Chapter 13. Conditions for Reopening: What Could Be Observed

13.1The lines every question carries, by the form

Meridian's question form asks what would reopen a decision. Each of the fifty defence questions this study reads carries lines answering that ask. The ledger records their presence as a property of the collection: on axis A5, cases with a condition for revisiting the judgment number 50 of 50, and the conditions in all number 798. Every setting in the collection is synthetic, in a fictional polity, and every judgment attached to it is AI-generated, dated and revisable. This chapter returns to eight questions for their Reopen when lines only, and quotes nothing else from them.

The presence of these lines in the questions is not a finding about the questions. The form asks for them, so their presence is guaranteed by the form. Chapter 2 sets out what Meridian's form guarantees, how the ledger's frame rule keeps the studio's sentences out of the records' own wording, and how two coders of different families recorded what they found. The counts in this chapter come from the ledger's own tables, computed by code, and each is given here with what it counts.

What is worth reading is not that the lines exist but what they hold. The ledger's numbers separate the conditions by field and by observability. In the field judgments[].claims.stopConditions — the stop conditions the thirteen judgments write in their own words — there are 650, of which 602 are observable. Across all 798 conditions, 629 (79%) are observable. The reopening lines are what a question shows of its own future, before any judgment has argued from the setting.

The frame row of the ledger's numbers answers a related question. Conditions whose wording is 30% or more the studio's frame: 0 of 798. On the ledger's measure, no condition is largely made up of the studio's sentences. The reopening lines are the records' own words. Where one is thinner than another, the difference sits in the wording the record carries; the frame measure does not show how that wording came to be there.

Observability is the axis worth reading here for a plain reason. A reopening condition is a promise about future observation: this decision may turn out to rest on a false assumption, and here is the thing that would show it. What the promise is worth to anyone reading the record afterwards depends on whether something observable sits underneath it — a dated check, a named role, a written measure. Where it does, the reader can ask whether the check was done and what it found. Where it does not, the line alone establishes less: it does not by itself rule out independent observation, a request for the supporting evidence, or a challenge to the institution's own interpretation. Chapter 2 sets out the pre-registered hypotheses and the verdict rule; this chapter takes one hypothesis in full, and takes the reopening lines apart first.

Two boundaries for what follows. This chapter does not weigh the options in any question, and it does not decide whether a reopening line is the right one to have written. Where a line is thin, this chapter says what would have to be added before a reader could check it. The eight cases return here only for these lines; their questions, options and judgments were read in earlier chapters.

The reading takes each line in three steps. First, what the line names: a role, a cycle, and a trigger. Every one of the eight names all three, and the naming is the part of a line that costs nothing to write. Second, what the line measures: whether the trigger is a comparison against a reference value written down somewhere, an event with a date or a description attached to it, or a judgment left to the person the line names. Third, what a reader could reach: whether the tally, the register, the reconciliation or the report the line depends on would exist somewhere outside the institution, and whether the value the trigger turns on would be visible in it. A line can be strong at one step and weak at another, and the eight lines read here are strong and weak in different places.

None of the three steps decides whether a reopening line is a good one to have written. A line that leaves a judgment to a named officer may be the only workable way to write the promise, because the judgment cannot be reduced to a number in advance. A line that carries a number may be easy to check and still check the wrong thing. What the steps give is a way of saying afterwards what a reader could and could not establish from the record alone, which is the only claim this chapter makes.

13.2Lines a reader could check

Four of the eight reopening lines name something a reader outside the room could watch for: a number, a date, an event or a named role, in each case doing a different piece of work.

Emergency budget ends (2026).

The named body is the Accounting Verification Office, the cycle is weekly, and two quantities are named as the things being compared: the omission rate of delivery confirmations, and actual cases of non-delivery. What a reader could watch is the weekly cross-check itself — whether it happened on the stated cycle, what it compared, and whether re-judgment was requested afterwards. The line leaves the sample unnamed and gives no values for either quantity, so a reader can see that a comparison was made and cannot, from this line alone, say how big a difference would count as a pattern. The condition is observable in its structure and unobservable in its threshold: what counts as different from the existing sample is a judgment the Office makes, and the line does not say what would make that judgment wrong. What the record would need is the sample described, or a rule tying the judgment to a stated difference.

Water parts contract timing (2026).

Here the number is fixed. Ten days after the contract, and again at final settlement, an independent cost auditor reconciles the cost evidence. Two moments are dated by interval rather than by calendar date, so a reader can watch for the first reconciliation on the tenth day after the contract was made and for the second once the contract is settled — provided a reader knows when each of those events occurred. The trigger is unexplained differences. The line does not say who decides that a difference is unexplained; it puts the finding in the auditor's hands and the reopening of price conditions behind it. The observable content is high by the ledger's standard — a named role, an interval in days, two events, a named quantity being reconciled. The judgment the line leaves open is what counts as explained. A reader would need the cost evidence, or the auditor's stated grounds, before saying whether the condition was met.

Arms verification data publication (2044).

The event here is a deadline rather than a date: at the application deadline for the next verification, and again at the receipt for the next round, a participation registration officer confirms cases of withdrawal. What a reader could watch is the register of withdrawals and the stated reasons — an artefact that could exist independently of the committee deciding the publication unit, with a reason column and a date. Two things sit inside the line that the line itself does not fix. The first is when the application deadline falls; the line refers to it as an event rather than stating a day. The second is what counts as formal withdrawal: a withdrawal that is not formal, or one that cites no reason, does not trigger the condition as written. The line names cases rather than a count, so a single qualifying withdrawal appears to satisfy it. What would make the condition checkable is the deadline written into the record, and a register whose reasons are themselves recorded.

Hospital or homes water (2026).

Two numbers sit in this line. The fairness inspector checks recovery progress and costs every two days; the trigger is a ten-day budget overrun that is expected. The reader can watch the two-dayly checks and the ten-day cost figures those checks produce. The condition turns on the word expected. An overrun that is expected is a forecast made by the inspector, so under this line an overrun that never occurs still triggers re-deliberation, because the condition is the expectation rather than the event. The line also leaves the baseline open: a ten-day budget overrun is written as an absolute rather than a percentage or a ceiling, so a reader would need to know what the ten-day plan was meant to cost before deciding that an expectation was reasonable. What would make the line fully checkable is the ten-day cost plan written down beside the inspector's reports.

Across the four lines the anatomy is the same: a named role, a cycle or an interval, and a trigger. Two of the four put numbers in the line; all four name a role; all four place the trigger's content in a judgment made inside the checking arrangement — a pattern of difference from the existing sample, differences left unexplained, withdrawals citing the publication method, an overrun that is expected. A reader outside the room could in principle watch the cycle: the weekly cross-checks, the reconciliations on the tenth day, the withdrawal register, the two-dayly reports. What such a reader could not do, from these lines alone, is settle in every case whether the condition was met. The gaps are of different kinds. A formal withdrawal citing the publication method is a qualifying event the line names, and needs only the date the deadline falls on to be followed. An expected expenditure over the ten-day budget is a comparison the line sets, and lacks the budget value and the grounds of the forecast behind it. The other two leave the criterion itself — what counts as a pattern different from the existing sample, what counts as unexplained — to a judgment the line does not define, and one of them also leaves the sample unnamed. That distinction matters for anyone using the record afterwards. A line with a named event and no date can still be followed if the date exists somewhere else and the reader can reach it. A line whose trigger is a judgment cannot be followed at all from the line, because the judgment produces the fact the line waits for. What would close the first kind of gap is a reference written into the record; what would close the second is a rule, or the judgment's own written grounds. That is a statement about the lines, not about the decisions they belong to. The ledger's field rows stand beside this reading: of the conditions in the question form's own reopening field, 27 of 148 are observable by the ledger's standard, against 602 of the 650 conditions written in the judgments' stop conditions.

13.3Lines that leave it to judgment

All eight lines name a role and an observation cycle. The ledger's standard counts a condition as observable when it is stated with a number, a threshold or a time, and its counts are aggregate; it does not classify these eight lines one by one. This chapter applies a stronger test — whether a reader outside the institution could reach the measure and the evidence the line turns on — and on that test the eight lines differ in degree rather than in kind. The four read next are set out in the order the questions give them.

Observation records keeper (2031).

Two things have to exist before this line can be checked by anyone. The first is the tally: a weekly record, case by case, of how long cross-checks took. That record would have to be kept somewhere a reader could reach, and the line names the verification lead who keeps it. The second is the agreed processing deadline — a number agreed somewhere else. The line refers to it; it does not state it. Without that number a reader could watch the weekly tallies and could not say whether any delay exceeded anything. A third gap sits behind the word cumulative: cumulative over what — a week, a month, a whole case? The line does not say. What would make this line observable is the agreed deadline written into the contract and quoted in the record, a cumulative window defined, and the tallies published or otherwise reachable. As written, the line promises a re-review whose condition a reader can only infer.

One line two demands (2031).

Four things are needed before a reader could check this line. Facility-level inventory figures would have to be recorded each week at a level of detail that identifies a facility. Operating restrictions would have to be recorded the same way. A baseline for additional demand would have to exist — last week's figure, or the demand as submitted — against which change can be measured. And the verifier's written confirmation would have to be produced, because the line puts the act of confirming in the verifier's hands and the observable event is therefore the confirmation rather than the underlying change. The line names the comparison, inventory against operating restrictions, and leaves out both the baseline and any threshold for what counts as a change. Without them a reader can see that a weekly comparison was done and cannot see whether additional demand moved. The allocation, the line says, is reopened in that case; a reader cannot say when that case arose.

Evacuation without stopping care (2026).

The tally is observable in principle: daily counts of intentions withdrawn or changed. The comparison is not, because capacity is not stated. Capacity of what — vehicles, lodging places, care shifts — and at what level: the whole contract, one day, one facility? The tally does not by itself establish the queue: that also needs the waiting list as it stands and what happens to it through admissions, departures and other changes. Even with the queue known, a reader would have nothing to compare it with. To become observable, the line would need capacity written down as a number for each support, or a rule stating that a queue beyond a stated length is reviewed whatever the contracted capacity happens to be. The daily tally is the easier half; the comparison is the half that decides whether review is requested. As written, the line gives the observer a count and withholds the reference point.

Unverified video disclosure (2026).

Three judgments sit inside one clause. What counts as a core contradiction; what the current viewing scope is, since the line does not describe it; and what it means for a contradiction to be one that cannot be verified within that scope. Each is the coordinator's to make, and each is a judgment rather than a count. The comparison itself is observable: two lists, compared every two days, with a record of the comparison. What would make the condition observable is a written definition of core, of the viewing scope and of what counts as unverified — or a simpler rule, that any contradiction left unresolved after the two-dayly comparison counts as core. As written, the line is a reportable act plus an unnamed judgement, and a reader can see the first and not the second.

Across these four lines different things are missing. One line refers to an agreed processing deadline and does not state it; another has no baseline for additional demand; a third has no capacity figure to set against the queue; a fourth does not describe the viewing scope it turns on. Not every trigger lacks a comparison rule: a delay beyond an agreed deadline and a queue beyond capacity are comparisons, even though the values behind them sit elsewhere in the arrangement. And whether the tallies, the comparisons and the judgments are recorded where a reader can reach them is not established by these sentences at all. Where the measure, the reference and the record are all present, a reader could watch the condition and say whether it was met. Where they are not, a reader can see that a check was performed and that someone inside the institution asked for re-judgment. The two kinds of gap fail at different moments: a missing measure fails when the check is read back, because the reader cannot say whether the number that came out was large or small, while a missing record fails earlier, because there is nothing to read back at all. The distinction matters for anyone using the record afterwards. The reopening line is what the decision said would bring it back; the check's record is what any reopening would actually have depended on. The ledger's rows record the same difference in proportion: 27 of 148 conditions in the form's own field are observable by the ledger's standard, against 602 of 650 in the judgments' stop conditions.

13.4Fantasy documents

Lee Clarke's Mission Improbable: Using Fantasy Documents to Tame Disaster (1999), as its reading note gives it, describes organisations writing plans for catastrophes they cannot control. Clarke calls these plans fantasy documents: they could not work as written, and they serve to persuade others, and the organisation itself, that the risk is under control. The note describes what such documents do for the people who write and read them. It does not say that anyone is lying. This study reads the note as saying that the document's function is reassurance as much as instruction.

That description bears on the lines read above in one respect. A reopening condition is also a document written in advance about a future event that has not happened. It says: this decision may rest on a false assumption, and here is what would show it. Its value, like a plan's value, depends on work being done later — a weekly tally, a reconciliation on the tenth day, a register with reasons attached — and on that work being recorded where someone can reach it. Where a line names a role and a cycle but no threshold, this study reads it as the point at which the document's function can drift towards the reassurance Clarke describes: the line shows that a check is scheduled, and leaves outside the record the standard by which its result would be judged, so a reader can follow the cycle and still cannot read back from the line what the check would have shown. The function drift is not a finding about these cases; it is what would have to be true for the comparison to bite, and this study marks it as its own reading.

Where the comparison does not fit. This study reads Clarke's fantasy documents as asserting, by their existence, that a risk is under control. The reopening lines assert the reverse in their structure: each states that a decision might be wrong, and names something that would show it. They are not uniformly untestable either. Of the conditions in the judgments' stop conditions, 602 of 650 are stated with a number, a threshold or a time, and even in the question form's own reopening field 27 of 148 do. The comparison also cannot be settled from this study's material. The cases are synthetic settings in fictional polities; the judgments are AI-generated and dated. Nothing in them establishes that any fictional body reassured anyone, or that any named check was ever performed. Clarke writes about organisations in the world. This study reads settings and judgments. The comparison is therefore a lens, not a verdict: it names what a line would be doing if the check were never resourced, and it makes precise what would have to be true for that to be so.

One further limit belongs here. Clarke's note describes plans for catastrophes an organisation cannot control, and the reading note does not supply a test that tells a reassurance document from an instruction document by looking at the words. What it supplies is a question: what does the writing do for the people who read it. A reopening line does less than a plan, because it promises no action at all — only that a fact, if it appears, will be treated as a reason to look again. Whether that promise holds depends on the arrangement around the line: who is named, what they are given to do the work with, and where their work goes. This chapter can read the line. It cannot read the arrangement, and it does not pretend to.

There is a further respect in which the lines and fantasy documents diverge. Every line read here names a role in the checking arrangement the decision sets up: an accounting verification office, a participation registration officer, a cost auditor, a fairness inspector, a verification lead, a demand verifier, a liaison officer, a verification coordinator. A fantasy document persuades inside as well as outside. A reopening line that names a checking role constrains those who must perform it, and it gives an outsider something to ask for — the weekly tally, the tenth-day reconciliation, the withdrawal register. Whether the outsider can obtain any of those depends on what the institution records, which is outside the line. The lines are therefore a written invitation to check that is addressed partly to people who cannot check, and partly to the institution itself. Clarke's reading names the risk in that arrangement; it does not settle it.

The comparison has one practical use, which is why it belongs in this chapter at all. It tells a reader what to look for when a reopening line is read back after the fact. A cycle that was performed and recorded leaves a reader able to follow the check; a cycle that was not performed, or was performed and not recorded, leaves the line's promise unmet. Neither state settles what the document was for: a check that was carried out can still serve to reassure, and one that was not can reflect a failure of instruction, of resourcing or of access to the record. Nothing in the record of a synthetic setting can tell a reader which happened. The record can only show what the decision said would bring it back, and how much of that a reader could see. Where the two questions of correction and observability meet — whether a decision can be changed once a signal arrives, and whether the signal can be seen at all — the first is Chapter 11's, which sets out Collingridge's dilemma of control and corrigibility; this chapter treats the narrower second.

13.5H2

H2 reads: few decisions write down in advance what would reverse them, and those that do write it in words no one could observe. The ledger states the question behind the hypothesis in brackets: is there a condition, and can it be observed? The supplied ledger gives the hypothesis and the verdicts on it and does not set out a case-level coding rule; on the reading that follows, a case would count as supporting H2 when it has no condition, or when the condition it has cannot be observed.

The ledger's verdicts are as follows. Strict: supports 0, against 50, not-applicable 0, unclear 0; verdict "against", share 0%. Lenient: the same counts, and the same verdict. Both readings put every case against H2.

What the verdict records is not that the reopening conditions are easy to check. It is the ledger's verdict on H2 as a whole, and it does not separate the frequency clause from the observability clause. Axis A5 counts a condition for revisiting the judgment in 50 of 50 cases. Both coders, strictly and leniently, returned an against verdict in all fifty cases. The verdict is the ledger's, computed by code from the two coders of different families, and this chapter reports it rather than revising it.

What an against verdict on H2 means for a reader is worth spelling out, because the hypothesis is compound and the record answers its parts differently. A verdict of against is the ledger's answer to the hypothesis as its coders read it. It is not a finding that every condition in these records can be checked, and it is not a finding about how decisions outside this collection are written. Most conditions in these records carry a number, a threshold or a time; the reopening lines of the question form carry one far less often. Both facts sit inside the same verdict, and a reader who wants more than the verdict has to go down to the lines.

That is the question such a reader actually brings: not whether H2 holds in general, but whether a particular line, in a particular question, would have let them tell whether the decision should have come back, and on what grounds. The eight readings above are an attempt to answer it for eight of the lines this collection carries, and the same three steps could be applied to any of the others.

What the verdict does not settle is the second half of H2, and the ledger's own field rows sit beside it. Of the 798 conditions, 629 (79%) are observable by the ledger's standard of a number, a threshold or a time, and that share is not spread evenly across the two fields. In the judgments' stop conditions, 602 of 650 are observable. In the question form's own reopening field, 27 of 148 are. This study reads that split as a statement about where observability lives in these records. The conditions that carry a number, a date or an event are, in the main, the ones the judgments write in arguing from the setting; the three lines the form asks for carry the smaller observable share. Both statements come from the ledger's tables, and both describe these synthetic questions and their AI-generated judgments rather than wars, states or decisions in general.

There is a reason the two fields may differ, and the reason is worth separating from the counts. A judgment writes its stop conditions while arguing from a setting: it has the premises, the options and the unknowns in front of it, and it can name the signals it would treat as failure. A question writes its reopening lines before any of that exists. It is writing about a future it has only sketched, and a role and a cycle can be promised before the work behind them is designed. This study reads the difference that way; the records do not state it. If the reading is right, the smaller observable share in the question form's field is not carelessness but the difficulty of writing a threshold before the arrangement that would produce the number exists.

That sets a limit on what the against verdict on H2 can be used for. The verdict counts cases, not lines. A reader who wants to know whether a particular decision could be brought back on checkable grounds has to work at the level of the lines, and no case-level verdict does that work for them.

The framing of H2 also matters when its verdict is read. H2's first clause is a claim about frequency — few decisions write such conditions down. On this collection that clause cannot be tested at all. What the collection can test is narrower: whether the conditions that exist can be observed. The result is a verdict of "against" with a share of 0% in both readings, and a reader should hold two things together: that no case was coded as supporting H2, and that most of the conditions a reader would actually meet in a question record are of the kind the ledger could not call observable.

Where this leaves the chapter's own question. If a reopening condition is what a decision says would bring it back, then observability is what decides whether anyone can later say whether the decision came back for the right reason. The eight lines read here divide on that point rather than on their presence. The ledger counts all eight lines as conditions, and counts 50 of 50 cases as having them. What the counts do not carry is the difference between the ledger's standard of observability — stated with a number, a threshold or a time — and the stronger checkability this chapter has been asking for: a reference value a reader could reach, evidence a reader could obtain, and a criterion a reader could apply without the institution's help. That difference is what this chapter has read in the lines themselves.

Chapter 15 sets H2 beside the verdicts on H1 and H3 through H8, and Chapter 16 sets the findings beside the earlier theories. The next chapter takes the eight hypotheses' neighbouring question: where, in these records, the thirteen judgments part from one another.