12.1How a verdict is reached
Each of this study's five hypotheses was written before any case was recorded and registered on 6 October 2026, beside the eight fields the decision ledger holds. Chapter 2 gives the fields and the hypotheses in full; this chapter reports how far each of them held, and what the counts say where it did not.
Before any count in this chapter is read, the marks at which a hypothesis is called supported or against are fixed. That is the section's claim and its addition to the debate. The four literatures in Chapter 1 study emergencies by their causes, their institutions and their effects; the reading notes on rules and discretion — Kydland and Prescott on a plan that ceases to be best when the future arrives, Bordo and Kydland on the gold standard as a rule carrying an escape clause — argue about what a commitment does under pressure. This study registers the same kind of commitment about its own reading: the marks below were set on 6 October 2026, before any case was coded, so that each verdict in the sections that follow is a count measured against a rule fixed in advance rather than a judgment formed after seeing the count. What this section adds to that debate is only the marks, and what each verdict word means.
The rule is the same for every hypothesis. The cases a hypothesis applies to are counted in four kinds. Supports means the coding bears out the hypothesis in that case. Against means the coding cuts against it. Unclear means the reading did not settle either way. Not applicable means the case's record cannot hold the field the hypothesis needs at all — the field is absent from the record, not merely empty there.
The rule set before any case was recorded runs as follows. A hypothesis is supported when supports reach 60% or more of the cases it applies to, and against when they reach 40% or less. Between those marks the verdict is unclear. A third route to unclear was registered as well: when unclear cases outnumber the decided ones, the verdict is unclear whatever the percentage shows. The percentages are taken over the cases the hypothesis applies to; cases read as not applicable stay out of them and are reported beside them, because a hypothesis that applies to few cases is a hypothesis about a small set, and the reader should be able to see how small. In this ledger the sizes differ sharply: across the eighty-four cases, one hypothesis reaches every case, and another leaves four to be judged on the lenient reading once the not-applicable cases are set aside.
Two readings of every count. The ledger was recorded twice, as two independent codings of one model run over the same records. In every count here the strict reading counts only what both codings found stated in the record's own words; the lenient reading also counts what at least one coding found implied from the setting of the case. Where the two codings left a value unsettled, it counts as unclear in both. Where the ledger gives a share, it is the share of the decided cases that support, reported as the ledger reports it. Where no share is given, the ledger gives none.
Two cautions travel with these verdicts, and each is used in the sections that follow. A share carried by two or three decided cases carries almost no weight, however large its percentage, and the decided count is given beside every share here. And the two codings were told to read each case's own sentences only: Skeptron's recurring frame sentences, which Chapter 2 describes, were kept out of the counts entirely, so that a verdict cannot be lifted out of the studio's method and read as something the records show.
12.2F1 and F2: the end of an act
The four literatures Chapter 1 maps study emergency acts through their effects, their authors or their legal form. None of them takes as its object the record a later archive keeps of an act's own terms. F1 asks whether that record names the condition, date or event on which the act would end or be revised, and F2 asks whether an act presented as temporary proved to have outlasted its first term. What this section adds to those debates is a third thing beside doctrine and effect: the state of the records on which arguments about exit clauses would have to be judged. Across eighty-four cases, from the code of Hammurabi to the Swiss ordinance of March 2023, the record names what an act did far more often than what would end it.
Strict: supports 28, against 4, unclear 29, not applicable 23; share 88%. Lenient: supports 30, against 6, unclear 8, not applicable 40; share 83%. Verdict: supports, on both readings. The registered prediction held on one reading only: it expected support on the lenient reading and unclear on the strict, the lenient reading supported at 83%, as expected, and the strict reading supported at 88% instead of being unclear.
The verdict does not depend on how generously the records are read. Where a case shows an act at all, its own text seldom hands a later reader a date or a condition on which the act would end. Across the ledger the exit condition reads observable 6, vague 2, not-named 28, unsettled 8, not applicable 40. The ledger separately reports that cases with a condition for revisiting the study's judgment stand at 0 of 84, with conditions in all totalling 0. The two counts measure different things — one asks what a case's own text says about an act's end, the other whether a record sets a trigger for its own revision — but they point the same way.
Where an actor wrote an end down, it is usually in the text the record summarises, and reading those texts first shows how differently an end can be written. The bank holiday of March 1933 was written the other way, in a proclamation that fixes its own end: "do hereby proclaim, order, direct and declare that from Monday, the Sixth day of March, to Thursday, the Ninth day of March, Nineteen Hundred and Thirty-three, both dates inclusive, there shall be maintained and observed by all banking institutions and all branches thereof located in the United States of America, including the territories and insular possessions, a bank holiday, and that during said period all banking transactions shall be suspended". The address of 15 August 1971 announces the end of gold convertibility with an adverb and no date: "I have directed Secretary Connally to suspend temporarily the convertibility of the dollar into gold or other reserve assets, except in amounts and conditions determined to be in the interest of monetary stability and in the best interests of the United States."
The amendment of 19 March 2023, Article 14a(2), reads "Die Garantie zur Verlustabsicherung beträgt höchstens 9 Milliarden Schweizer Franken" — in this study's English, the guarantee against losses amounts to at most 9 billion Swiss francs. That cap is not an end-date at all; it is a limit on the state's exposure, and the same article names who bears the first losses on the assets to be wound down. The joint statement of 12 March 2023 names no term for its measures.
Against these texts the ledger's reading is plain: six cases carry an observable end, two carry general wording, and twenty-eight name none. The twenty-four cases the book reads closely show the same shape — not-named 14, observable 4, vague 1, unsettled 4, not applicable 1. Four carry a condition a later reader could check: the condition in the record of the restriction of 1797, the next harvest of the green sprouts loans, the ninety days of Mexico's deferral, and the days the proclamation of March 1933 names for the bank holiday. One carries general wording, attached to Korea's stand-by arrangement of December 1997. Fourteen name none at all.
Group by group, F1 holds in every group that contains acts. Strict: crisis history 9 for, 1 against, 6 unclear; credit history 15 for, 3 against, 10 unclear; markets 1 for, 0 against, 1 unclear; money 3 for, 0 against, 2 unclear. The lenient readings move in the same direction — credit history 16 for, 5 against, 4 unclear — and the price windows, which hold no act at all, are unclear on both readings. Skeptron's price windows record observed prices up to one closing price; they hold no decision and no actor, and this study reads them only as a group. The single strict reading to go against F1 comes from the crisis history: the bank holiday, whose proclamation names the days of the holiday in its own words.
F2 asks whether a temporary act outlasted its first term, and here the counts stop the question before it starts. Strict: supports 0, against 0, unclear 5, not applicable 79. Lenient: supports 0, against 1, unclear 3, not applicable 80; share 0%. Verdict: unclear on both readings. The registered prediction — that F2 applies to fewer than ten cases and is supported among them, naming the bank restriction, Britain's suspension of gold and the closing of the gold window — held only in its first half. It applies to fewer than ten cases, as expected; among those it names, the counts found no support. Seven cases stand as term-stated, two as temporary without a term, and twenty-seven say neither that an act was temporary nor that it carried a term. On what became of an act, the ledger reads not-said 38, not-said implied 1, unsettled 3, not applicable 40, with one case read as ended as stated — the bank restriction, whose period 1797–1821 is given by the Bank of England's own chronology as the record's later presentation of it — and one case in the crisis history read as reversed or ended early.
The lesson F2 teaches is a limit of the question rather than a finding about emergencies. This study reads the ledger's counts as showing that an argument about whether temporary measures become permanent needs records that state a term and record an outcome; these records do neither in the great majority of cases. Where the source texts do supply terms — the proclamation's four days, the Swiss ordinance's six months, the Federal Reserve exception's date — they are documents the cases summarise rather than documents the cases hold. Whether such a term should have been written at all is the doctrine's question; whether the record keeps one is what F1 and F2 can answer, and on that the answer is that it usually does not.
12.3F3 and F4: dates and figures
Emergency acts in money and credit are recorded as single dated acts and, where a sum appears, as single figures. The work of separating one moment from another — an announcement from an effect, a headline sum from what was available — falls on the reader of the record, and where an actor's own text survives it is usually that text, not the archive's summary, that carries the second date or the second number.
F3 asks whether a case's own text supplies two dated moments at all. Strict: supports 12, against 23, unclear 26, not applicable 23; share 34%. Lenient: supports 13, against 28, unclear 3, not applicable 40; share 32%. Verdict: against, on both readings. The prediction registered with F3 did not hold.
Skeptron's frame instructs the reader to keep the moments apart; that sentence is never counted as a finding. F3 asks something narrower: whether the case's own text names two dated moments, so that the separation is possible from the record alone. The count neither contradicts the frame nor is borne out by it. What it shows is that the instruction does work the records do not do for the reader. Where a case carries two dated moments — named (stated) 18, not-named 53, not-named (implied) 7, unsettled 6 — a later reader can check one moment against the other; where it does not, the reading is left to the frame, to the actors' own texts, or to Chapter 2's method.
The source texts show what a second date looks like in an actor's own sentence. Roosevelt's proclamation of 6 March 1933 sets its own term in inclusive words: the holiday runs "from Monday, the Sixth day of March, to Thursday, the Ninth day of March, Nineteen Hundred and Thirty-three, both dates inclusive", with all banking transactions suspended in between. Nixon's address of 15 August 1971 announces the suspension of convertibility "temporarily", and calls the ten per cent import surcharge "a temporary measure", without a date or a condition of return. The Swiss Federal Council's emergency ordinance of 16 March 2023 states its own duration in Article 15(2): "Sie gilt für die Dauer von sechs Monaten ab Inkrafttreten." — this study's English: it applies for a period of six months from entry into force.
Skeptron's records are what an archive keeps in 2026: short, dated, thin on the second moment. Among the twenty-four cases read closely, six carry two dated moments: the order of 194 lending grain from the third month to the seventh and taking it back in the tenth; the bank holiday's three dates in March 1933; the closing of the gold window announced on 15 August 1971 with the preceding days' talks kept apart; sterling's suspension on 16 September 1992 and the inflation target of 8 October; the LTCM meeting; and the Greek ceiling of 28 June 2015 measured against the level of 26 June. Seventeen of the twenty-four name two dated moments in no words of their own, and one is unsettled between the two readings. The Volcker announcement records a change of operating procedure on a single day, with no second date beside it.
Group by group, strict: crisis history 4 for, 7 against, 5 unclear; credit history 5 for, 14 against, 9 unclear; markets 0 for, 1 against, 1 unclear; money 3 for, 1 against, 1 unclear; price windows unclear. The lenient readings follow the strict ones — credit history 5 for, 19 against, 1 unclear — which is unusual in this ledger and means the case against F3 does not rest on any caution in reading the records. The money group, six cases, is the only group to support F3 at all, at 75% strict and 80% lenient. The price windows name a closing price up to which prices were observed and hold no act, so their two dated moments belong to the frame and not the case.
On figures, F4 asks whether a record sets the sum announced apart from what was available, drawn, paid or settled later. Strict: supports 7, against 26, unclear 24, not applicable 27; share 21%. Lenient: supports 7, against 31, unclear 16, not applicable 30; share 18%. Verdict: against, on both readings. The prediction registered with F4 did not hold. The field behind it reads announced-and-available-apart (stated) 7, single-figure (stated) 26, single-figure (implied) 5, no-figure 30, unsettled 16. A single figure is the norm wherever a record gives a sum at all: one number for a loan, a line, a ceiling or a commitment, with nothing beside it to show how much could be drawn or how much had been drawn.
The source texts again separate announcement from availability where an actor's own text survives. The Federal Reserve's release of 14 September 2008 raises the total offered under one facility from $175 billion to $200 billion — a ceiling, not a drawn sum. The FDIC's release of 10 March 2023 gives total assets of about $209.0 billion and deposits of about $175.4 billion at 31 December 2022, and says at the same time that the amount of uninsured deposits was undetermined at closing: the announcement and the available figure stand apart because the second could not yet be stated. The Swiss Federal Council's amendment of 19 March 2023 caps the state's guarantee in Article 14a(2): "Die Garantie zur Verlustabsicherung beträgt höchstens 9 Milliarden Schweizer Franken." — this study's English: the loss-protection guarantee amounts to at most 9 billion Swiss francs, with the first losses borne by the acquiring bank itself. On LTCM, Greenspan's testimony of 1 October 1998 draws the same line in the other direction: "This agreement was not a government bailout, in that Federal Reserve funds were neither provided nor ever even suggested."
Among the twenty-four cases read closely, seventeen carry no figure, three carry a figure with availability kept apart, one carries a single figure, and three are unsettled. This study reads the three with availability kept apart as showing that where a public body announces a staged programme, the record can and sometimes does hold the announcement apart from the availability.
Group by group, F4 goes against in the crisis history, the markets and the money groups and is unclear in the credit history; the price windows' own row reports strict against 16 and unclear 13, and lenient against 20 and unclear 9. Crisis history strict: 1 for, 2 against, 3 unclear. Markets strict: 0 for, 1 against, 0 unclear. Money strict: 1 for, 3 against, 1 unclear. The credit history is the widest split in the ledger between the two codings: strict 5 for, 4 against, 7 unclear, and lenient 5 for, 5 against, 4 unclear, at 50% share. It lands on the group that holds the international programmes and the deferrals. Where a figure exists, its wording is the hardest thing in these records to read the same way twice.
12.4F5: who paid
The literature on who pays in a crisis argues about effects rather than about documents: Reinhart and Rogoff's catalogue documents that banking crises are often followed by long and costly recoveries; Radelet and Sachs argue that the sudden withdrawal of foreign lending exposed the countries of East Asia to runs whose cost fell on them; Feldstein argues that heavy conditionality in the Asian programmes intruded on national policy. Short public records cannot settle those arguments; they can show only whether their own words say who waited and who paid. On that question F5 supports: strict reading 31 supporting, 1 against, 29 unclear, 23 not applicable, a share of 97%; lenient reading 35 supporting, 3 against, 6 unclear, 40 not applicable, a share of 92%.
The field behind it reads named (stated) 1, named (implied) 2, not-named 34, not-named (implied) 1, unsettled 8, not-applicable 38.
Roosevelt's fireside chat of 12 March 1933 addresses holders of money: "I can assure you that it is safer to keep your money in a reopened bank than under the mattress." The proclamation of 6 March 1933 orders "all banking institutions and all branches thereof located in the United States of America" to observe the holiday and that "all banking transactions shall be suspended"; those who held the balances are not named in the order, and appear only in the address that explains it. Silber's note on the episode argues that reopening restored confidence not only because banks were declared sound but because the government and the Federal Reserve signalled an implicit guarantee of deposits, backed by the Federal Reserve's willingness to lend against a wider range of assets under the Emergency Banking Act: what a guarantee did, not what the record holds.
The Joint Statement of 12 March 2023 names where losses will fall: "Any losses to the Deposit Insurance Fund to support uninsured depositors will be recovered by a special assessment on banks, as required by law." One class is named: the banks carrying the assessment. The record for the case of 10 March 2023, the closure of Silicon Valley Bank and the appointment of the FDIC as receiver, reads the bearer as not named. This study reads the difference as one of object rather than of fact: the statement is three officials speaking; the record is a five-question summary that did not ask whom the assessment would reach. Metrick's note describes a bank with a concentrated base of uninsured depositors and an exception to the usual insurance limit that protected all depositors — depositors left out of the record.
The oldest act names its parties literally. Hammurabi's law 48, in L. W. King's 1910 translation, reads: "If any one owe a debt for a loan, and a storm prostrates the grain, or the harvest fail, or the grain does not grow for lack of water; in that year he need not give his creditor any grain, he washes his debt-tablet in water and pays no rent for this year." Both sides are named in the rule's own words: the debtor who pays nothing that year and the creditor who receives neither grain nor rent. The record reads the bearer as not named; this study reads the difference as the summary's silence rather than a gap in the law. Diocletian's edict of 301, whose preamble the record reads as blame laid on traders, is the opposite shape: those who pay are named there as the cause of the trouble, not as its bearers.
By group. Crisis history strict: 9 for, 0 against, 7 unclear, share 100%; lenient 9 for, 1 against, 2 unclear, share 90%. Credit history strict: 17 for, 1 against, 10 unclear, share 94%; lenient 20 for, 2 against, 3 unclear, share 91%. Markets strict and lenient both 1 for, 0 against, 100%, one unclear on the strict reading. Money strict: 4 for, 0 against, 1 unclear, share 100%; lenient 5 for, 0 against, 0 unclear, share 100%. Skeptron's price windows record observed prices up to one closing price; they hold no decision and no actor, and this study reads them only as a group. Price windows strict and lenient are both unclear, because the windows hold no act and therefore no bearer. The credit history holds the single case counting against F5 on the strict reading, two of the three on the lenient reading, and the greatest disagreement between the readings, unsettled 3 against 20 read as not-named. On the ledger's bearer field the two codings agree in 62 of 84 cases, with 54 read as unknown and 22 unsettled; "not named" and "unknown" are different answers. Skeptron's price windows are twenty-nine cases with no act, recording observed prices up to one closing price, with no decision and no actor, and this study reads them only as a group.
Among the twenty-four cases read closely the bearer field reads not-named 17, unsettled 5, not-applicable 1, named 1. The single case read as naming a bearer is the bank holiday of March 1933; Chapter 3 reads that case in full.
Where a bearer is named it is named incidentally: because the act addressed a class of people, as a price edict addresses traders, or because the closure names those shut out of their money. This study reads that as an absence in the form of the record, not an argument that anyone was spared: the question is not whether a rescue worked but whether the record lets a reader find who bore the wait. Two works beside this result cannot be tested by it: Tucker's argument that rescue decisions are political choices needing rules set in advance, and Higgs's account of powers that expand in crisis and not fully return. Both leave open who carried those choices, and the records state no outcome and no condition of ending from which that could be read. What F5 supports is a fact about the form of the record.
12.5The test
The study's five fields, taken together, make a single test of the records. One case in forty-two passes all five, and one in the twenty-four the book reads closely.
The denominator of 42 is the ledger's own count of cases with an act — the records where an authority did something to money, credit or a market, as distinct from the records of a market's movement or a window of observed prices. The single case that passes all five conditions is the bank holiday of March 1933: an act presented as temporary (the record gives no term, though the proclamation names the days 6 to 9 March), an observable end in the reopening, a bearer both readings found named, and three dated moments kept apart in the record — the order of 6 March, the Act of 9 March, and the reopening of banks from 13 March. The composite is therefore not a description of how emergency acts are usually written. It is a description of one record in the ledger, and of one in twenty-four of the cases read closely.
Two further counts belong beside it. The ledger reports 82 of 84 cases with a deadline in the record, and 84 of 84 cases that list what is missing, with 232 items listed in all. A deadline is not a condition of an end: a case can carry a calendar check — the fourth question of every Skeptron case checks a weekday by arithmetic — without the record saying anything about when an act would be revised or lifted. And the ledger reports 0 of 84 cases with a condition for revisiting the judgment, with conditions in all totalling 0. Nothing in these records sets a number, a date or a threshold that would require the reading to be redone.
Agreement between the two readings on the fields that make up the test is high where it matters most. The two readings agreed on whether an act was present in 82 of 84 cases, 98%; on temporary in 76 of 84, 90%; on the exit condition in 76 of 84, 90%; on who bore the cost in 76 of 84, 90%; on two dated moments in 78 of 84, 93%; on what became of the act in 81 of 84, 96%. Agreement is lower on figures, 68 of 84 or 81%, which is consistent with F4's split in the credit history. Where the two readings disagreed, the ledger counts the case as unclear rather than choosing between them, so no verdict anywhere in this chapter rests on a case the two readings left open.
What the test shows is narrow and should be stated narrowly. Among forty-two records of emergency acts in money and credit, spanning the code of Hammurabi to the takeover of Credit Suisse in March 2023, one record carries enough in its own words for a later reader to judge the act by its end. The other forty-one leave at least one of the five conditions unnamed. That is a fact about records, and Chapter 13 sets it beside the literatures that study these acts.
12.6The series' hypotheses
The ledger carries eight hypotheses as well, belonging to the series rather than to this study's question, and recorded on the fields of Skeptron's records. They are named here by what they say, with the counts as the ledger gives them. Where a hypothesis asks about a field these records cannot hold, the verdict not applicable is a statement about the field, and not a finding about emergencies.
The first of the eight hypotheses concerns who bears the cost while a decision waits. Strict: supports 0, against 0, unclear 84, not applicable 0. Lenient: supports 8, against 0, unclear 76, not applicable 0; share 100%. The hypothesis is unclear on both readings, and the counts show why. The field that decides it is who bore the cost, and on that field the two readings reached settled values in 62 of 84 cases, with 54 read as unknown and 22 unsettled. The field of who decides is in even worse shape: the two readings agreed in 56 of 84 cases, with 36 cases read as unknown and 28 unsettled, and the ledger reports cases with the decider in the record at 0 of 84. Eight lenient supports against zero against is not support; it is eight readings where a setting implied a bearer that no sentence named. A verdict of unclear here, on a hypothesis the series's own records mostly cannot hold, is not a finding about who pays.
Few decisions write down in advance what would reverse them, and those that do write it in words no one could observe. Strict: supports 84, against 0, unclear 0, not applicable 0; share 100%. Lenient: the same, 84 supports and share 100%. This is the only hypothesis in the ledger that both codings support in every case, and it is also the only one where the codings agree completely. The field behind it — a condition for revisiting the judgment — is present in none of the eighty-four cases, with conditions in all totalling 0. The hypothesis is supported at full strength, and its support is the same fact as F1's: no record sets a trigger for its own revision.
Where judgments diverge, material is missing. Not applicable on both readings, in all 84 cases. Divergence is a field the ledger does not carry for these records: the two readings did not read the cases as divergent among themselves, and there is nothing for the hypothesis to apply to. The verdict says nothing about whether missing material causes disagreement.
The harder a decision is to undo, the further apart are those who decide and those who bear the cost. Strict: supports 0, against 0, unclear 0, not applicable 84. Lenient: supports 2, against 0, unclear 0, not applicable 82; share 100%. The strict reading cannot apply at all, because reversibility was read as unknown in 40 cases and unsettled in 35, and the decider is usually absent as well. The lenient share of 100% rests on two cases out of eighty-four; a percentage carried by two decided cases is reported here as the ledger reports it and is given no weight. On this ledger the hypothesis cannot be tested, and the honest report is that.
A restriction outlives the memory of its reason. Strict: supports 0, against 59, unclear 25, not applicable 0; share 0%. Lenient: supports 6, against 61, unclear 17, not applicable 0; share 9%. Verdict: against, on both readings. Here the field does exist. The two readings agreed on whether a reason is recorded in 67 of 84 cases, 80%, reading 59 cases as stated, 2 as implied yes, 6 as implied no, and 17 unsettled. The six lenient supports are the cases where a reason was inferred to be absent, and they are far outnumbered by the cases where the reason stood in the record's own words.
The longer the horizon, the less is decided now and the more conditions are set for revisiting. The ledger gives this hypothesis the verdict read across the books in Volume 0, because the series as a whole carries the cases its horizon needs and this study does not. What this ledger can report beside it is that on horizon the two readings reached a settled value in 49 of 84 cases, 25 of them read as unknown and 35 left unsettled, and that on conditions the count stands at 0.00 per case at every horizon the readings could read: days to months on 16 cases under the lenient reading, years on 5, decades on 3, all with zero conditions per case. Where a horizon was legible at all, no case set conditions for revisiting.
The more a decision is left to an automatic rule or to AI, the more the human's place — approval, exception, review — is empty or a name only. Strict: supports 0, against 0, unclear 0, not applicable 84. Lenient: supports 5, against 0, unclear 0, not applicable 79; share 100%. On who decides, the readings are human-decides (implied) 30, ai-or-rule-decides (implied) 5, unknown 10, unsettled 38, human-decides (stated) 1; on whether the human's place is specified, they are no (implied) 53, unsettled 23, no (stated) 5, not-applicable (implied) 3. Every one of these eighty-four records is a public record; where a case shows an act, that act was taken by officials and institutions, and none of the case texts puts an automatic rule or an artificial system in the chain of decision. The five lenient supports come from settings one coder read as rule-governed. A verdict on five cases and a field the records do not hold is not a finding about automation.
The options that cut the immediate loss most often give up evolvability and resilience among the four higher goals. Not applicable on both readings, in all 84 cases. The ledger reports options in all eighty-four cases at zero. A hypothesis about what the most loss-cutting option sacrifices cannot be judged where no record names an option at all; the series' hypothesis about a trade-off among options has nothing to trade off in these records.
The overall shape of the series' verdicts on this ledger is therefore this: one hypothesis supported at full strength, one against, one clear on a field the records cannot hold, one clear where the field is mostly absent, two untestable here because the record cannot carry the question, one carried by two cases on a lenient reading, and one read across the books in the series.
12.7The predictions that failed
Five registered readings were not borne out by the counts: the strict half of the prediction registered with F1, and the whole predictions on F2, F3, F4 and F5. The study reports each with what it teaches.
On F1 the lenient reading held, as the prediction expected, at 83%; the strict reading was expected to be unclear and instead supported at 88%, so the strict half of the prediction failed. What the study learns is that where both readings found the record's own words stating an end for an act, those words were common enough to carry the verdict under the stricter test as well; the cases the prediction expected to fall away when only stated wording is counted did not fall away.
The prediction that an act presented as temporary would prove to have lasted longer than first said, or to have become permanent, was not borne out on either reading: strict unclear with zero supports among the five applicable cases, none of them decided, and lenient unclear with share 0% among the four applicable cases, only one of which was decided. What the study learns is a limit of the field rather than a result about emergency acts. A prediction about an act's term needs records that state a term and record an outcome, and these records give twenty-seven cases that say neither. The lesson is that the question "did it outlast its term?" cannot be asked of a set of records that mostly do not write terms down.
On two dated moments the counts went against the prediction on both readings, and what the study learns is that the separation of announcement from result is a reader's job, not a record's gift. Seventeen of the twenty-four close cases do not name two dated moments, and the field across the ledger reads not-named in the great majority of decided cases. This is the prediction most closely entangled with Skeptron's own frame, and the result should be read with the warning given in the previous section: the coders read the cases' own sentences only, so the count measures the record, not the studio's method.
On figures the counts went against the prediction on both readings, and what the study learns is that a single figure is the ordinary form of a public sum. Only seven cases in the ledger keep announcement and availability apart, and they cluster in the credit history, where the two readings split most — 5 for, 4 against, 7 unclear on the strict reading, and 5 for, 5 against, 4 unclear on the lenient. A figure announced without its availability beside it cannot be checked later against what was drawn; that is what a reader loses, and it is the reason F4 is worth registering even though it failed.
On F5 the prediction expected the verdict to be unclear rather than supported: it held that these records would name depositors, debtors and holders more often than the epidemic records named the sick, and so fall below the 70% found in the first study of the series on its own question of who bore the cost while an act waited, while staying above 40%. The counts went the other way — strict 31 supporting, 1 against, share 97%; lenient 35 supporting, 3 against, share 92%. What the study learns is that a record naming a class of people — those shut out of their money, those excluded from protection, the banks carrying an assessment — is enough to settle whether the record names a bearer, and that naming a class is not the same as measuring what it cost. Among the twenty-four cases read closely one names a bearer in its own words, the bank holiday of March 1933, and it names it incidentally, in a closure that addresses holders of money.
One further item, not a registered prediction, should be named as a question left open: the series' hypothesis that longer horizons bring more conditions for revisiting was given the verdict read across the books in Volume 0, and its own horizon table shows zero conditions per case at every horizon legible in this ledger. The study learns from that only that this book's cases cannot settle it.
The findings are set beside the literatures in Chapter 13.