Chapter 11. A Typology of Restraint

11.1Why a Typology and What It Cannot Do

Chapters 6 to 10 read ten working records one at a time: each scene on its own terms, then through eight questions, then against a tradition that has had two thousand years to disagree with itself about what holding back means. This chapter sets the ten scenes beside one another, to make visible what the individual readings could not show separately: that restraints which look alike from the outside are made of quite different materials.

A typology sorts things by the features they share. It does not rank them. No scale runs from the least to the most legitimate restraint in these ten scenes, because no such scale is available from the records: each restraint protects something, costs somebody, and carries a condition, in the studio's reading of it, under which that reading would be revised. A typology describes forms and not virtues. A cell in the table at the end of this chapter says what a restraint is made of, where it comes from, how far it reaches and how hard it is to undo; it does not say that the restraint should exist, that its holder acted well, or that the person who waited for it was wrong to wait.

The material of the sort is ARCHON's own material. Each scene carries eight lenses; the first is record-inspired, and each of the ten first lenses ends with a single-line reading of the scene, already quoted in the chapters that read the scenes one at a time. They are returned here as a set, and the returns are named as such. Each is a dated interpretation made by the studio, marked there as a reading that can be corrected — "Interpretation That Can Be Corrected" on the lock's cards, and elsewhere "Revisable interpretation", "Revisable Interpretation", "Correctable Interpretation", "Correctable interpretation" and "An interpretation open to revision". They are hypotheses about the shape of a scene, open to the evidence each scene's counterexample lens asks for. They are quoted here as a set, without endorsement.

Three observations can be made without ranking anything. First, the readings do not describe the same kind of object: a program acting without a remembered author, a promise, a calendar constraint, a state of a person. A single word such as "control" would flatten them. Second, several name a mismatch between the size of a thing and the size of its authority: a button that stops more than its appearance suggests, a name that hides more than it states, an absence that holds time without an order. Third, the readings are silent about outcome: none says the restraint worked, and none says it failed. That silence is what makes them usable as sorting material rather than as verdicts.

There is a further reason for the typology, which concerns method rather than scenes. A typology is itself a stored decision: a choice about how to divide a field, taken once and then left to govern what a later reader sees. Simon's decisions are drawn from premises of value and of fact (Administrative Behavior, 1947); the table below is such a premise. The term "programs" in the sense of stored routines belongs to March and Simon (1958); this study's own extension is that a program is a premise that keeps running after its authors are gone. The table should therefore be read as a device that can be wrong, not as a map of how restraint actually is. A reader who finds a scene fitting badly in a cell is being shown the limit of this chapter, not a fault in the scene.

11.2First Axis: Where the Restraint Comes From

The first axis asks whose decision is still running, and whether anyone can still name it. The ten scenes are sorted into nine source types named from the record's own narrative. The element the studio says is missing is set beside each type as further description, not confirmation, since what the studio cannot establish cannot confirm a source. A restraint's origin is normally recorded in its reason, and that reason is the first thing lost when a person changes roles, when a rule is copied into a new system, or when a decision is remembered only by its result. What remains is the control and, beside it, the thing nobody can say about it.

Type one: code with no living author. The journal-entry lock is the scene in which the restraint is a program behavior. The original reason could not be remembered accurately, so the behavior was checked again. What is missing is the boundary the lock was meant to protect.

Simon's account of a stored decision applies almost without qualification: the lock is a premise a later action follows, with no one present to re-decide it. Chesterton's parable of the fence in The Thing (1929) has already been set against this lock in Chapter 6, and the two readings were left unsettled there. The typology records only the form: here the restraint comes from stored code.

Type two: a standing rule or principle. Two scenes fall here. The purchase-page image prohibition is an instruction standing in the way of a request; the unscheduled-goods return is a principle presented as a boundary at the lowest execution layer. In the first the prohibition is presented as an explanation that can be questioned; in the second the principle is stated as settled, and the record shows a small quantity of goods received physically without scheduled information.

The two share a source form and split where the typology should name it: the image prohibition's origin can be interrogated because it is presented as an explanation; the goods principle's origin is presented as already settled, and what is missing is the reason for a divergence elsewhere in the chain. Chesterton's fence fits the first better than the second: asking who posted the notice differs from tracing a divergence between a delivery and a schedule.

Type three: a decision to stop at a threshold. The label-issuance stop is neither inherited nor automatic. Because the origin can change depending on the production manufacturer, it was decided not to confirm it and not to issue labels before verification. The restraint is a decision taken at a point in a process, and what is missing is the fact the decision waits on.

This type differs from authorless code because the decision has a date and a decider, and from the standing rule because it is tied to one item rather than stated as policy. Chapter 7 already drew the distinction between a gap in a document and wrong information attached to a physical item; here the source of this restraint is the point at which information becomes physical.

Type four: a control on screen. The option sold-out is a small state change with an unstated reach. A question was asked about whether a sold-out product is under stoppage, and guidance was shared that the product is not sold when an option sold-out is processed. The studio's reading is that the button may exercise authority wider than expected. What is missing is how wide.

The source is neither memory nor policy but configuration: a state change whose wider reach the record does not state, and whose setting the scene's lenses supply as assumptions. Simon's stored premise again, in a different medium.

Type five: a gap in a flow. The unsplit order form is the only scene in which the restraint is not something anybody set. Automatic intake status was created, but with no split order form, exchange dispatch came to a stop. The stoppage is a consequence of an output that did not appear, and what is missing is the output the process was supposed to produce.

Perrow's Normal Accidents (1984) is the nearest tradition here: the book argues that accidents in tightly coupled, complex systems are normal consequences of the design rather than failures within it. The source question differs, though. In this scene there is no restrainer to name from the record, only a flow that halted between intake and dispatch.

Type six: a withdrawn promise. The same-day delivery copy is a restraint made of words that have been removed. A request came to pause brand search ad asset registration and to remove same-day delivery copy from search ads. The restraint is the absence of the promise in future exposure, and what is missing is the state of affairs that would tell anyone whether the promise could still have been kept.

Type seven: a calendar. The holiday transfer is a restraint produced by a date and by the absence of an approver on that date. The planned end-of-month transfer date fell on a holiday, so whether the transfer could be made the day before was asked. What is missing is the pair of facts the question turns on.

Type eight: a person's absence. The absent delegate is a restraint held by an empty chair. With a person in charge absent, the argument to defer met the argument to at least set an outline. What is missing is the thing the absence makes unknown.

Type nine: a person's hesitation. The hesitating-to-leave scene is the only one in which the restraint has no external author, no configuration and no date; it exists in the interval between an order that was not given and an action not taken. Extension call forwarding was requested because of a delay by the manager, and the remaining employee was not sure whether leaving was allowed. What is missing is the reason for staying, if there was one.

Taken as a set, the nine types run from the most external to the most internal source. Code, configuration and calendar do not consult anyone before they act. A standing rule consults its author if the author can be found. A decision to stop at a threshold consults a fact that may not yet be available. A gap consults nobody. A withdrawn promise consults customers who have already read it. A person's absence consults a schedule. A person's hesitation consults a relationship. Nothing in this ordering is a judgement about seriousness: a hesitation may cost more than a lock, and Chapter 12 takes up the question of cost that this axis only names.

11.3Second Axis: Who Can See the Boundary

The second axis turns from the source of the restraint to the visibility of its boundary. The boundary is the scope a restraint protects: the transaction the lock guards, the page the prohibition governs, the schedule the gate enforces, the site the label attests to, the delivery the copy promises, the option the state touches, the output the flow completes, the delegation the meeting needs, the work that would justify waiting, the deadline the payment must meet. In each of the ten scenes the studio names a single missing element, quoted in the previous section; in several of the ten that element is the boundary a restraint protects, and in others it is a cause, a constraint or an output instead. The chapter therefore treats the correspondence as a reading to be checked rather than as a property of how the scenes were recorded.

The visible boundary and the invisible boundary produce two different kinds of worker. Chapter 4 already described Simon's zone of acceptance, the range within which a person follows an instruction without re-deciding it; this chapter uses the term without restating the argument. In every scene the information lens's conditions assume a division of the same kind, in which one role holds the basis of a judgement and the other receives only its conclusion; the library repeats this division under different names in each scene: an approval-feature changer who knows and a closing-materials user who does not; an image change approver and a purchase-page production owner; an inbound exception approver and a receiving logistics centre; a production specification confirmer and a label issuance handler; an ad copy approver and an ad asset registrar; a sales-state setter and a sales-availability notifier; an automated process designer and an agent explaining customer delay; a meeting convener and a person in charge executing the outcome; a manager who sets end of work and an employee who waits; a fund execution approver and an approval-request preparer. The roles change names from scene to scene. The assumed structure does not change.

What a receiver can use is named in the judgments of the information lenses, and two of them, from a software scene and a payment scene, are set side by side here. The first is from the lock's threshold of information:

The second is from the transfer's information threshold:

The two sentences are the studio's template for the information lens, with the roles changed; what they show together is that the studio's answer to the visibility question is the same in a software scene and in a payment scene: a conclusion travelling without its scope is not information, and a receiver who does not know the boundary may not even know what to ask. Chapter 10 gave the sharpest case of this: the record states only that the employee who hesitates was not sure whether leaving was allowed, and the risk named in that scene's information lens is that a person who does not know the work that requires staying may not know what to ask.

The gain of a visible boundary is that a person acting inside a restraint can act on its purpose rather than on its appearance. If the label issuance handler could see whether the work order information matched the actual production site, the difference between a necessary hold and a blanket hold would be available at the point where the hold is applied rather than after it. Chapter 7 quoted the studio's own caution against a blanket hold on items already confirmed; that caution is a statement about visibility. The risk of the gain is that visibility is not free. To show the boundary to someone else is to spend the time of the person who holds it, to create an obligation to update it, and to expose the part of the basis that is still unverified. ARCHON's judgments say this in almost identical terms across all ten scenes: the risk of disclosing is that the response stops at writing down unverified values without anyone taking on the responsibility of refreshing them, and the risk of not disclosing is that the instruction the receiver is given takes on a different meaning in their hands.

Weick's essay on the collapse of sensemaking in organizations, "The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster" (1993), is the tradition this axis leans on, already introduced in Chapter 4 with the caution that it describes disasters and that these scenes are not. The scenes in this study are not disasters, and none of them records a collapse. What the essay describes is a crew's structure of roles and its sense of what was happening collapsing together as the fire turned, a collapse Weick calls a cosmology episode. This study's own reading, marked as such, is that the scenes here are a mild form of that condition: people working inside a structure they cannot fully see, holding a situation together with a conclusion whose basis is elsewhere. The visibility question is therefore not a question about transparency as a virtue. It is a question about whether the person nearest the stoppage has enough of the situation to act inside it.

11.4Third Axis: The Scope of the Stop

The third axis asks what each restraint stops, and how far. Scope is not the same as firmness, and it is not the same as visibility. A restraint can be highly visible and still reach further than anyone intends; a restraint can be narrow and almost invisible. The scenes make both combinations, and the typology records them separately.

Scope has two components in these records. The first is reach: whether the restraint touches one item or a linked set. The second is the condition under which it can be lifted, which in these scenes comes down to a question of who is allowed to raise a hold. ARCHON asks this question in every scene under an authority-to-stop lens, and in every scene the same three positions are set out, with what each gains and what each risks. The three positions are: a temporary hold issued by the person who discovers a problem, released by the decider with a basis recorded; the concentration of both stopping and resuming in one named authority; and each site stopping and resuming on its own judgement. All three positions are quoted here once, from the lock's authority-to-stop lens, which sets them out in the studio's clearest form:

The other scenes repeat these three positions, with a gain and a risk in each case, in near-identical words.

The repetition belongs to the template the studio uses for every scene; it is not a finding about the scenes, and a caution is owed the reader: the authority-to-stop lens is marked hypothetical in every scene, and its conditions assume a finder who cannot stop without another's approval. What the repetition shows is that the studio asks the same question everywhere and supplies the same three positions to answer it; that the scenes themselves contain two roles of which one can see a problem before the other can act on it is an assumption of the lens, not a fact reported by any record. The gain in every case is time between discovery and spread. The risk in every case is the same shape: an undefined hold becomes an indefinite veto, which is to say that the person who holds acquires a confirmation power nobody can name or time. The studio's judgment in each scene adds the caution that if what and how much to stop is not defined, the hold right becomes an opaque authority that holds up other teams.

Hobbes's Leviathan (1651) is the tradition this axis is read against, in the form the study has already used in Chapter 2: men without a common power to keep them all in awe are in a war of every man against every man, and the end of obedience is protection. The formulation of a restraint whose holder keeps something back and in doing so keeps everything else moving is this study's own reading of Hobbes through the katechon; the cited chapters of Leviathan do not state it. The question Hobbes forces is who holds the power once it has been transferred by covenant; what happens when that power is lent to another is Han Feizi's question, which the next sentence answers. Han Feizi's chapter "The Two Handles" in the Han Feizi answers that question with a flatness the study noted in Chapter 5: reward and punishment are the two handles, and they are dangerous when handed to anyone else. Applied to these scenes, the second and third positions are precisely the lending and the withholding of the handle. Concentrating both stop and resume in one authority keeps the handle with one person, which gains traceability and risks a delay whenever that person is unavailable — a risk the holiday transfer scene shows in concrete form, where the approver's absence is the fact the whole question turns on. Letting each site stop and resume on its own judgement gives the handle to everyone, which gains speed and risks linked work colliding where two sites hold opposite views of the same boundary. The first position is the attempt to keep the handle with the decider while putting the trigger in the discoverer's hands, and its risk is the one repeated in the three quotes above.

Scope, in this sense, is where a restraint meets its people. It is not the only place they meet; Chapter 12 takes the meeting that concerns cost, and this chapter will not anticipate it beyond noting that a wider reach usually means more people waiting on a single authority, which is a cost, and not only a benefit.

11.5Fourth Axis: How Firm the Restraint Is

The fourth axis asks how hard a restraint is to undo. Firmness is a property of the material a restraint is made of, and the ten scenes fall along a range rather than into two clean groups.

At the hard end are restraints whose objects resist reversal. A printed label is the clearest case: the studio's reading of the label scene states the reason directly, that once provisional information is printed, disposal and reissuance and mix-up management follow. A payment that has already been made is at the hard end for a different reason: the money has left, and a transfer cannot be unsent. A commitment made outside an organisation — the personnel, budget or delivery dates the meeting scene's judgment warns that an outline may settle — is hard because it binds parties who are not in the room. A returned delivery is hard because the goods have physically moved and the work of identifying what does not match the schedule falls to the receiving site. A stopped dispatch is hard because exchange dispatch came to a stop after automatic intake status was created; the record does not state that a customer is waiting, and the role that explains the delay is supplied by the scene's hypothetical lens. A stopped sale is hard in the direction of loss, because an option that cannot be bought cannot be unbought by the person who wanted it.

At the soft end are restraints made of words that have not yet hardened into anything. A request to replace a purchase-page main image is soft: it is a request, its approval status is unstated, and nothing has been committed to anyone outside the conversation. A copy edit is soft in the same way: an asset can be replaced, wording can be changed, and the material can be re-registered. The same-day delivery scene is instructive because it sits at the seam. The removal of the copy is a soft act — an edit to an asset — and its object is a promise that has already hardened in the minds of customers who have read it. The scene's own judgment makes exactly this point: even if the copy has been deleted from one asset, if it remains elsewhere, customer expectations do not change. Softness of mechanism and firmness of effect are not the same property, and a typology that treats them as one will mis-sort this scene.

The relation between firmness and the cost of reversal is named here and not settled. Chapter 12 takes the question of cost and its carrier, and this chapter deliberately leaves it there: what this axis establishes is only that firmness is a property a restraint has before anyone decides whether to reverse it. A printed label and a copy edit are both restraints; the difference in what they cost to undo is a fact about their material, not a fact about their rightness.

Cacciari's The Withholding Power: An Essay on Political Theology (2013), already introduced in Chapter 3, supplies the metaphor this chapter uses. Chapter 3 already gave Cacciari's withholding its full statement; here the metaphor is read as a spectrum rather than as a single power, and that is where it becomes useful. It lets the study place a program lock, a calendar, a withdrawn promise and a superior's absence on one axis without pretending they are the same kind of thing: they differ in material, in source and in reach, and they are alike only in that each of them delays a coming, and each of them delays some remedy along with the harm it holds back — this study's extension of Cacciari's reading, which the reading holds as an ambiguity of the withholding power. The gain of the metaphor is that it unifies stops that would otherwise have to be compared by analogy one at a time.

The risk is equally clear, and Chapter 3 stated it: the tradition makes a mundane control look world-holding. A small configuration error, a missing intermediate output, a date on a calendar — none of these is the power that withholds until taken out of the way, and describing them in that vocabulary flatters them. It also flatters whoever holds them: Chapter 2.5 already argued that a theory in which the restraining power is justified by what it postpones will justify any postponement whatever it costs. The warning the study carries about that move rests on Schmitt's account of sovereignty as the decision on the exception in Political Theology (1922) and his reading of the katechon in The Nomos of the Earth (1950) and later writings; the plain statement owed to the reader here is that Schmitt served the Nazi regime from 1933, and that critics read his theory of the exception in that light. Perrow's Normal Accidents (1984) supplies the counterweight to the metaphor's grandeur: some stops are normal to the design of a tightly coupled system — this study's reading, not a claim in Perrow's book, is that they cost the people who work inside them — and they are not thereby world-holding or thereby justified. The metaphor is used in this chapter as a spectrum of delay, from a lock that holds a period boundary to a hesitation that holds an evening, and it is used nowhere as a description of what any restraint is worth.

Firmness also raises a question the typology cannot answer from this sample: nothing in the ten records says what reaches a question library or what is resolved inside a conversation, and ten anonymised scenes cannot support a claim about the relative frequency of hard and soft restraint in working life. That three of the ten scenes are soft in mechanism is a property of these ten scenes and nothing more.

11.6The Typology Set Out

The table below sets the ten scenes against the four axes. The source column gives the type named in section 11.2. The boundary column gives the element the studio names as missing in each scene. The scope column gives what the restraint stops and how far, as far as the records state it. The firmness column gives the position on the range described in section 11.5. No column makes a judgement, and no row ranks against another.

The table is a working instrument. It is meant to be used in the chapters that follow as a way of asking the right question of a given scene quickly: where does this restraint come from, what is its boundary, how far does its stop reach, and what is it made of. Chapter 12 will take the question this table names in the scope column and nowhere answers — who waits, and who works again afterwards — and Chapter 13 will take the question the table names in the source column and nowhere answers — what would make this restraint be taken out of the way. Those chapters will not be able to use the table as evidence, because the table contains no evidence about cost and none about removal. It contains only forms.

The closing statement of this chapter is therefore a negative one, and it is the one the typology is built to support. No cell of this table makes a restraint right or wrong. A restraint that comes from code with no living author is not thereby illegitimate, and a restraint that comes from a person's hesitation is not thereby unjust. A hard restraint is not thereby serious, and a soft one is not thereby negligible. A wide stop is not thereby dangerous, and a narrow one is not thereby safe. Each cell names a form; each form protects a boundary, costs somebody, and carries a condition, in the studio's reading of it, under which that reading would be revised. The ten scenes were chosen as evidence about how organisations decide and act, and what they show about restraint is that restraint takes many forms, that the forms differ in ways that matter to the people inside them, and that the question of whether any of them should stand is not a question a typology can answer. It is the question the rest of this study asks with the records open.

11.7The Typology beside Earlier Ones

Six earlier sortings are available, each by a property the table in section 11.6 does not carry. They were built for different jobs.

Merton and the rule that outlives its purpose. Merton's "Bureaucratic Structure and Personality" (1940) argues that officials come to treat rules as ends in themselves and to keep to them after they have stopped serving their purpose — the displacement of goals. The typology stands nearest to this in the type called code with no living author. The two disagree not about facts but about what may be asserted: the typology says a control's reason is out of memory and stops there, because that is what the record states, while Merton supplies a mechanism by which such a control survives. The ten scenes cannot decide between them; the lock's record reports no official treating anything as an end. What Merton shows that this typology does not is a reason for persistence. If displacement of goals is at work, the restraint may protect nothing and the case for taking it out of the way would be strong, and the records contain nothing that would let a reader establish either.

Gouldner and the three patterns. Gouldner's Patterns of Industrial Bureaucracy (1954), from a study of a gypsum plant, distinguishes three patterns of rules: mock rules imposed from outside that no one keeps, representative rules that both sides support, and punishment-centred rules that one side imposes on the other. The classification turns on reception rather than origin. The source column can say that a prohibition is a standing rule; it cannot say whether that rule is representative or punishment-centred, because the record shows an explanation posted and a request to change an image, with no evidence about who supports the rule or who resists it. Gouldner answers a question the first axis does not ask — not where the restraint came from but who it is between — which bears on what it protects: a mock rule protects nothing and a punishment-centred rule protects one side against another.

Adler and Borys and the two kinds of rule. Adler and Borys's "Two Types of Bureaucracy: Enabling and Coercive" (1996) argues that formal rules can enable or coerce: enabling rules help users master the work, show their reasons and leave room to repair, while coercive rules force compliance and hide their reasons, and the difference lies in how the rules are designed. This is the closest of the six to the second axis. The difference is one of grain: enabling and coercive are properties of a rule's design, assigned from outside by someone who can see how it was built, whereas the visibility of the boundary is a property of a scene, assigned from the element the studio names as missing; the records do not report how any rule was designed. What their sorting shows that the table does not is that concealment can be a design choice rather than an accident of memory, and that a reason hidden by design and a reason lost to forgetting look identical from the position of the person who cannot see the boundary.

Kaufman and the residue of protection. Kaufman's Red Tape: Its Origins, Uses, and Abuses (1977) argues that red tape is mostly the residue of protections that people demanded, that one person's red tape is another's safeguard, and that cutting it means deciding whose protection to drop. It bears on the fourth question, when a restraint should be taken out of the way, more directly than any of the six. The firmness axis names the cost of reversing a restraint and stops there; Kaufman says the cost is not only work but a loss of protection, landing on someone who did not ask for the cut. Chapter 12 takes that question up with the records open, and section 11.6 already said the table contains no evidence about cost.

Crozier and the zone of uncertainty. Crozier's The Bureaucratic Phenomenon (1964) argues that in the organisations he studied, those who control a zone of uncertainty hold power, and that impersonal rules protect people from one another's arbitrary power and so make the organisation rigid, in vicious circles that are hard to break. This overlaps the third axis, who may say wait, and the fourth, how hard a restraint is to undo; section 11.4 asked the first through Hobbes and Han Feizi, without the vocabulary of uncertainty. What Crozier's sorting shows that the typology does not is that rigidity has a beneficiary: the same property Chapter 12 will cost is one someone may be relying on, and the table records firmness as a fact about material and cannot weigh that.

Lipsky and the front line. Lipsky's Street-Level Bureaucracy: Dilemmas of the Individual in Public Services (1980) argues that front-line workers make policy through the discretion they use under rules and scarce resources, and that the people they serve bear its costs. It answers the third question, what it costs and who pays, which section 11.6 left to Chapter 12, and it bears on the third axis, since discretion widens or narrows a stop from inside it: the authority-to-stop positions quoted in section 11.4 name who may raise a hold, and Lipsky names what happens when someone with no formal authority raises one or lowers one.

The two differences that remain. Two features of the typology have no counterpart in the six sortings. The first is the visibility of the boundary as a column of its own; of the six, only Adler and Borys come close, through reasons that are shown or hidden, and none records scene by scene what the person nearest the restraint cannot see. The second is the revision condition: each of ARCHON's judgments carries a dated condition stating what would make the reading change, and a status marking it as an interpretation that can be corrected, so that every cell of the table inherits a statement of its own withdrawal. The six sortings do not arrive with such a field, and this study does not treat that absence as a fault in them.

Against the four questions the first chapter left standing, the division of labour is plain. The typology answers the first, what restrains here, and partly the second, what it protects, by naming a boundary without naming its beneficiary; it names cost of reversal and the authority to lift a hold, and defers both. Gouldner and Adler and Borys answer parts of the second, Kaufman and Lipsky the third, and Merton offers a test for the fourth that the records cannot apply. The earlier sortings do not compete with the table so much as occupy the cells it leaves empty: a reader who wants to know who pays for the lock must leave section 11.6.

11.8The Ten in the Library

The ten scenes read closely are ten of 186 scenes in ARCHON's public question library. The question is where they stand on the cells this study cares about: what holds work back, why its reason exists, who may lift it, and whether those who wait can see what it protects.

How the ledger was made. Each scene's public record was taken as ARCHON serves it. Six fields, and four hypotheses about them (K1 to K4 below), were registered on 2026-10-05 before any coding began, under a rule set in advance: 60% or more of the cases a hypothesis applies to counts as support, 40% or less as against, anything between as unclear; unclear also when unclear cases outnumber decided ones. Each scene was read twice, by independent codings of one model, and a cell counts only where both codings agree; where they differ the cell is unsettled. Whether there is a restraint, where it comes from and whether the scene says why were read from the record summary and the scene's title alone; whether a condition for lifting the restraint is named, and with whom applying it rests, also from the first lens's judgment and the condition it carries for revisiting; whether those who wait can see the boundary, as far as the record summary bears it out. The other seven lenses, and the studio's list of what it says it cannot establish, were left out as the studio's frame. Strict counts keep only the values both codings found stated in the record's own words; lenient counts also keep the values both gave where at least one inferred it from the setting.

How the coding was checked, and why it was done twice. The first coding gave the coders the studio's list of what it cannot establish as part of the record. Two further readings tested it. A larger model of the same family read all 186 scenes a third time and agreed with each coding about as well as the two agreed with each other: Cohen's kappa, which discounts the agreement chance alone would give, was 0.48 against 0.52 on source, 0.25 and 0.41 against 0.36 on reason, 0.24 and 0.18 against 0.19 on whose hands, and 0.25 and 0.36 against 0.32 on boundary. A stronger reader of the same kind settled nothing the two codings had left unsettled. A model of another family then read 40 scenes drawn at random, and found in 18 of them no restraint in the record itself. In 17 of those 18 both codings had found one, and in 11 they had agreed it was a gap in a flow. In nearly every one the codings had taken the studio's list of what it cannot establish for something held back: the frame that section 0.6 says this study does not count as a finding. The fields were therefore coded again with that list out of the coders' material, the fields and the hypotheses unchanged. On the same 40 scenes the second coding agrees with the other reader far better than the first did: on whether there is a restraint, kappa rose from 0.00 and -0.04 to 0.42 and 0.35, and on whether the scene says why, from 0.03 and 0.13 to 0.44 and 0.39. On whose hands and on boundary it rose less, to between 0.10 and 0.17. A model of a third family then read the same 40 scenes on the second coding's terms. The two readers of other families agree with each other better than either agrees with the codings: kappa 0.65 on whether there is a restraint, 0.58 on source, 0.51 on reason, 0.63 on whether a condition for removal is named and 0.47 on whose hands it rests in; on whether those who wait can see the boundary they agree least, at 0.30. The counts below are the second coding's. The first coding's counts are kept with the ledger, and where they differ the difference is named.

Restraint. In 152 of 186 scenes both codings found something that holds work back or keeps something from happening: in 106 stated in the record's words, in 46 with at least one coding inferring it. In 14 both found none; 20 are unsettled. The first coding, reading the studio's list as part of the record, had found a restraint in 176. Even without the frame the codings find a restraint more readily than the readers of two other families, who found none in 18 and 21 of the 40 scenes they read. Of the 28 of those scenes in which both codings found a restraint, one reader found none in 8, the other in 10, and both in 6; in the 5 where both codings found none, all four readings agree. The count of 152 is therefore best read as an upper bound: on the two other readers' reading of their sample, something like half of the scenes show a restraint in the record itself.

Where restraints come from. The codings agreed on source in 120 of 186 scenes (65%). The largest agreed value is a gap in a flow, work stopped because a piece of information or an output is missing: 38 scenes, 6 of them stated. Then a decision to stop: 22, 14 stated. A system setting or code: 14, 5 stated. Other: 12. A standing rule or principle: 11, 4 stated. A calendar or deadline: 6, 4 stated. A person absent or hesitating: 3, all stated. 14 are not applicable and 66 unsettled. In the first coding a gap in a flow had been agreed in 80 scenes.

Why the restraint exists. The codings agreed in 132 of 186 scenes (71%). In 89 the record does not say why; in 27 it says why in so many words, and in 1 more both codings gave a reason with at least one inferring it; in 1 the reason is said to be unknown. 14 are not applicable and 54 unsettled.

A condition for lifting or narrowing. The codings agreed in 157 of 186 scenes (84%). In 139 both found such a condition named in the record's words, and in 2 more with at least one inferring it; in 3 none is named; 13 are not applicable and 29 unsettled. The condition may come from the record itself or from the judgment's condition for revisiting.

Whose hands it rests in. The codings agreed in 115 of 186 scenes (62%). In 81, applying the condition rests with the holder of the restraint or a third party, in all but one with at least one coding inferring it; in 14 the record does not say; in 3 those who wait could invoke it themselves; 17 are not applicable and 71 unsettled.

Whether those who wait can see the boundary. The codings agreed in 117 of 186 scenes (63%). In 73 the record does not say; in 26 those who wait cannot see what the restraint protects and how far it reaches, 2 of them stated; in 5 they can, 3 of them stated; 13 are not applicable and 69 unsettled. The first coding had found in 89 scenes that those who wait cannot see the boundary. Without the frame, the commonest reading is not that they cannot see it but that the record does not say.

The study's test. A restraint that passes all four at once — its reason stated in the scene, a condition for removal named, that condition in the hands of those who wait, and its boundary visible to them — passes in 0 of the 152 scenes that show one. Among the ten scenes read closely, 0. The first coding gave 0 of 176.

The counts bear unevenly on the study's questions. What holds an ordinary organisation back is answered as to form in most scenes. What it protects is answered only where the record says why, which is seldom. When a restraint should be removed is answered as to form in 139 scenes where the record states it and 2 more where it implies it, and as to hands, in favour of those who wait, in 3. What holding costs and who pays is carried by none of these cells.

The four hypotheses registered before the coding.

K1: in most scenes that show a restraint, the record does not give both the restraint's reason and a condition for its removal. Strict: unclear — 62 supports, 15 against, 97 unclear, share 81%. Lenient: supports — 91, 22 and 59, share 81%.

K2: where a condition for removal is named, applying it rests with the holder or a third party, or the record does not say with whom, rather than with those who wait. Strict: unclear — 9 supports, 0 against, 160 unclear, share 100%. Lenient: supports — 92, 3 and 74, share 97%.

K3: in most scenes that show a restraint, those who wait under it cannot see, on the record's terms, what it protects and how far it reaches. Strict: unclear — 47 supports, 3 against, 124 unclear, share 94%. Lenient: supports — 97, 5 and 70, share 95%.

K4: in most scenes that show a restraint, the scene's own record does not say why the restraint exists. Strict: unclear — 61 supports, 19 against, 94 unclear, share 76%. Lenient: supports — 90, 28 and 54, share 76%.

The share is the proportion of the cases the codings decided that support the hypothesis; it is not a proportion of all 186 scenes. On the strict count no hypothesis is decided, because the unclear outnumber the decided; on the lenient count all four are supported. The verdicts are the same as the first coding's.

Where the ten stand among the 186. On the same cells the ten scenes read here look like this: all 10 show a restraint; on source, 2 each in standing rule or principle, decision to stop, system setting or code, and a person absent or hesitating, 1 calendar or deadline and 1 gap in a flow; on whether the record says why, 5 not said, 4 unsettled, 1 stated (the Label Issuance Stop); on removal condition, 10 named; on whose hands the condition rests, 8 the holder or a third party, 1 not said, 1 unsettled; on whether those who wait can see the boundary, 4 no, 4 unsettled, 1 not said, 1 yes (again the Label Issuance Stop). None of the ten passes the study's test.

The selection of section 0.6 does and does not represent the library. It represents the library on the test that matters most to this study: the ten fall short of it as the 152 do, and 0 of 152 against 0 of 10 is not a selection effect. On reason and on hands the ten fall where the library falls. On the boundary they are clearer than the library, because the ten were chosen for restraints whose records say enough to be read closely: the codings agree that those who wait cannot see the boundary in 4 of the ten, while across the library the commonest agreed value is that the record does not say. And the selection does not represent the library as a sample: the library's largest agreed source, a gap in a flow, has one scene among the ten. George and Bennett's account of case-study kinds distinguishes heuristic case studies, which use cases to find new variables, paths and questions, from plausibility probes, preliminary studies that ask whether an untested idea is worth fuller testing. The ten scenes function here in the second capacity: they are not a basis for claims about the library's frequencies. Set against the 186, they ask whether the reading developed in Chapters 6 to 10 would survive a wider field; on removal and on the silence of records the lenient count says it survives and the strict count cannot say.

What the counts cannot show. Five limits bind every number above. First, the records are anonymised summaries; a cell counted as not said may record a gap in the summary rather than a silence in the workplace. Second, the two codings are one model run twice, not two readers: run separately, neither given the other's output, but sharing a way of reading. A larger model of the same family did not read differently, and readers of two other families, on a sample of 40, agreed with each other better than with the codings, and least of all, with each other as with the codings, on whether those who wait can see the boundary. The cells this study's thesis rests on are the cells the records least let any reader settle. The test, and the verdicts on the four hypotheses, are the same on every reading: no reader of any family found a scene that meets the test. Third, the fields were coded twice, and the second coding was made after the first coding's counts had been seen. The fields, the hypotheses and the rule for verdicts did not change; only the material the coders were given did, to keep out what this study's method had already set aside as frame. The verdicts are the same on both codings; the counts of restraints, of gaps in a flow and of boundaries unseen are not, and a reader should weigh them with that in mind. Fourth, a cell is a reading of a record, not an observation of a workplace: each number counts how a record was read on 2026-10-05. Fifth, the strict and lenient counts are two different questions posed to the same reading. On the strict count this library does not establish what holds ordinary organisations back or who may lift what is held. On the lenient count it suggests that reason and hands are what records most often leave out, and that whether those who wait can see the boundary is, more often than not, not recorded at all. The ten scenes are not an aberration in their library, and the library is not thereby a description of working life. Chapter 12 takes up what holding costs and who pays, which no cell here carries, and cannot use these counts as evidence.