Chapter 13. Conditions for Revisiting: What Could Be Observed

13.1Every record has a condition, by its form

Every one of the 53 records ends with a line telling the reader when the record's judgment should be reconsidered. The ledger counts 53 of 53 (100%) with a condition for revisiting the judgment, and 53 conditions in all. Read at a glance that looks like the uniformity of a careful practice: a record that always states what would make it change course. Chapter 2's rule removes the glare. The record form asks for such a line, and a writer completing the form must supply one. What a form guarantees is not evidence about how decisions are made, so the fact that all 53 conditions exist is not a finding.

The same rule governs the absences, and there are three of them. The form has no field for a deadline, so the ledger counts 0 cases with a deadline in the record. It has no field for a decider, so the ledger counts 0 cases with the decider in the record. And no record lists what its setting leaves out, so the row for missing items reads 0, with 0 items listed in all. The ledger reports those zeros, but the form does not guarantee them: a deadline, a decider or a list of missing information could appear in another field. A condition has to do its work without a decider named, without a date by which the matter must be settled, and without a statement of what the record does not know.

What is a finding lies inside the conditions themselves. Axis A5 of the ledger codes two things about every condition: that it exists, and whether a reader could observe it. Five conditions in the whole set name something a person could look for in a log, a schedule, a rule book or a list of posts. The remainder name states of the world that would have to be judged before anyone could say the condition had arrived.

The ledger also measures a third property: how much of a condition's wording is shared with WSTiger's studio frame. Two of the 53 conditions share 30% or more of their wording with that frame. Chapter 2 sets the rule for reading such wording: a field whose wording is at least 30% shared across a studio's records is not evidence across domains, because it shows what a studio's writers tend to write rather than what this setting demands. The rule keeps that wording from being evidence across domains; it does not say the wording carries no evidence about any one setting.

Within the record's fields, the condition occupies a peculiar place. The Scenario describes how the setting came about; the Question and the Options are the decision in front of someone now; the Example judgment is a reading of that decision, which is WSTiger's own and is dated and revisable. The Invalidation line looks forward in one particular way: it names the condition under which the reading would have to be made again. This chapter reads eight of those lines, quoting nothing else from the records they belong to.

Two rules hold throughout what follows. The sentence in each block quote is the record's own, copied exactly; what follows each one is this study's reading, not the studio's. And no line here is evidence about a real firm, trade or market event: every record is a synthetic setting published for study, and its condition describes a setting, not a history.

13.2Conditions a reader could check

A condition that a reader could check names something a person could go and look for: a number in a limit, a date on a calendar, an event in a sequence, or a named position whose action would settle the matter. The four records quoted in this section are the ones the plan groups as naming that kind of thing. Each is read here on its Invalidation line alone. Earlier chapters read these same records on other axes — where authority sits, what a shared measure leaves out, when underperformance becomes evidence, what can be undone — and this section adds only what the condition itself does and does not fix. The ledger counts five conditions of this kind across the 53.

Who holds the switch

What the sentence names is an event with a sequence in it. An envelope is a limit already set on an agent; a halt is an action taken in response; and the condition fires only when the first happens before the second. Both halves are checkable in principle: a desk that logs exposure against an envelope and timestamps its halts has produced exactly the two pieces of evidence the sentence asks for, and their order settles the condition. What a reader would watch for is a log line showing the agent against its envelope, and any halt line with its time; the condition is met when the breach line comes before any halt line, or when no halt line appears at all.

What the sentence does not fix is almost everything around those lines. It does not say whose envelope, in what sense a halt triggers, whether one instance or a run of instances would prompt reconsideration, or who would be reading the log. The ledger's rows explain why those gaps cannot be filled from the record: no decider is named in the record and nothing is listed as missing, so the sentence is read as it stands.

One asymmetry is worth naming, because it follows from the wording rather than from anything the record adds. The condition is met by an envelope breach that comes before any halt, including a breach after which no halt ever fires. A desk that halts early, or halts close to the edge, is never prompted to reconsider, however narrow the margin. This study reads that as a condition aimed at the failure of a defence rather than at the risk the defence is meant to hold back — a defensible shape for a condition, and one that keeps a timely intervention from prompting reconsideration. Chapter 8 read this record beside One anomaly three agents and Three venues halt apart, on where authority sits; the condition here adds a test that would live outside the record, in the desk's logs.

One measure all banks

The sentence names two things that could appear in a body of records: differences between measurement methods across banks, and the absence of market sales after simultaneous limit breaches. A reader could look for both in principle — methods can be compared line by line, and breach days can be set beside sale days in a sequence. What a reader would watch for is how far the methods differ, and whether breach days are followed by sales.

Three gaps keep the condition from being checkable as written. The word sufficiently carries no amount, so a reader cannot know when the accumulation has reached the point of the condition. The phrase did not lead to is a causal claim about a market, not a field in any one firm's records; deciding it requires an argument about what would have happened otherwise. And no body is named that would accumulate the records in the first place — the sentence assumes a repository that no single setting in the record provides.

The condition is also written at an altitude above any one bank. The record's Scenario names a bank supervisory authority recommending one measure to large banks, and its Question asks whether responses to limit breaches should be set differently for each bank; on this study's reading, the condition asks for evidence about how banks differ, not a decision any one bank has to make. Chapter 6 read this record as one firm's choice inside a shared measure, and the Invalidation turns the question round: what the individual bank could decide is not what the condition waits for.

Live underperformance criteria

Two triggers and one check are set out here. The triggers are a confirmed risk-limit breach or a confirmed structural divergence, and the procedures that follow them. The check is a classification: profit-and-loss alerts divided into four types — risk, data, execution, and hypothesis. Of the three, the check is the most checkable part of any condition in this chapter's case set. A four-way division is a rule a reader could apply to an alert stream and count. What a reader would watch for is whether alerts are assigned to the four types, who assigns them, and on what data.

The triggers still lean on judgment. The words confirmed and structural divergence are conclusions, and the record does not say who reaches them, on what evidence, or by when. The phrase recovery and re-validation names a state without any measure attached to it. And the prohibition in the middle of the sentence — not reverting to the original state before recovery and re-validation — is about operational restoration, and the record gives no way to observe it.

Chapter 7 read this record on the point at which underperformance in live trading becomes evidence, and the condition adds something the earlier reading did not need: an operational shape. An alert triage is a change to how the desk would work, not a statement about the market, and it is the kind of condition that would remain checkable whatever the market did. This study reads that as the strongest form the observability row can take — a condition that names the machinery that would make it testable.

Stop distance and size

This is the longest of the eight conditions quoted in this chapter, and the one that carries a named external reference. It sets out a trigger — a stop-loss criterion changing legitimately because market structure changed; a requirement — re-determining the position quantity; a prohibition — not raising the existing risk budget; a next check — separate checking of excess losses where the stop-loss fill is not guaranteed; and a source, named explicitly as being for conceptual confirmation rather than measurement.

What a reader could watch for: whether the risk budget moved, and by how much; whether stop-loss fills were in fact not guaranteed, and how many excess losses fell into that class; and whether a re-determination of quantity was derived from the budget or from the desire to take the trade. The excess-loss class is the kind of thing a desk logs — a fill that did not protect the position, and the loss beyond the planned stop. It is countable, and the record asks for it to be checked separately.

What the condition leaves to judgment: the words legitimately, due to a change in market structure, and arbitrarily are all conclusions, and no two of them are the same conclusion. The record itself marks the limit of its own named source, describing the methodology reference as being for conceptual confirmation, not measurement, so the one external name in the condition does not settle any quantity. The condition here is written as a rule about quantities, and the part of it a reader could count is exactly the part that touches the loss.

What the four conditions share

Across the four the arrangement differs: the switch condition is a single event-sequence test, the banks condition mixes judgment-dependent terms through both of its limbs, and in the other two the checkable content sits nearer the end. Triggers turn on what is confirmed, on what changes legitimately, and on what did not lead to; the parts that could be inspected are a division into four types, separate checking, and log lines recording a breach and any halt. One of the four names a mechanism — an alert classification — that would make a condition checkable by design, and that mechanism is a change to the desk rather than to the market. Two name external facts whose observability depends on records held somewhere other than the setting itself.

One caution applies to the whole section. A condition can earn an observable rating while most of its words still ask for judgment, as Stop distance and size does. The observable row is therefore a row about a first step, not a row about the sentence as a whole.

13.3Conditions no one could check

The four conditions quoted in this section name states of the world: whether war's possibility is resolved, whether a limit falls within a risk budget, whether a portfolio's rule-keeping trades hold up over a follow-up sample, whether a rolling limit stops operations excessively. None of these is a field in a record, and each would have to be interpreted before anyone could say the condition had arrived. Parts of the same conditions also name inspectable procedures — an independent review, an approval trail, separate log fields — and the readings below separate those from the states. The four quoted here are the clearest examples of what a condition leaves to judgment and of what it would take to become observable.

Calm route marine insurance

The setting, as the plan gives it, is a route where accidents have been rare for ten years, with stable annual premium arriving and rumours of war circulating between two countries. The condition waits for one of two things: war's possibility being resolved, or the contracts being confirmed as spread across different risks.

Both limbs leave everything to judgment. There is no named authority whose statement would settle whether war's possibility is resolved — not a government, not a court, not a market — so resolved is a description a reader would have to give. And whether contracts are spread is a question about correlation, about how many of them would be hit by one common event; the record does not write down the contracts, their routes, their buyers or their sums, so the confirmation cannot be carried out inside the setting. What a reader would watch for, if the condition were to become observable, is a pronouncement with a date attached, and a list of contracts scored by how much they share.

The vagueness sits precisely on the axis of time that this record was chosen for. The interval between rumour and resolution is the interval in which premium keeps arriving; a condition that cannot name when resolution arrives is a condition that cannot say how long that interval lasts. Chapter 10 read this record on who collects now and who pays later. The Invalidation adds one thing to that reading: the moment at which the question would be reopened is the same moment the record does not know how to identify.

Leverage limit raised repeatedly

Two limbs again, and they are not equally checkable. The first asks whether an independent review confirms that the new limit falls within the risk budget. That is the very question the review exists to answer, so the condition points to a conclusion rather than to a fact; it does not name the review body, what the review must examine, what the risk budget is in figures, or when the review must finish. The second limb is more procedural than the first: whether a limit was upheld without retroactive approval depends on how approvals are recorded, and that is a fact a log could hold.

What a reader would watch for is the approval trail — whether a limit change was recorded before or after the exposure it governed — and the figures behind any budget claim. What would make the first limb observable is not hard to describe: a named review body, a figure for the budget, a deadline for the review, and a record that separates prospective approval from retroactive. None of these appears in the record.

Chapter 5 read this record as the present form of a limit raised repeatedly until the raise becomes the practice. The condition responds to that drift with a review, and the record does not say what that review is meant to change. A review that arrives after the exposure has been taken cannot reopen the interval the drift ran through, and the condition does not say when the review would happen.

Good results good decisions

Here the second half is more checkable than the first, and it is worth separating the two. A trade log that stores a result assessment and a process assessment in independent fields is something a reader could open and inspect: two fields, holding different things. That is a design detail, and the record states it as a next verification, which is close to a test.

The first half carries four judgments in one sentence. The phrase a sufficient follow-up sample gives no size and no period. The word deteriorates gives no measure and no reference point — against a benchmark, against an earlier window, against an expectation. The phrase rule-compliant trades depends on the rule book, which the record does not reproduce. And the final clause, that past assessments are not revised to fit individual results, is a question about a process rather than a number: a reader would have to compare what was written at the time with what the review later concluded.

What a reader would watch for is the sample actually accumulated, the measure applied to it, and whether any assessment was rewritten. What the condition would need to become observable is a size, a measure, a direction, a threshold, and a rule fixing when the assessments are written. The condition here keeps the two assessments apart by design, which is a change to the desk's own machinery rather than to its circumstances.

Midnight loss limit refill

The setting places a volatile period across a midnight initialisation, with an agent reaching the limit minutes before and recovering the whole of it minutes after. The condition waits for records to show one of two things: that a rolling limit stops normal operations excessively, or that losses do not recur right after initialisation.

The second limb is nearly countable. Losses are logged with times, and right after initialisation could be given a window — minutes or hours — against which the losses in that window would be counted. The condition does not give the window, and it does not say how many losses in how many days would make the point. The first limb is not countable at all as written. The phrase normal operations is undefined in the record: proposals approved, orders placed, a rate of activity, none of these appears. The word excessively has no amount.

What a reader would watch for is the timing of losses around the initialisation hour, and the volume of the desk's activity on days when the rolling limit binds. What the condition would need is a window, a measure of operations, and a level above which the rolling limit counts as excessive. Chapter 12 read this record on the clock inside its own setting — the reset at midnight, the volatile period crossing it — and the condition asks whether the fix for that clock causes a problem of its own, in a way the record does not say could be seen.

What the unobservable conditions leave to a reader

Across the four, the unobservable element is concentrated in a single kind of word: resolved, within, sufficient, excessively. Each of these would become more observable by naming a threshold, though a threshold alone would not settle whether war's possibility is resolved, whether an independent review confirms a budget claim, or whether contracts share different risks; those need definitions, evidence and sometimes a named authority, and the records supply none of them. It would be wrong to read that as an argument that thresholds are what a decision needs — a threshold changes what a decision is about, and the distinctions the records' own Example judgments work with are not always the kind of thing a threshold carries. What the ledger does show is that the conditions inherit the record's silences: no record lists what is missing, and no record carries a deadline, so a condition that waits for records to accumulate waits for a period whose length no one has written down.

13.4Fantasy documents and thinking in time

Two works in the list stand beside these conditions. One concerns written artefacts produced before the outcome that cannot be executed as written; the other concerns the disciplined use of history in decisions. Neither can be settled on the records, and each illuminates a different part of what this chapter has found.

Clarke on fantasy documents

Lee Clarke's Mission Improbable: Using Fantasy Documents to Tame Disaster (1999) describes organisations writing plans for catastrophes they cannot control — Clarke's term is fantasy documents — plans that could not work as written but that serve to persuade others, and the organisations themselves, that the risk is under control. The plan's note gives three elements to hold onto: the artefact is a plan; it addresses a catastrophe beyond the organisation's control; and its work is persuasion rather than operation.

Set beside the four unobservable conditions, the fit is partial and the fit is real in one respect. A condition that cannot be acted on as written — the possibility of war is resolved, without a rule for resolution; a sufficient follow-up sample, without a size — has the shape of an artefact that cannot be executed. If a sentence cannot be checked, one thing it can still do is record that the risk was considered, and that is close to the persuasive work Clarke describes. This study reads that comparison as a question about function, not as a finding: the records do not say what the Invalidation line is for.

The comparison fails in three ways, and the failures are worth stating because they bound the lens. First, purpose: the plans Clarke describes are written to persuade that a catastrophe risk is controlled. These records are synthetic settings published for study, and no organisation in any of the settings is documented as reading them. Second, audience: in Clarke's account a fantasy document persuades others and can also persuade the organisation itself; this study reads the difference as one of the reader's task — working through the decision, with the condition as one field among several — rather than of audience. Third, scale: Clarke's claim is about plans for catastrophes an organisation cannot control, while several of these conditions concern local procedures — a fill that did not guarantee, an approval recorded as retroactive — and the records do not say which failure would matter more. Clarke's vocabulary belongs to catastrophe plans, and the limits on its fit here come from the differences between those plans and these records rather than from any rule about wording.

Neustadt and May on thinking in time

Richard E. Neustadt and Ernest R. May's Thinking in Time: The Uses of History for Decision-Makers (1986) addresses decision-makers who misuse history, above all by seizing on one analogy. Their recommendation is practical: separate what is known, unclear and presumed; ask how the present situation came about; and see the issue as part of a stream of time running from the past into the future. This study reads the two lenses as pointing in different directions: Clarke asks what a document does to its audience, while Neustadt and May ask how a reader can discipline their own use of history.

Applied to the conditions read above, the first recommendation produces a clean inventory. The known column is thin: the provisions the conditions state rather than evidence already gathered — separate checking of fills where the stop did not guarantee, two independent fields in a trade log, and the procedural fact of whether an approval was retroactive. The unclear column is larger: whether measurement methods differ sufficiently among banks, whether a new limit falls within a risk budget, whether a rolling limit stops normal operations excessively. The presumed column holds the events the conditions wait for: records accumulating, a review being held, a sample being sufficient, war's possibility being resolved. A condition can wait for such an event without asserting that the event will arrive. This study reads the inventory itself — what is stated, what is judged, what is waited for — as the most useful thing Neustadt and May's method yields for these eight records.

The second recommendation, asking how the present situation came about, works differently. The conditions rarely look backwards. Some of the conditions quoted in this section wait for evidence to accumulate — records accumulating in Midnight loss limit refill, a follow-up sample in Good results good decisions — which looks to a future supply of evidence rather than asking how the present state was produced. The record form does supply a Scenario, which describes how the setting came about, and the Invalidation does not refer back to it. The third recommendation, the stream of time, is the one this study's case design makes visible from outside: the same settings appear in past, present and future forms across the chapters, and no single record contains that stream. Chapter 12 read those three-time settings for what changes across the times; what this section adds is that no condition in any of them says how the setting it belongs to came to exist.

The comparison fails in two ways. Neustadt and May address decision-makers who misuse history by seizing on one analogy, within a method that also reaches beyond analogy, and this study reads them as assuming a decision-maker who can revise the analogy; the records name no one who will act on these conditions, and a condition a reader cannot observe can still be revised by clarifying its evidence, its authority or its threshold. And their method addresses the misuse of history by analogy, whereas the settings here are constructed by design, so what looks like an analogy problem between, say, a 1739 route and a 2026 desk is in fact the studio's choice to build the same setting at two ends of a period.

This study reads the two works as pointing in different directions at the same object. Clarke's answer is sceptical about what such documents are for; Neustadt and May's is to discipline the use of history in a decision. This chapter takes the second path: it reads each condition as a claim about evidence, states what a reader could not do with it, and records what it would take to change. What neither lens can settle is whether a condition that no one could check was written as a test or as a demonstration of care — the records make no claim about their own function.

13.5H2

The ledger's verdicts on H2 are identical under both readings. Strict — counting only what both coders of different families found stated — the counts are supports 48, against 5, unclear 0, not applicable 0, with a share of 91%, and the verdict is supports. Lenient, adding what the setting implies, the counts are again supports 48, against 5, unclear 0, not applicable 0, a share of 91%, and the verdict supports. The verdict rule and the pre-registration are set out in Chapter 2; Chapter 14 takes the verdicts on all eight hypotheses, and this chapter reports H2 alone.

What the verdict rests on deserves care, because H2 bundles two claims that the records cannot be made to test in equal measure. The first claim is that few decisions write down what would reverse them. On this study's 53 records that claim is foreclosed by the form, which asks for a condition in every record: 53 of 53 have one, 53 conditions in all. By Chapter 2's rule, that presence is form-guaranteed and cannot be reported as a finding in either direction. The second claim is that the conditions which do exist are written in words no one could observe. That claim is testable, and the ledger's observable row — 5 of 53 (9%) — is the evidence beneath it.

This study reads the verdict as resting entirely on the observability half of the hypothesis. The four conditions quoted in section 13.2 name something a reader could observe, and the four quoted in section 13.3 do not. A reader taking the verdict at face value would draw a stronger conclusion than the counts carry — that decisions rarely write down their reversal conditions — when what the counts support is narrower: that the conditions written in these records state no number, threshold or time, with five exceptions in fifty-three, even though parts of some remain inspectable.

The strict and lenient counts being identical follows from the way the observability row is scored rather than from any judgment the coders made.

Three limits on the verdict stand, and they are the limits of the study rather than of the coding. The 53 records are synthetic settings from one studio, chosen by topic rather than sampled, so the counts describe these records and nothing beyond them. And the verdict is a statement about how these sentences were built; it is not a claim about how decisions are made in real organisations.

The chapter ends where the study itself is to be judged. The Conclusion writes the study's own conditions for revisiting, in wording a reader could observe, so that the same test this chapter applied to the records can be applied to the study. The next chapter sets out the verdicts on all eight hypotheses.