Chapter 2. Method: The Ledger, the Frame and the Source Texts

This chapter sets out the instruments with which the rest of the study reads the records: what a Skeptron case is; how the decision ledger codes a case; how the studio's recurring sentences were found and why they are kept out of the coding; the eight fields and five hypotheses this study registered before it coded anything; which source texts sit beside which chapters and which originals could not be had; two cases worked field by field; and the limits of every one of these. The Introduction states the study's question; this chapter states how that question is turned into code, and what the code cannot see. This chapter recommends no option and judges no act. It describes records and the readings made over them.

2.1What a Skeptron case is

A Skeptron case is a dated event, a short scenario, and five questions with Skeptron's answers, beside the public source Skeptron checked. Each case is a public-record question produced by the studio Skeptron, and each is read here in an English translation made once for this study. Skeptron writes its cases in Korean; the translation is not re-made question by question, and the English text quoted here is that one translation, with quotation marks checked against the records by machine.

A record carries more than its scenario and its answers. It names its own id and title, the domain it belongs to (crisis-history, credit-history and the others), its nature (public_record_question), its version, its event period, its time scope, and an asOf date on which Skeptron's reading was current. It also carries a fingerprint of the record itself and a retrieval timestamp, so that a later reader can tell which state of a record was read. The fields a case exposes are the ones this study names throughout: the Scenario, the Answer, the Unknown, the Decision window and the Notice, together with the list of sources and the list of questions.

The five questions follow one shape across every case, and their order is fixed. The first is the date of the record: which chronology the adopted source presents, and how the time at which the source was written and released is kept apart from the time of the act. The second is a concrete fact: what the source states about who acted, on what and within what scope. The third is the boundary of what can be judged: what a later reader may not conclude from what the record actually holds. The fourth is a calendar check: what day of the week the adopted date falls on, computed by calendar arithmetic; the record itself says that the consistency of the date formula is recorded separately from the evidence that the event occurred, and this study treats that check as arithmetic only. The fifth is the condition of a later comparison: what observations would have to be gathered for a comparison with the present matter to be kept, and what would withdraw it.

Each answer carries four things beyond its words. It has an answer kind — record, calculation or boundary — so that a reading is never confused with a measurement. It has a method line naming what was done: a cross-check of sources for people, documents, actions, numbers and units; an independent day-of-week formula; a comparison condition set against the source. It has a source hash, which ties the answer to the version of the source text it was checked against. The Bank Holiday of 1933 carries an asOf of 2026-09-13 and names Federal Reserve History's essay "Bank Holiday of 1933" as its source, checked on that day; Korea's Stand-By Arrangement carries an asOf of 2026-09-14 and names International Monetary Fund Press Release 97/55, checked on that day. Both records were fetched for this study on 6 October 2026.

The Scenario of a case states the event's date and what the source separates from it. The Unknown field lists what is not verified and what must therefore not be filled in with actual input. The Decision window says what kind of review is being made — in these two cases, a post-hoc review that keeps the facts contained in the original text separate from the replication check being carried out now. The Notice frames the whole exercise as a historical verification question, distinguishing verified records from calculations and from synthetic control groups, and forbidding the present extraction from being read back as a past prediction.

Two things a case is not. It is not the source text itself; it is Skeptron's dated reading of that source, held at the source's own hash. And it is not a verdict about the event. Where the case's text marks a boundary — that a proportion of resources in reopened banks is not the same metric as the individual recovery of deposits, that an approved sum is not the same thing as cash available at once — that boundary belongs to Skeptron's reading, and this study quotes it as such.

2.2The ledger

The decision ledger is the instrument that makes comparison possible. It codes each case on nine axes. In one sentence each:

• Who decides: what the case's own text lets a reader say about who took or is said to have taken the act — an individual, a team or organisation, an institution or state, a multilateral body or a civilisation, or nothing.
• Grounds and what is missing: what the case's own text gives as the ground of the act, and what it lists as missing — the Unknown field is the main place where a case names its own gaps.
• Deadline, horizon and who bears the wait: whether the record names a deadline or term, how long the act's horizon runs, and who waits while it runs.
• Whether the act can be undone: what the record shows about the constraint on reversing the act and about the action that would reverse it.
• What would change it and whether that could be observed: the condition under which the act would end or be revised, and whether a later reader could watch for that condition.
• The time of the record and of the judgment: the date of the act as the record gives it, and the date of the reading — Skeptron's asOf, not the act's date.
• Divergence: where two readings of the same case differ, and whether that divergence sits against what the case itself lists as missing.
• The human's place: whether the record specifies the human role in the act — approval, exception, review — or leaves it empty, a name only, or absent altogether.
• The four higher goals: what an option offered in a record gives up among the goals above the immediate one, of which evolvability and resilience are two of the names the series' hypotheses use.

Not all of these axes are held in Skeptron's records. That is itself a result, and the study reports it rather than filling it.

Three kinds of cell sit on top of the axes. Some cells are copied: a title, an event period, a source name, an asOf date are taken from the record without interpretation. Some cells are counted by machine: the ledger counts, for example, how many cases list what is missing, how many carry a deadline in the record, how many set a condition for revisiting the judgment, and how far two readings of a cell agree. Some cells are read by coders: decider, bearer, horizon, reversal constraint, reversal action, place, human role and reason recorded are chosen by reading, not copied.

The codings are one model run twice, as two independent codings. Each coding of a cell carries a value, a basis — what in the record or in the setting the value rests on — and a reason for the value. A cell is settled when both codings chose the same value; "stated" means both found it in the record's own words, "implied" that at least one of them inferred it from the setting. Unsettled cells count as unclear in the verdicts. The strict reading counts only what both codings found stated; the lenient reading also counts what the setting implies.

The verdict rule is fixed in advance. A hypothesis is supported at 60% or more of the cases it applies to, goes against at 40% or less, and is unclear in between; it is also unclear where unclear cases outnumber decided ones. Chapter 12 applies that rule to this study's hypotheses and to the series' hypotheses; this chapter states the rule, and nothing here pre-judges a count.

Agreement is not even across the cells. The ledger's figures, as it gives them:

Read plainly, the weakest agreement is in the human's place, where the codings share a value in 46 of 84 cases; in the horizon and in the reversal constraint, where they share one in 49; and in the reversal action, at 54. The strongest is in whether a reason is recorded, at 67, and in who bears the cost, at 62. Where two codings share a value in fewer than 60% of cases — place, at 55%; horizon and reversal constraint, at 58% — this study treats that cell as a reading of a record rather than as a count of it.

Beside these cells the ledger counts five things that a machine can see in the record's words. It finds a deadline in the record for 82 of the 84 cases. It finds that all 84 cases list what is missing, in 232 items in all. It finds that no case names a decider in its own words — 0 of 84 — which is why the decider cell above is so often "unknown" or "unsettled" and why the coders' implied values sit far from any stated one. It finds that no case sets a condition for revisiting the judgment: 0 of 84, and 0 conditions in all, with 0 stated with a number, a threshold or a time, because the code counted no condition to count. The tension between the last two figures is worth holding on to: 82 cases carry something a machine can read as a date or a period, and none of them sets a condition for the act's own end or revision. Those are different things, and the study keeps them apart throughout.

The ledger also holds the series' hypotheses, which are named by what they say and judged in Chapter 12:

On H3, H4, H7 and H8 the ledger reports that the fields they need are absent from the records in most cases: H3 is not-applicable on both readings; H4 is not-applicable on the strict reading and supports on two cases out of the lenient; H7 is not-applicable on the strict reading; H8 is not-applicable on both. On the option cells the ledger records 0 options in all, with both codings agreeing that 0 reduce the immediate loss, and nothing settled about what such options would give up. A verdict on a field the records cannot hold is not a finding about the world, and this study says so wherever the counts force it.

2.3Skeptron's frame

Skeptron's frame carries the studio's own instructions about how a case should be written, and those instructions, not any single event, are what this section reads. The sentence that an event's date and the date of its document are not to be merged, and the sentence that announcement, decision, implementation and result are not one event, are both of that kind. What makes them method rather than content is that a machine found them by recurrence, not that a person chose them: a run of sentences was marked where most of its five-word runs stand in five per cent or more of the cases of its series. That rule found sentences in the Frame of case after case with only names and dates changed. This study treats that recurrence itself as the mark that a sentence carries method rather than content. The frame is therefore Skeptron's way of reading, detected by recurrence, and it is written into case after case.

The sentences fall into recognisable kinds. Some keep the date of an event apart from the date of its document: the record's event chronology and the creation and publication of the source material are not merged into the same field. Some refuse to merge announcement, decision, implementation and result into a single event. Some refuse to fill in month-days, times, coordinates or personal information that the original does not hold. Some state the calendar check and its limits, on the same pattern as the fourth question of every case, which says nothing about the event and is not read as evidence in this study. And every fifth question ends on the same kind of sentence: what observations would have to be gathered for a comparison to be kept, and what would withdraw it — the archive's condition of a later comparison, written into each case with only the names changed. An example from one record's Frame:

This sentence stands in the Frame of that record because it stands in the Frame of a large share of Skeptron's history cases with only the names of the missing items changed. It is Skeptron's method, and it is not a finding about the bank holiday of 1933 or about any other event. It is not counted as support for anything. A finding, in this study, is what a case's own text shows about its own act; a frame sentence is what the studio wrote into most of its cases about how to read. The two are kept apart at every point of the coding. The coders were told to read the case's own sentences only; the frame was kept out of the case's own words for this study's purposes, and the coders read each record without the studio's frame. Where a cell in this study is settled on the record's own text, that is what was read.

One point of method deserves to be said plainly, because it touches this study's own fields. Skeptron's frame contains the sentence that announcement, decision, implementation and result are not one event. That sentence is close in spirit to what this study's field on two dated moments measures, and a careless reading could treat the two as the same thing. They are not. F3 asks whether the case's own text gives two different dated moments among its decision or announcement, its taking effect, and the settlement of its amount or result — a question about dates the record states in its own scenario and answers. The frame sentence is Skeptron's instruction to itself about how to write a case; F3 is a question put to what a case actually holds. This study therefore codes F3 from the case's own dates only, and reports the frame's presence as method, never as support for a hypothesis. Read that way, the counts are ordinary and plainly stated: across the whole ledger, records name two dated moments in eighteen cases, and in fifty-three they name none. The frame cannot raise or lower those figures, because it is not among the sentences they count. Where a case's Scenario itself supplies the dates, as The Bank Holiday of 1933 does with 6, 9 and 13 March, that is the record speaking, not the frame. And where a record's own scenario happens to resemble a frame sentence — as that same Scenario resembles the instruction not to merge announcement, decision, implementation and result — the resemblance is a fact about Skeptron's writing, not a finding about the bank holiday.

2.4The study's own question, fields and hypotheses

The Introduction sets out the study's question. This chapter states the eight fields into which that question is turned, and the five hypotheses registered beside them.

The eight fields are read from each case's own text, and each has a rule for what it counts:

• act: does the case's own text show an act by an authority on money, credit or a market — a law, charter, order, proclamation or edict; a suspension or closure of payment, convertibility, a bank, an exchange or trading; a loan, guarantee, credit line, rescue or takeover arranged by an authority; a relief, cancellation, deferral or restructuring of debts; a change of rate, peg or monetary regime; a sanction or a fine? A market's own movement, with no such act, is "no".
• kind of act: where there is an act, what kind it is, taking the main one.
• presented as temporary: does the case's own text present the act as temporary, provisional or for a stated term — "term-stated" where it gives an end date, a period or a term; "temporary-no-term" where it calls the act temporary, provisional, a suspension or a holiday without a term; "not-said" where it does neither.
• condition of its end: apart from the frame, does the case's own text name the condition, date or event on which the act would end, be lifted or be revised — "observable" where that is a date, a number or threshold, or a named event or decision; "vague" where it is named only in general words; "not-named" where the case names none.
• what became of it: does the case's own text say that the act lasted longer than first said or became permanent; that it ended or was lifted as stated; that it was reversed, abandoned or ended early; or none of these.
• two dated moments: does the case's own text give two different dated moments among the decision or announcement of the act, its taking effect, and the settlement of its amount or result.
• a figure: does the case's own text give a sum, limit, rate or count for the act or the event, and does it set the figure announced apart from what was available, drawn, paid or settled later.
• who bore the cost: does the case's own text name who bore the loss or the cost while the act was in force, or while it was not yet taken — depositors, note holders, debtors, creditors, shareholders, taxpayers, workers, a city or a country.

The five hypotheses were registered on 6 October 2026, at 09:55, before any case was coded for them, in these terms, set out as a list rather than as a block quotation because block quotations in this study are reserved for Skeptron's records:

• F1: In most cases that show an act, the case's own text, apart from the studio's recurring sentences, does not name an observable condition, date or event on which the act would end or be revised.
• F2: In most cases that present an act as temporary and say what became of it, the act lasted longer than first said or became permanent.
• F3: In most cases that show an act, the case's own text names two different dated moments among its decision or announcement, its taking effect, and the settlement of its amount or result.
• F4: In most cases that give a figure, the case's own text sets the figure announced apart from what was available, drawn, paid or settled later.
• F5: In most cases that show an act, the case's own text does not name who bore the loss or the cost while the act was in force or not yet taken.

Each hypothesis carries a test. The test is applied to the cases the hypothesis applies to, and each case yields one of four verdicts: supports, against, unclear, not applicable. A case that shows no act does not apply to F1, F3 or F5; a case that presents no figure does not apply to F4; a case that does not present an act as temporary or does not say what became of it does not apply to F2. Unclear cells count as unclear in the verdicts, and a hypothesis is supported at 60% or more of the cases it applies to, goes against at 40% or less, and is unclear in between, with unclear cases outnumbering decided ones also producing unclear.

Beside each hypothesis, before coding, the editorial session registered predictions, word for word, on 6 October 2026 at 09:55. They are given here exactly as they were registered:

• F1 supported (lenient), unclear strict: the cases name what an act did, seldom what would end it; the condition of a later comparison is the studio's recurring sentence.
• F2 applies to fewer than ten cases; supported among them (the Bank Restriction, Britain's suspension of gold, the closing of the gold window).
• F3 supported, at 70% or more of the cases with an act (lenient): Skeptron's titles in the crisis and credit histories are built on such gaps (an order and a reopening, a charter and an opening, a notice and an effective date); more than E3's 62% in the epidemics study.
• F4 supported among the cases with a figure in the crisis and credit histories (approved and available, two lines, a cap and the existing limit); unclear across the whole ledger.
• F5 unclear: these records name depositors, debtors and holders more often than the epidemic records named the sick, so lower than E5's 70%, but not below 40%.

The counts and the verdicts are left to Chapter 12, and where a prediction failed the study will say so there.

The ledger holds 84 cases in five groups: Skeptron's crisis history, 16 cases; Skeptron's credit history, 29 cases; Skeptron's markets, 4 cases; Skeptron's money, 6 cases; and Skeptron's price windows, 29 cases. The price windows are held in a group of their own, kept apart from the groups that hold cases with an actor, and they are kept apart for a reason that has nothing to do with their date. A price window is not a decision: it records observed prices up to one closing price, for Bitcoin, Ether or Litecoin, with no actor and no act. It has no decider, no act, no term and no bearer, and almost every cell of this study's fields is therefore not-applicable to it. This study reads the windows only as a group, never computes a return, a price or a trend from them, and treats what their own answers say a reader may and may not do as part of the record. Chapter 10 reads that group.

2.5The source texts

Some chapters receive, with every request, the public-domain texts in which the actors wrote. They are evidence of what the actors wrote at the time, and they are used on strict terms: this study says of each text only what it says, in its order, and names the text and its part. Where a text is in Latin, Greek, German or classical Chinese, this study gives its sense in its own English and says so. Source-text passages appear in quotation marks inside this study's own sentences; block quotes are reserved for Skeptron's records alone. Modern authors are described in this study's own words from their reading notes and are never quoted.

The texts and their editions:

The originals that could not be had are a list in itself, and the cases that rest on them are read from Skeptron's record and the reading notes alone. The IMF's press releases, including Press Release 97/55, and the IMF's own histories are not available; the European Central Bank's statements are not available; the Swiss government's press release on Credit Suisse and the Swiss National Bank's statements are not available; the British government's statements of 1992 are not available; and the Mexican and Korean governments' own documents are not available. The Federal Council's emergency ordinance of 16 March 2023 and its amendment of 19 March 2023 are given in German, and this study gives their sense in its own English, saying so. Korea's Stand-By Arrangement names Press Release 97/55 as its adopted source, and that release is not among the source texts, so the case is read from the record and the reading notes only. This chapter shows that consequence directly in 2.6: the figures quoted there are the record's own figures, not figures read from a release.

2.6Two cases judged

Two cases are worked here field by field, to show what the coding does and where the codings part.

The Bank Holiday of 1933. The record's Scenario gives three dates and warns against merging them:

Coded field by field, from the case-by-case table of "The ledger's numbers":

Two further facts from the ledger's group and summary tables belong to this case's reading. Within the crisis-history group the ledger records one case read as "reversed-or-ended-early (stated)" for what became of the act; the case-by-case table records The Bank Holiday of 1933's own cell as unsettled, and this study does not merge the group figure into the case.

Korea's Stand-By Arrangement. The Answer that sets the approved sum apart from what was available at once:

This study gives both figures as the record writes them. The record's own wording for the approved total is "15.5 billion SDR"; the study's summary of the case elsewhere writes the approved sum as SDR 15.5 billion. Both are reported here; neither is converted or reconciled, and the sum stands in the currency and unit the record gives. What the Answer does hold, and what it holds apart, is plain in the record itself: an approved total over three years, a conversion at the time to about 21 billion US dollars, an immediately available amount of 4.1 billion SDR, and subsequent tranches tied to review and performance conditions. The record's third question states the same boundary in its own terms: the approved total may not be treated as cash that flowed in on that day in full, and SDR and dollars may not be calculated using the same number.

Coded field by field:

The two cases stand differently on the study's axes. One is a suspension whose record names three dated moments and a named bearer; the other is a credit whose record sets announced and available apart in its own words, names a term and a condition in general words, and names no second dated moment and no bearer. Both are settled on their records' own text, apart from the one cell of The Bank Holiday of 1933 on which the codings differ.

2.7Limits of the method

Everything above is reported here so that a reader can tell what rests on what.

The codings are one model run twice, as two independent codings. This study never calls them two models or two families, and where they disagree, as in the outcome-of-term cell of The Bank Holiday of 1933, the disagreement is reported as a cell that did not settle. A verdict built on unsettled cells counts them as unclear; that is the rule, not a repair.

The translation is one, made once, and it is a translation. Skeptron writes its cases in Korean; this study reads them in English, and where a Korean term carries a nuance the English cannot hold, that limit is in the reading and not in the source. Quotations from the records are machine-checked against the records as fetched on 6 October 2026.

Skeptron selected its own cases and its own sources. The 84 cases in the ledger are the studio's choices, and the 24 this study reads closely are drawn from them on the study's own reasoning about what its question can bear. The choice of which public source each case adopts is Skeptron's as well: Federal Reserve History for the bank holiday, Press Release 97/55 for Korea. This study does not re-select, and where a source it could not have names something the record does not, the difference is reported rather than resolved.

Source texts go with some chapters and not with others, and the absence is not even. The suspensions of 1797, 1933 and 1971 have texts; the Federal Reserve's release of 14 September 2008, the Chairman's testimony of 1 October 1998, the FDIC's release of 10 March 2023, the Joint Statement of 12 March 2023, the Federal Council's emergency ordinance of 16 March 2023 and its amendment of 19 March 2023, and the Federal Reserve Board's transcripts of Paul A. Volcker's press conference of 6 October 1979 and his statement of 17 October 1979 also have texts. The Swiss government's press release and the Swiss National Bank's statements of 2023, the IMF's press releases and histories, the European Central Bank's own releases, the British government's statements of 1992, and the Mexican and Korean documents do not. A chapter that has a text can say what the actor wrote; a chapter that has none can say only what the record holds. Chapters that receive no source text work under that second limit, among them Chapter 7, whose case rests on an International Monetary Fund press release that is not among them, and this study does not fill the gap from memory or from secondary accounts.

Twenty-nine of the 84 cases are price windows with no actor. They carry no act, no decider, no term and no bearer, and they are kept out of the groups that do. Their presence in the ledger at all is Skeptron's, and this study reads them only as a group, never computing a return, a price or a trend from them. A count taken across all 84 cases without this division would be counting observations as if they were decisions.

This study is written by AI. Its readings of the works named in it stay within what those works' reading notes give, and where this study reads past a note, it says so in the form "this study reads … as …". It never gives an author a quotation, a page number, a date or a claim the note does not have, and it never adds an event, a number, a person, a date or an outcome that neither the records nor the source texts nor the reading notes give. What a rescue did afterwards is said only where a record, a source text or a reading note says it, and named from there.

Finally, the frame is not evidence. The sentences that recur across Skeptron's cases are the studio's method, detected by recurrence, and no count in this study treats them as support for a hypothesis. F1 through F5 are read from the case's own sentences; the coders read each record without the frame; and where a case's own words happen to resemble a frame sentence — as The Bank Holiday of 1933's Scenario resembles the instruction not to merge announcement, decision, implementation and result — the study treats the resemblance as a fact about Skeptron's writing, not as a finding about the event.

The suspension is the first kind of emergency act this study meets; Chapter 3 reads it in the Bank Restriction of 1797 and the bank holiday of 1933.