14.1The answer in brief
Put briefly, the question was whether an archive's summary of an emergency act tells a reader how long the act was meant to last and who paid for it. On Skeptron's records the answer is mostly no, and the no has a shape. The act is kept: in 42 of the 84 cases the record shows an authority acting on money, credit or a market, and in the credit history it does so in 25 of 29. The end is mostly not kept: where there is an act, the record seldom names an observable condition for its end. The bearer is kept least of all. And the actors' own texts, where they survive, name the end and the bearer more often than the archive's summaries do.
14.2The end of an act
The first hypothesis held. On the lenient count it stands at 83% (30 to 6, with 8 unclear and 40 outside its reach); counting only what both readings found stated, it stands at 88% (28 to 4). It holds in each group with acts: 9 to 1 in the crisis history and 16 to 5 in the credit history on the lenient reading. The archive does keep endings, but mostly in the studio's own recurring sentences about what would keep or withdraw a comparison, which this study does not count.
The actors' texts tell a different story where the study has them. They name the end in at least four forms. Three name a calendar: the bank holiday of 1933, the Board's exception of 2008 and the Swiss ordinance of 2023. Some name an event: the law as it stood in 1810 tied the resumption of cash payments to the ratification of a definitive treaty of peace, and the Bullion Report proposed two years, peace or war. Some name another body's decision: the Privy Council's minute of 1797 ran "until the sense of parliament can be taken", and the Act of 1931 "Unless and until His Majesty by Proclamation otherwise directs". And some call the act temporary and name nothing: the address of 15 August 1971 suspended convertibility "temporarily". The oldest of the acts read here names its end most exactly: Hammurabi's law 48 relieves the debtor "in that year" in which the storm prostrates the grain.
The second hypothesis, that a temporary act outlasts its term, could not be judged: it applies to too few cases, on the lenient reading one against and three unclear, and none in support. Records that do both are rare. The Bank Restriction is the clearest exception in the close reading: its record gives the period 1797–1821, and the minute of 1797 and the Bullion Report of 1810 give its conditions. What the study can say is that the archive rarely lets a reader ask the ratchet's question at all.
14.3Dates and figures
Two predictions about the records' form failed, and their failure is a finding about the archive. The archive's titles had led this study to expect two dates in most cases with an act. Instead the third hypothesis fails, with about a third of those cases in its favour on either reading. The dates are in the titles and in the frame; the cases' own sentences give one date more often than two. The money cases are the exception, supporting it 4 to 1. The bank holiday of 1933, with its three dates, is the exception that shows what the rule leaves out.
The study also expected the records to keep a headline figure apart from what could be drawn. Fewer than a fifth do on the lenient count; only the credit history is evenly split. Where a record does keep the figures apart, it is because its source did: Korea's SDR 15.5 billion approved over three years against SDR 4.1 billion available at once; the two lines of up to 100 billion francs in the Credit Suisse record, one with priority in bankruptcy and one with a federal guarantee. The Swiss amendment of 19 March 2023 goes further than the record, capping the federal loss guarantee at 9 billion francs and making it payable only after the acquiring bank has borne 5 billion of losses on the assets to be wound down.
14.4Who paid
The fifth hypothesis held most strongly of all. In 35 cases with an act the record does not name who bore the cost, against 3 that do, on the lenient reading (92%); on the strict reading 31 against 1 (97%). The pattern is the same in the crisis history (9 to 1) and in the credit history (20 to 2).
This is where the actors' texts add most. The joint statement of 12 March 2023 says that "Shareholders and certain unsecured debtholders will not be protected." and that losses to the deposit insurance fund from supporting uninsured depositors would be recovered by a special assessment on banks. Greenspan's testimony of 1 October 1998 says that the agreement on LTCM "was not a government bailout, in that Federal Reserve funds were neither provided nor ever even suggested", and grants that any public involvement that softens private losses raises the question of moral hazard. The FDIC's release of 10 March 2023 divides depositors into insured and uninsured and gives each a different path. Hammurabi's law names the debtor and the creditor. The archive's summaries keep few of these sentences.
14.5The test
One case of the 42 with an act passes the test: the bank holiday of 1933. Its record presents the act as temporary, names an observable end in the reopening, names who bore the cost, and keeps three dated moments apart. Where every family of reader had to agree, the test still passes in one case of 39. The study predicted that fewer than one case in ten would pass; the prediction held, and the margin is wide.
14.6What each kind of act showed
The suspensions read in Chapter 3 named their end in four different forms, and the archive kept the form for one of them. The minute of 1797 tied the end to Parliament, the law as the Bullion Report found it in 1810 tied it to peace, and the Report proposed two years instead; the record keeps the period 1797–1821 and says the obligation was suspended. In 1931 Parliament left the end to the Crown and gave a fixed term only to the Treasury's orders on the exchanges, while the archive kept the date and the change of regime. The 1971 address promised an end without naming one, and the summary of it records the weekend of talks before the broadcast. The proclamation of 1933 named its days, and its record kept three dates.
In Chapter 4 the authority seldom lends in its own name: in 1907 it refuses, in 1998 it convenes, in 2008 it widens its facilities while a holding company fails, and in 2023 it closes a bank and appoints a receiver. Who was protected and who was not is in the actors' texts for the cases of 1998 and 2023, and mostly not in the records.
The reliefs of debtors read in Chapter 5 are the oldest acts in the archive and, read in their own texts, the most exact about their term: a year of crop failure, a season of lending and a month of return, the next good harvest. Their records add the warning that a law or a proposal is not a record of its enforcement, and give no rate of interest and no result.
The foundings read in Chapter 8 are emergency acts meant not to end. The records keep the date of each charter or agreement apart from the opening of the institution or a country's entry into it, and say nothing of how the charter itself could be amended or ended. The rules abandoned in Chapter 9 were each left on a dated day; the records give the day, and for 1992 a second date, the inflation target of 8 October, but seldom what was meant to replace the old rule or for how long.
The market records of Chapter 10 hold no act, and so they show the ledger's fields at their limit: a crash of two days given as two percentages that are not to be added, a company's charter, conversion and collapse kept apart, and twenty-nine windows of prices that end on one close each.
14.7The groups compared
The archive's groups differ less than the study expected. The share of cases with an act is highest in the credit history, as predicted; the crisis history records more market movements without an act. The price windows, with a single closing figure each, push the fourth hypothesis down for reasons unrelated to emergencies. The first and fifth hypotheses hold in every group with acts. The third goes against in every group but the money cases.
14.8What the actors' own texts added
The source texts were the study's second witness, and they changed what it could say. Where the summaries were silent about endings, the texts usually were not. They named bearers the records left out. They gave the reasons the records left out: the minute of 1797 records the unusual demands for specie and the fear of a want of cash for the public service; the Board's release of 2008 records the weekend's discussions about the unwinding of a major financial institution; Volcker's press conference of 6 October 1979 explains the move from the federal funds rate to the supply of reserves, with the warning that the daily rate in the market would be apt to fluctuate. Where the study had no source text, as for Mexico in 1982, Korea in 1997 and Greece in 2015, it could say only what the records and the reading notes say.
The texts also show the limits of the second witness. Some were had only as transcriptions not checked against the printed page, and the study read them for their sense and did not quote them. Others, for the cases of the 1980s and 1990s, belong to institutions whose terms do not allow them to be reproduced here.
14.9The series' hypotheses on this archive
Of the series' hypotheses, two can be read here and the rest cannot. The hypothesis that few decisions write down what would reverse them is supported in all 84 cases on both readings, but this is because Skeptron's history records hold no field for a condition of revision at all; by the series' own rule, a property of the archive's form is not a finding. The series' guess that restrictions outlive their reasons fails here, because the summaries nearly always say why an act was taken. Who decides, who bears the cost and whether an act can be undone are mostly unknown or unsettled in these records, which keep events rather than decisions; the hypotheses that rest on those fields are unclear or do not apply.
14.10The predictions that failed
Seven predictions were registered before coding. Four failed. The first hypothesis was predicted to be unclear on the strict reading; it is supported on both. The second was predicted to be supported among a handful of cases; it cannot be judged. The prediction of at least seven in ten for the third came out near three in ten. The fourth was predicted to be supported in the crisis and credit histories; it goes against overall and is unclear in the credit history. The fifth was predicted to be unclear; it is supported at 92%. The test and the shape of the groups came out as predicted.
The failures have one source. The study's designer read the archive's titles, and the titles promise gaps, figures and bearers that the cases' own sentences mostly do not carry. The first study's predictions failed in the same proportion, four of seven, for a related reason: there, the archive's later readings stood in for the actors' record of the hour.
14.11Beside the first study
On the epidemic records, 70% of the decided cases left the cost-bearer of a waiting act unnamed; on these records, 92% do. A gap in time was named in 62% of the acts on disease, and two dates are kept apart in 32% of the acts on money. The first study's records of acts on disease were richer in dates; this study's records of acts on money are poorer in bearers. Both studies find that a later archive keeps the act and drops much of what would let a reader judge it in its own time. The readers of other families moved one of the first study's hypotheses from one side of the line to the other; here none moved.
14.12The archive's own sentences
Much of what these records say about an end is said by the studio, not by the case. Almost every case closes with a question about the condition of a later comparison, and its answer names what to observe and on what finding to withdraw the comparison: for Korea in 1997, liquidity by maturity, the borrowing entities and the record of what was actually carried out; for the Bank Restriction, quantities, prices and whether trade was possible after the control. These sentences are careful and often useful, but they are the archive's conditions for its own comparisons, written in 2026, and they recur from case to case with only names changed. This study counted none of them as the actors' condition for ending an act. Had it counted them, the first hypothesis would have gone the other way, and the study would have reported the studio's method as a finding about history.
The same is true of dates. The studio's own method insists on separating announcement from result. The third hypothesis asks whether a case's own sentences keep two such moments apart, and most do not. The titles and the frame promise the gap; the cases' sentences mostly give one date. A reader who takes the archive's method for its content will think it richer in time than it is.
14.13The readers
Two readings by one model coded every case, and models of two other families read a sample of thirty. On this study's eight fields the two readings agreed in 81% to 98% of the cases, and the families agreed with one another in 88% to 95% of the cells they both settled. Where all had to agree, every verdict kept its direction. That agreement is a property of the question as much as of the readers: whether a record names a date, a sum or a bearer is easier to read than whether an act came before its cause was known, the question on which the first study's readers divided. The series' axes that ask who decided, who bore the cost of waiting and whether a decision could be undone were harder, and the readers often left them unsettled or unknown.
14.14What the records can offer to someone who decides
The study gives no advice, but its finding has a plain use. Someone who decides an emergency act in money today will be judged later from records like these. The actors whose acts are easiest to judge in this archive are those whose own texts named a term, an event or a decision on which the act would end, and named whom it protected and whom it did not. The archive kept such sentences when the actor wrote them plainly, and lost them more often when the actor did not. What an act says about its own end is the part of it most likely to survive into a later reader's hands, and the part most often left unwritten.
14.15Contributions
The study has made three contributions. It has counted what one archive preserves of four thousand years of emergency acts in money and found the act preserved and its term and its losers mostly lost. It has set the actors' words beside the summaries and shown what the summaries drop. And it gives the series a second study on one archive and one method.
14.16Limits
The evidence is a modern archive's summaries, the actors' texts where the study could have them, and the reading notes on the scholarly works; it is not the events. One model coded the ledger twice and other families read a sample. The cases were chosen, the price windows hold no acts, and some source texts are uncorrected transcriptions. Its counts describe these summaries and these texts, not emergencies in general. It cannot say whether any act was right, or whether naming an end would have changed what followed.
14.17Further research
Three lines follow. The first is to read the actors' texts for every case of the ledger, not only the twenty-four read closely, and code them on the same fields, so that the gap between the archive and the texts can be counted rather than shown. The second is to read the same question in archives that keep decisions rather than events, such as the series' archives of policy and defence, where a field for the condition of revision exists. The third is to follow the temporary acts of this archive forward in time, from their own texts to the texts that ended them, so that the ratchet's question can be put to acts whose terms are known.
14.18What would change this study's answer
The answer would change if a fuller reading of the archive's records, or of other archives, found the end of an emergency act named in most cases; if the actors' texts for the cases without them turned out to be as silent as the records; or if readers of other families, coding all 84 cases rather than a sample, settled the unsettled cells in the other direction. The study's counts and its registered hypotheses are published with it, so that any of these can be checked.