who decides, and by when

Every one of the sixteen records puts the decision in one office, and the office is cut to the shape of the problem. The owner of the relief decision after an outage is an application relief lead; the owner of the decision about a bounced notice is a procedural protection officer; the owner of the decision about an ended programme's bills is a programme close-out finance officer. Sixteen cases have sixteen different owners, each a title ending in one of four office words (lead, officer, coordinator, manager), and in every case the owner and the actor are the same. No case gives the decision to a committee, a minister or a political leader, and none names a person.

This is a choice of framing, and the crisis-management literature helps show what it includes and excludes. Boin, 't Hart, Stern and Sundelius write of leaders who are expected to minimise a crisis's impact while others look for someone to blame, and who must restore a sense of normality. The AEGIS records contain none of that politics. Their owners are not leaders in that sense; they are the offices that hold the procedure in question. The records sit closer to Col's local government, which, in her account of communities left on their own for hours or days, protects citizens by taking the initiative in risk assessment, decision making and communication. What the records add, on this paper's reading, is a further narrowing: the decision is lodged not with a local government but with one desk inside it, the desk whose own procedure produced the risk.

The case of the disbanded support team shows how far the set carries this. The protective task there is ownership itself:

The record asks how the ownership of these enquiries, and the promises to answer them, should be handed over. The first of its three choices repeats, at the level of each enquiry, what the record does at the level of the decision:

An owner and a deadline: this paper reads the choice as treating these two things as what makes a commitment to an applicant hold, and reads the set's own design as that sentence applied to itself.

The deadlines say as much about the set as the owners do. All sixteen cases have one, all sixteen are worded differently, and none names a date. Each is counted from an event in the administration's own working time: the close of support applications, a payment decision, the publication of a map, the next batch of notices, the first appeal review, the end of a supplement period, the publication of a waiting list, the resumption of a queue. Ten are counted in days, five in hours, and one at a clock hour. The shortest span is 2 hours; the longest is 168 hours, a week.

Read closely, a deadline in these records falls just before the point after which the harm the decision addresses would be fixed. The case of the results map is the plainest:

Once the map is published, the exposure the record describes could no longer be prevented, and the deadline falls before that point. One of the links the record draws makes the same point in a sentence of its own rather than the studio's formula:

The bounced notice has the same shape, and its record repays close reading, because the deadline it sets is a deadline about a deadline:

The record's question is from what moment a notice should count as given when its delivery has not been confirmed, and the office must answer it by this point:

The applicant's clock is already running, set by a machine from the moment of sending. The office's clock is set to stop two days before the applicant's runs out. The record thus holds two times at once: the applicant's, which the procedure imposes, and the office's, which the record imposes on the office. The warning literature describes the point at which a message must be received, attended to and understood before anyone can act on it; Lindell and Perry place it at the start of the protective decision, and Mileti and Peek hold that what people do depends on the substance of a warning and the course by which it reaches them. In this record the message has not reached the applicant at all, and the protective decision belongs to the sender. This paper reads the case as the warning problem turned around: the office must act on the possibility that its own warning failed.

The relief case after an outage pushes the deadline past the deadline it repairs:

Here the application deadline has already passed for those the outage stopped, and the decision has a window running to the next evening.

Two deadlines in the set run the other way, and both belong to disaster support. The case of the temporary recovery copies sets its deadline as a check a week after the switch back to normal operation, and the case of the temporary link sets it two days after normal service resumes:

These are deadlines of closing down. The emergency measure was the protective action; what is left is to stop it doing harm after its time. This paper reads this as close to what Boin and his co-authors describe as establishing a sense of normality after a crisis, but the records place it with an officer and a date on the recovery calendar rather than with leadership.

Taken together, the owners and deadlines answer the first half of the paper's question in a consistent way. In this record set, a protective decision belongs to the single office that runs the procedure at risk, and it is due shortly before the next operational step would make the harm permanent, or shortly after recovery, when an emergency measure would otherwise outlive its use. That is the set's contribution to the debate on where protective decisions are lodged: it pushes them down past the local government to the desk, and it ties them to the desk's own calendar.

1.3What is unknown, and what reopens

Each record names exactly one thing that is unknown at the deadline. In six of the sixteen, the unknown concerns the applicant: what the person did, meant or understood. Four of these six are in the civilian-protection cases and two in disaster support. The proxy applicant shows the pattern:

The rejection notice places the unknown in the reader's understanding:

and the restored bookings place it in an intention that left no trace:

The office has to decide by its deadline without access to a fact that bears on the matter, because that fact lives with the applicant. Lindell and Perry's model suggests the same point from the other side: before further processing, a warning must be received, attended to and comprehended. On this paper's reading, that sequence on the applicant's side is what the office cannot observe in time. The remaining unknowns lie inside the administration (copies outside an inventory, payments never registered, the time it will take to find a cause) or between offices; they are the kind of unknown an office could in principle find out by looking harder. The six on the applicant's side are of a different kind, and this paper reads their presence as the set's most distinctive feature: a protective decision in these records is often a decision about a person the office cannot yet hear from.

What the records offer an office facing such an unknown is a choice of three, and in five of the sixteen cases one of the three is a way to hold. The proxy applicant offers "Withhold the decision"; the bounced notice offers to withhold adverse consequences in disputed cases; repeated resubmissions offer to hold doubtful cases for confirmation; the restored bookings offer to suspend the booking-change function for a time; the temporary link offers withdrawal after a grace period. Sunstein's argument in Laws of Fear is that risks lie on all sides of a situation and that precautionary steps create dangers of their own. The records do not settle that argument, and they do not rank the hold above the other choices. On this paper's reading they are nonetheless consistent with it: the hold is one option among three, never the frame of the whole decision, and most cases offer no hold at all. What the records do not do is say what a hold would cost, or who would bear it.

The conditions for reopening complete the picture. Every one of the sixteen is a sentence beginning with "when", and every one names the moment the harm the decision was meant to prevent would come into view. For the outage:

For the bounced notice:

For the disbanded team:

Each condition is the failure itself, described as an observable event. Each record pairs it with something to watch, so that the failure can be seen if it happens; for the bounced notice, the record asks the office to keep three things apart:

This is close to what Haasnoot, Kwakkel, Walker and ter Maat call a signpost: something monitored to see whether a plan needs to adapt. It is also close to Lindblom's administrator, who expects to reach goals only partly and to repeat the comparison as conditions change. The records build that expectation into every decision. Where they part from the adaptive literature is in precision. None of the sixteen conditions names a number, a share or a threshold. The case of the backlog after resumption shows the gap clearly, because it has a number to hand. Its bounded calculation reads:

Its condition for reopening reads:

The calculation gives a daily net fall under stated assumptions, but the condition uses no such figure. "Keeps exceeding" leaves open for how long, and by how much, before a new look is due. This paper reads the absence as a choice that suits authored exercise cases, in which no real flows exist to set a threshold against, rather than as an oversight; but it means the reopening conditions describe the shape of a failure without saying when it has arrived.

Two further absences bear on the debate. The first concerns learning. Kahneman and Klein conclude that the likely quality of an intuitive judgment depends on how predictable the environment is and on whether the judge has had the chance to learn its regularities. Every AEGIS record asks the office to record what happens after the decision, which is the raw material of that chance; and every one of the sixteen records says, of its own predictive performance, that it was not measured. The records set up the conditions for learning and stop there. The second concerns the waiting. Moynihan, Herd and Harvey describe the costs that citizens bear in dealing with the state, and argue that how those costs are shared between state and citizen is often a political choice. In several of the AEGIS scenes the one in the interval between a decision and its possible revision is an applicant: the person whose support waits for a confirmation, whose supplement period runs while a notice bounces, whose booking hangs on a restored copy. The records name the owner of the decision and the time it is due. They do not say who bears the delay.

These findings answer the second half of the paper's question. In this record set, what is unknown at the deadline is, in six of the sixteen cases, something only the applicant could say, and what would reopen the decision is the visible arrival of the harm it was meant to prevent, watched for through an act of recording but measured against no threshold. That is the set's contribution to the debate on revision: it takes the adaptive view, that every decision should carry its own way back, and applies it uniformly, while leaving the cost of the interval unassigned.