Chapter 11 sorted the ten restraints by form: where each one comes from, who can see the boundary it protects, how far it stops, and how firm the stopping is. It named cost without tabulating it, and deferred the relation between firmness and the cost of reversal to this chapter. This chapter takes up that question. It does not ask whether a restraint is good, and it does not decide who in any scene was right. It asks what holding costs, in which direction that cost travels, and whose hands it lands in. Two questions are put to every restraint in the records: who waits while it holds, and who works again after it has held.
12.1Two Costs That Are Confused
A control that stops work has two costs, and ordinary speech runs them together. The first is the cost of stopping: the time and attention lost while the restraint holds — the queue behind the gate, the explanation owed to someone blocked, the day that ends without an answer. The second is the cost of reversing: the work of undoing what has already been done — work-order information that a label would carry if it were printed before verification, a transfer that would have to be renegotiated if made before the approver's schedule was known, a delivery promise that would have to be withdrawn if its conditions were not verified. These are not the same cost; the distinction is this chapter's own, drawn from the records rather than named by them.
They are borne by different people at different moments. The cost of stopping is paid while the restraint holds, in waiting, by the layer that cannot lift it: the journal-entry lock whose reason no one remembers is felt by the closing-materials user, and neither keeping the lock nor removing it has a sufficient basis. The cost of reversing is paid after a step that cannot simply be undone, in rework or in obligations to a counterparty. Work-order information printed on a label cannot be lifted back into the file it came from; the label judgment reasons that once printed, disposal and reissuance and mix-up management follow, though in this scene no label was printed. In the label case and the transfer case alike, the step contemplated becomes part of the world rather than part of the file. Neither cost can be measured here, because the records contain no hours, quantities or amounts.
Herbert A. Simon's account of administrative life gives one reason these costs stay apart. Decisions in an organisation are drawn from premises, and the organisation shapes later decisions by setting those premises. James G. March and Herbert A. Simon, Organizations (1958), describe programs in the sense of stored routines — routines that a stimulus calls up and that settle what to do without fresh deliberation — and such a program can keep working after the people who wrote it have gone, so that a hold is felt not as a decision but as an obstruction. The application of that point to the hold is this study's. The cost is real but diffuse, and it surfaces in somebody else's waiting rather than in the account of whoever set the program. Charles Perrow's point about interactive complexity and tight coupling is that the system stopping is the design behaving as designed. Perrow does not say that such a stop is cheap. The point here is that the cost of stopping and the cost of removing a safety device are separate quantities that no single description of "the cost of the control" can carry.
In each of the two scenes examined here, the cost-of-rollback lens names an option first, and the option has the same shape: to link what happens after a decision back to the decision itself. The first is the journal-entry lock.
Link Follow-On Waiting, Rework, and Persons in Charge to the Decision Record
The gain this option offers is comparison rather than blame: it puts what happened afterwards next to what was chosen.
Outcomes and follow-on costs can be compared in the same choice.
The risk the same lens names is that the comparison curdles into something else:
If records turn into a per-person penalty table, the site can hide costs.
The second scene is the goods returned at a gate for lack of scheduled information; the same shape appears in that scene's own phrasing.
Link the follow-up waiting time and rework in the decision record with the person in charge
Results and follow-up cost can be compared under the same option.
If the record turns into an individual demerit board, the field can hide the cost.
Now the strongest objection to keeping the two costs apart, stated at full strength: they are one burden recorded twice. A stop that had been lifted promptly would leave nothing to reverse, so the reversal cost is the stopping cost delayed, and dividing it in two lets an organisation write down the decider's share and drop the cleaner's, or re-label the cleaner's share as diligence and the decider's share as design. The distinction, on this objection, does not describe the world; it gives the layer that stopped a complete-looking account. The records meet the objection from inside. The risk sentence quoted above names exactly the instrument that would deliver such a account — a record that turns into a per-person penalty table, after which a site hides its cost — and the alternative the lenses give, of evaluating only by the processing speed of the near layer, is refused in the same judgement as a mixing of two things: the workload created by the decision, and the capacity of the field to perform. The objection is therefore not answered by abandoning the distinction; it is answered by refusing both ways of escaping it.
Two supports stand beside the records. First, the ledger's own cell on this question. Of the 186 cases, the two codings agreed in 115 (62%); among the settled values, applying a named removal condition rests with the holder or a third party in 81 cases, in all but one with at least one coding inferring it, while 14 cases do not say, 3 name those who wait, 17 are not applicable and 71 are unsettled. The cost of waiting therefore falls on people who cannot lift the restraint themselves, and the cost of reversing falls on people who were not asked to choose. Second, Michael Lipsky's Street-Level Bureaucracy, whose reading note says that front-line workers make policy through the discretion they use under rules and scarce resources, and that the people they serve bear its cost. Lipsky meets this distinction as an argument about who acts after a decision: the near layer is not merely waiting but choosing how the decision is applied, so the reversal cost is a second decision-site rather than a by-product. Lipsky does not meet the records as a measure of cost, because his costs fall on those served while these lenses place the follow-on work on the working layer itself, and name no served party.
The two costs are separable only if both are recorded against the decision that produced them; separated and separately recorded, the distinction serves the account of the layer that stopped. The second question — who works again — cannot be answered by those who wait unless the condition and the hands that hold it are visible to them, which the ledger's cells seldom show them to be: across the library the record mostly does not say, and the codings agree that those who wait can see the boundary in 5 scenes.
Three things should be said about this material. These lenses are hypothetical: the condition attached to them supposes that the person who decided and the person who does the aftermath are different people, and nothing in the scenes reports that as something which happened. The records contain no numbers, so this chapter can name the direction of a cost and the layer that carries it, and cannot measure either. And the two questions this chapter asks are asked because the records ask them. "Who waits" is the question of the cost of stopping. "Who works again" is the question of the cost of reversing. Every restraint in the ten scenes can be described by putting both answers in the same sentence, and in eight of the ten scenes the same layer carries both answers.
12.2The Condition the Records Name
Across the cost lenses, the records name a single condition under which follow-on cost becomes visible at all: the person who decides and the person who cleans up are different, and the outcome of a decision goes to one of them while the explanation and the rework go to another. This condition is hypothetical in every lens it appears in. It is restated here in that status, not as a finding about any workplace, and the restatement in each case below carries the same mark.
The first scene is the automatic intake that produced a status but no split order form, leaving exchange dispatch stopped, read in Chapter 8. The roles in the lens are an automated process designer and an agent who explains the delay to a customer. The scene's cost lens names the same option, gain and risk as the two scenes of section 12.1, in its own wording; the risk it names is a per-person demerit board.
The condition behind the lens is hypothetical, and it is also specific: it supposes that the credit for the decision is assigned to the designer while the additional explanation and rework are assigned to the agent. Note what that does to the record's own caution in this scene, which Chapter 8 quoted: the studio points out that shipping delay can have causes other than intake logic. If the delay has a second cause, the agent explaining it absorbs that cause as well, and the design question stays where it was.
The second scene is the employee who stays at a desk after call forwarding has been set, read in Chapter 10. Its cost lens again names the linking option, its gain and its risk, here an individual penalty sheet.
Here the hypothetical condition is of a particular kind. It supposes that the outcome of the decision flows to the manager who sets the end of work, while the additional explanation and rework flow to the employee who waits. The cost in this scene is not primarily rework. It is time — a person's time held at a desk by an absence rather than by an order. Chapter 10 read that scene with care and refused to conclude that waiting was forced or intended. This chapter adds only a cost question to that reading: if the record never carries the waiting, no one can tell afterwards whether the waiting was necessary, and the question is settled by whoever was not standing there.
The third scene is the month-end transfer whose date falls on a holiday, with an approver about to be absent, read in Chapter 9. The roles in that lens are an approval-request preparer and a fund execution approver, and its cost lens names the same option, gain and risk, the risk here an individual penalty board.
This condition is hypothetical as well, and it attaches to a decision with a third party in it. The decision turns on when a transfer can be made, and on when the approver is available. The outcome flows to the fund execution approver; the extra work of chasing the schedule, and of speaking to whoever is waiting for the money, flows to the preparer. As Chapter 9 noted, the follow-on work here is not only internal.
Max Weber's account of officialdom in "Politics as a Vocation", set out in Chapter 4, gives this condition its name in another vocabulary: the ethic of responsibility requires attending to what a decision causes, while the official who executes an order does so on the responsibility of the superior who gives it and cannot shift that responsibility onto anyone else. This study reads such an arrangement as one in which a decision can travel without the decider travelling with it; that phrasing is this study's, not Weber's. Applied to these three scenes, the point is narrow. If a decision is made in one layer and carried out in another, the layer that made it can record its own reasons without ever meeting the waiting and rework those reasons produced. Whether that happens is a question for the people in the organisation, not for this study; what can be said here is that the arrangement makes the meeting optional.
12.3The Penalty Table and Its Danger
Section 12.1 quoted the option that appears in the cost lenses: to link handover, waiting and rework to the decision record, with the gain of comparing outcomes and follow-on costs inside one choice, and the risk that a record kept for comparison becomes a per-person penalty table, after which a site hides its cost. Section 12.2 found the same shape in three more scenes; the option, its gain and its risk belong to the studio's template for the cost lens, and the five scenes name them in five different wordings. No further quotes are needed here.
There are two alternatives, both present in every cost lens in the records. The first is to evaluate by the speed of the layer nearest the stop alone. Its gain is arithmetic: delays at the processing steps are easy to tally, because they happen in front of a stopwatch. Its risk is that the measurement mixes two things — the workload created by the decision, and the capacity of the field to perform. A site measured on its own throughput is not measured on its share of somebody else's decision. The second is to say that if the outcome was good, the follow-on cost is not worth asking about separately. Its gain is speed of recognition. Its risk is that the organisation loses the chance of producing the same success at a lower cost, and loses the next signal that rework is coming.
Both alternatives have something real behind them, and this chapter does not choose between them. The first is honest about what it can count. The second is honest about what investigation costs: a question about follow-on cost is a question someone must answer, and answering takes time that is itself part of the cost. What they share is that neither produces the comparison the linking option proposes. Under the first, the near layer's numbers stand in for the choice; under the second, the outcome stands in for it. In neither case do the decision and its aftermath appear in the same place.
Now the danger of the linking option. Each of the five scenes names it in its own words: a per-person penalty table, an individual demerit board, a per-person demerit board, an individual penalty sheet, an individual penalty board. The wording matters less than the fact that all five say the same thing. When a record carries outcomes and follow-on costs against named individuals, it is not only a description; it is an instrument, and instruments can be used. Chapter 5 described Han Feizi's two handles — in this study's translation of the passage, "The two handles are punishment and favour", and the English is this study's — as the two levers of power, dangerous when lent to anyone other than the one who rules. A cost record comparing outcomes against individuals is exactly such a handle: whoever can read it can distribute praise and blame through it without writing either down.
The danger runs both ways. If the handle stays with the layer that set the restraint, the field learns to report what that layer rewards, and the record fills with what is safe to say. If it is lent to the layer nearest the stop, the record fills with what is safe to say from the other side, and the decision layer's reasoning becomes the only thing never written down. In both cases the record does the opposite of what it was meant to do. Han Feizi's chapter argues that a ruler must hold the handles rather than lend them; this study extends that argument to instruments of measurement, and the extension cuts against the field as much as against the office, because a site measured by follow-on cost has reason to hide it.
Weber's ethic of responsibility, set out in "Politics as a Vocation", places the answer for consequences on the one who decided — but a consequence answered for afterwards requires something that can be examined afterwards, which is why the record is where answerability is won or lost. That cuts the other way from Han Feizi, and the tension between them is this chapter's real position. Refusing to record is not neutral: if follow-on cost is never written down, the balance of power is not restored, it is left to whoever accounts for events afterwards, and that account favours whoever is already able to speak. But a record that answers only for the near layer's speed inverts the office's own logic, because Weber's ethic asks the decision-maker to answer for consequences rather than measure someone else's pace.
What is lost when follow-on cost is never asked about? Three things, all visible in the records' own alternatives: the possibility of the same success at lower cost, because no one can compare the choice with its aftermath; the signal of rework, because unrecorded rework is absorbed rework; and the organisation's only observable half of a comparison it cannot otherwise complete, a point this chapter returns to in the next section. A fourth thing is lost, less visibly: the layer that absorbs cost silently learns that cost is absorbed silently. The cost does not vanish; it moves into people's days, where no one reads it.
12.4Luck, Outcomes, and Credit
Every one of the ten scenes asks a version of the same question: if the outcome was good, was the choice right? The records answer it the same way in each scene, and this chapter quotes the answer once, from the journal-entry lock, where the sixth of the scene's eight lenses, the Outcome and Luck lens, reads as follows.
A good outcome is an important observation, but it is not a certificate for the entire judgment process.
When "while looking for the cause, record the behavior with a limited test" was chosen, do not write it as if it had actually been observed, but leave it as conditions to compare in the next similar case.
The scene of the employee who waits says the first of these in nearly the same words, and so does every other scene: the sentences belong to the studio's template for this lens. They carry three separations, and the separations are what makes the cost question answerable. The first is the basis at the time: what the decision-maker actually knew, in particular the missing element each scene names — the transaction boundary the lock was protecting, the reason goods and scheduled information diverged, the final automated output, the actual work that requires staying, the payment deadline and the approval-available time. The second is the external conditions: what happened afterwards, whether or not by chance, that had nothing to do with the choice. The third is the follow-on cost: who worked again, and what that work consisted of.
The sentence this study quotes from Sun Tzu's The Art of War, quoted in full in Chapter 5, is used here about a button, a lock and a calendar, so it should be taken narrowly. Its narrow point is that knowing both sides means one need not fear the result of a hundred battles: the passage says, "If you know the enemy and know yourself, you need not fear the result of a hundred battles". This study reads the passage, applied here, as saying that a good outcome cannot stand in for the knowledge a choice rested on; that extension is this study's. Applied to the records, a good outcome produces a story about the choice, and the story may be wrong, because the choice may have been made without the knowledge the story credits to it. The third quoted sentence says the same thing in operational terms: when an option was not chosen, the organisation has nothing to observe about it, so it must write down what would count as evidence in the next similar case.
Charles Perrow's argument that some accidents are normal to a system rather than caused by bad people gives the second separation its sharpest form. In a tightly coupled system with interactive complexity, outcomes are properties of the design as much as of anyone's decision. A bad outcome may be what such a design produces under ordinary conditions, and a good outcome may be what such a design produced this time because the conditions were favourable. Perrow does not say that decisions do not matter. This study draws from that argument that a design produces outcomes under ordinary conditions, and so that reading outcomes backwards into decisions misdescribes the system. Chapter 4's point stands: a system that stops is doing what it was designed to do, and a system that limps through is not thereby well designed.
This is where the cost question and the luck question meet, and it is the strongest reason the records insist on the third separation. Of the three quantities — basis, conditions, follow-on cost — the follow-on cost is the only one that actually occurred to somebody, although Section 12.3 showed that it is not reliably observed. The basis at the time may have been written down, or may have evaporated into someone's memory; the unchosen option never existed, so there is nothing to observe about it; and the external conditions are not in anyone's control. The follow-on cost, by contrast, is a thing that actually happened to somebody. It is the one piece of evidence about the choice that cannot be reconstructed afterwards from a story.
There is an asymmetry inside this that the records name without solving. In every cost lens, the outcome of the decision is assigned to the decision layer while the explanation and rework are assigned to the near layer. If only outcomes are recorded, the decision layer's account is systematically enriched and the field's account is systematically impoverished. The result is not a conspiracy; it is the ordinary consequence of who is present when the good result arrives. A person who succeeds and then moves on carries the credit; a person who stays and explains carries the cost, and if neither is written down, only the credit is retold. The records' insistence on recording follow-on cost is therefore not primarily a disciplinary measure. It is the correction of an evidential bias that would otherwise make these scenes look better than they are.
12.5The Comparative Cost Map
The ten scenes fall on one map, in two columns: who waits and who works again. This is an observation from the records, not a law, and no cell makes any person in any scene responsible for anything. One caution applies throughout: the first column draws on the conditions attached to the authority-to-stop lenses, which are hypothetical — they state what each lens assumes about who can stop, not what any scene reports as having happened.
| Scene | Who waits | Who works again |
|---|---|---|
| Journal-Entry Lock | The closing-materials user, who found the problem and cannot stop it without the approval-feature changer's approval | The closing-materials user, who must explain and, if the control is changed, redo work under a revised rule |
| Purchase-Page Image | The purchase-page production owner, waiting on the scope of the change prohibition and the approval status of the request | The production owner, who explains the state of the page and reworks assets |
| Unscheduled Goods Return | The receiving logistics centre, which holds goods that arrived physically without scheduled information | The centre, which identifies the actual goods and answers for them; the field, if an exception agreed elsewhere was not communicated |
| Label Issuance Stop | The label issuance handler, who cannot print until the origin is verified | The handler, who must explain the hold to the next step; no printed label appears in the record |
| Same-Day Delivery Copy | The ad asset registrar, waiting on the applicable scope of the withdrawal | The registrar, who explains the withdrawal of the delivery promise; the records report no re-registration |
| Option Sold-Out | The sales-availability notifier, while the range of the effect is verified | The notifier, who explains what was stopped and what was not |
| Unsplit Order Form | The agent explaining customer delay, whose customers wait while dispatch does not begin | The agent, who explains the delay; the records report no fixed flow |
| Absent Delegate | The absent person in charge, whose absence is the reason the argument to defer meets the argument to set an outline | Whoever must ratify, correct or absorb an outline that has already committed something |
| Hesitating to Leave | The employee who waits, whose own time is the cost | Nobody in the record; the cost is absorbed as waiting rather than as rework |
| Holiday Transfer | The approval-request preparer, waiting on the approver's availability, and whoever is waiting for the money, including the payee, while a calendar and an absence collide | The approval-request preparer, who reworks the schedule and renegotiates the timing |
Read down the second column, one observation stands out. In eight of the ten scenes, the layer nearest the stop is both the layer that waits and the layer that explains. The cost lenses place the waiting on that layer, not on the decision layer, whose recorded cost is the outcome and whose follow-on cost appears only when the two layers are compared. The cost that becomes visible to somebody — the queue, the apology, the desk at the end of the day — is not visible at the moment of decision, because at that moment the decision layer is not standing in the queue.
The two scenes that do not fit neatly are the two where the cost is not rework at all. In the waiting scene the cost is a person's time, held by an absence rather than by an order; in the meeting scene the cost is a suspension, and who cleans up depends on whether the outline has already committed somebody outside the room. Chapter 10 read both; this chapter adds only that cost in an organisation is not only the work after a decision. It is also the time before one.
The counter-finding comes from the same records and must be set beside the observation. Not every delay can be attributed to the person who decided. The cost lenses say so in all five scenes examined here: one must distinguish whether the missing element was verifiable at the time from circumstances that arose afterwards. One scene goes further — a shipping delay may have its cause in inventory rather than in the intake logic, in which case fixing the flow does nothing and the agent keeps explaining. A map of who waits and who works again shows the direction of cost, not the source. The direction is reliable across these scenes; the source is not established by the map.
Weber's hierarchy describes the asymmetry in one direction and the records show it running in another. In a hierarchy, answerability runs upward: the official executes and the decision-maker answers for consequences. The map shows consequence running downward, because that is where waiting and explaining are available to be done. Simon's bounded rationality supplies the mechanism: the decider cannot see the near layer's workload, and the near layer cannot see the decider's basis. Each gap looks like inefficiency from the other side, and neither is closed by moving responsibility, because moving responsibility moves the blindness along with it.
12.6What the Chapter Refuses
This chapter ends by naming what it will not do, because the cost question is one that invites conclusions the records cannot support.
It offers no ranking of burdens. A day held at a desk and a payment delayed are not commensurable, and the records contain no hours, quantities or amounts that would allow a comparison to be made. Any statement that one scene's cost exceeds another's would be this study's invention rather than the records' finding.
It reaches no verdict on any person in the scenes. The people in the records appear as roles — an approval-feature changer, a closing-materials user, a receiving logistics centre, an agent explaining customer delay, a fund execution approver — and the readings of those scenes are interpretations, not findings of fact. The counterexample lens in every one of the ten scenes refuses to settle what is inside a person's mind, and this chapter keeps that refusal. Nothing here establishes what any person intended, and nothing here is a judgement on anyone.
It makes no claim that recording cost is always better than not recording it. Section 12.3 showed both sides: a record that compares outcomes against named individuals can hide cost as effectively as it reveals it, and a record that only measures the near layer's speed can misdescribe the choice. The chapter leaves the choice to the organisations that would have to live with either instrument.
It recommends no option. Each cost lens in the records offers three positions with gains and risks attached to each; the linking option, the evaluation-by-speed option and the option of not asking about follow-on cost at all are all presented here with what each gains and what each risks, and none is declared better. The same holds for the scenes as a whole: Chapter 11 sorted them without ranking them, and this chapter costs them without ranking them.
And every reading quoted in this chapter carries a date. Each judgment in ARCHON's record is marked with the date 2026-09-20 and with a status of interpretation that can be corrected; the observations offered in this section — the two-cost distinction, the condition named in the cost lenses, the penalty-table danger, the evidential bias in outcome records, and the map of who waits and who works again — are of the same kind. They are dated readings, correctable by evidence that has not arrived, and this chapter holds to that mark rather than escaping it.
What remains is the question this chapter deliberately leaves behind. Knowing who pays for holding is necessary for any judgement about holding, and it is not sufficient. The tradition this study has been reading puts a different question at its centre: not who carries the cost of a restraint, but when the restraint itself should be taken out of the way. That question has its own conditions, its own evidence and its own refusals, and Chapter 13 takes it up, leaving here all questions of cost.
12.7Who Pays, beside the Study of Organisations
The cost of holding travels in a direction: waiting and rework land on the layer nearest the stop, while the outcome of a decision accrues to the layer that made it. This is an observation and not a law, with two cautions: the condition of a separate decider and a separate cleaner is hypothetical in every lens it appears in, and not every delay can be traced to the person who decided. Four works from the study of organisations sit beside these positions. Each asks a question the records do not ask, and meets them in one place and not in another.
Michael Lipsky's Street-Level Bureaucracy describes front-line workers who make policy through the discretion they use under rules and scarce resources, with the people they serve bearing its cost. This meets the chapter's map directly. In eight of the ten scenes the layer nearest the stop both waits and explains, and Lipsky supplies a vocabulary for what that layer is doing while it waits: not merely executing a decision made elsewhere, but choosing, under constraint, how the decision is applied to the person in front of it. The records come close in the scenes where an agent explains a shipping delay to a customer or a registrar explains a withdrawn delivery promise, where the explanation itself is work that no decision recorded. Where Lipsky does not meet the records is in the direction of the cost. Lipsky's costs fall on the people served; the records' cost lenses place follow-on work on the working layer itself, and name no served party except, in the transfer scene, a payee who waits for money. The records also show no discretion being exercised — they show roles and conditions the lenses suppose — so applying Lipsky here would be an interpretation, not a report.
Albert O. Hirschman's Exit, Voice, and Loyalty offers three responses to an organisation's decline: leaving, speaking up, and the loyalty that holds leaving back and makes speaking up more likely. This bears on when a restraint should be taken out of the way more than on cost. A worker who can leave can impose a cost that no cost lens records; a worker who speaks up can surface the waiting and rework identified as the only evidence about a choice that cannot be reconstructed afterwards. The records contain neither exit nor voice. The employee who hesitates to leave after call forwarding is set is, if anything, the picture of loyalty holding exit back, but the record does not say so. What the records do show is a structure in which voice has nowhere to go: the finder who cannot stop without another's approval, in five scenes. Hirschman's terms name what that structure forecloses, without establishing that anyone in these scenes wanted to use it.
Amy C. Edmondson's account of psychological safety — that team members who believe it safe to take interpersonal risks speak up about errors and learn more — is the nearest work to the records' own repeated pattern of a warning that must travel from finder to decider. It meets the records at the point where a cost becomes visible: a cost that is never spoken becomes a cost that is absorbed. Where it does not meet them is in level and in evidence. Edmondson describes a property of teams; the records describe roles in hypothetical lenses, and no team, no speaking up, and no learning are reported anywhere in them. To say that these scenes show a lack of psychological safety would be to report a measurement the records do not contain.
James Reason's account of organisational accidents meets the records' condition most exactly and departs from them furthest. Reason holds that accidents occur when the holes in successive layers of defence line up, and that many of those holes are latent conditions created far from the front line. The records' condition — that the person who decides and the person who cleans up are different — is structurally similar: a weakness created at one layer and paid for at another. But the similarity is formal. Reason writes about accidents; the ten records report no accident, no outcome and no aligned holes. The counter-finding that circumstances arising after a decision are not the decider's cost shift is Reason's point seen from the side of attribution: if a hole was latent, the front line inherits it without having made it. Reason thus strengthens the chapter's refusal to assign blame and does not strengthen any of its observations about what happened.
Set together, the four works show where the chapter stands in the debate the study of organisations conducts. The chapter sides with the accounts that treat cost as structural and directionally patterned, against any account that reads cost as individual diligence — which is why both the penalty table and the silence on follow-on cost were refused. It remains undecided between the view that rules can be redesigned to share cost, the question Lipsky and Edmondson make live, and the view that the pattern persists because it is latent, the question Reason makes live, because the records cannot distinguish the two. For the study's four questions this adds little and settles nothing: the records still do not say what each restraint protects, and the cost question still returns only direction and bearer, never amount. What the four works add is a set of named hypotheses — discretion, exit and voice, safety to speak, latent conditions — that a reader can carry to the people in such scenes, and that this study leaves there.