Introduction

0.1What this study asks

Every public act on disease is taken at an hour, and every hour has its own state of knowledge. A council that bars travellers from a city where plague is reported acts on rumour and memory; a physician who gives cowpox to a boy acts on the stories of milkers; a committee that calls an outbreak an emergency acts on two days of reports. Later, the cause may be found, the measure may be counted, the deaths may be tallied and revised. By then the act has a reputation, and its reputation is usually written in the knowledge of the later hour.

This study asks what the earlier hour held. Its question is narrow on purpose: when a public body acted on disease, what was established at the moment it acted, and does the record of that act let a later reader keep apart three moments that the reputation tends to fuse, the moment of the act, the moment of the evidence and the moment of the count? The question is put to records, not to events. It does not ask what the plague of Florence was, or whether the month of separation at Ragusa worked. Its subject is the archive's account: what that account permits a reader to affirm about each act as of its own date, and where the account goes quiet.

The narrowness is the study's protection. A question about whether acts on disease were wise would need evidence that no archive of short cases can hold: the counterfactual city, the epidemic that did not happen, the treatment withheld from a comparable patient. A question about what a record allows can be answered from the record itself, case by case, on the same terms, and the answers can be counted. The counting is what lets the study speak of patterns rather than anecdotes, and it is also what lets a reader check the study: each count has a denominator, each cell a case behind it.

0.2The series and its question

This study is one of a series. Every volume of the series opens a different studio's archive and asks of it one shared thing: which choices cannot wait, and which later facts should reopen them? Each reads its archive on the same axes (who decides, on what grounds and with what missing, by when and over what horizon, who bears the wait, whether the act can be undone, and what would make it be revisited), and each registers its hypotheses before its cases are coded. The series' theme is time: the time between a question and its answer, and the time between an answer and the evidence that would show whether it was right.

Public health puts that theme in its starkest form. Its acts are often irreversible for the people they touch, and its evidence often arrives too late to have guided them. Turned on this archive, the series' question becomes one about health authorities facing an outbreak, before it comes and while it lasts: what has to be settled today, and what new fact ought to reopen it? This study's own question sits beneath it. Before asking what should make a decision be revised, it asks whether the record of a decision lets anyone see what the decision stood on.

The series' axes were designed for records of decisions: records that name a decider, list options, state a deadline and a condition for revisiting. Skeptron's history cases are not records of that kind. They are short histories that a studio wrote in 2026 about documents of many ages. The study codes them on the series' axes so that they can be set beside the other archives, and it reports plainly where those axes fit them badly. It adds seven fields of its own, fitted to acts on disease, and its own hypotheses about them.

0.3Sub-questions

The question divides into four.

The first concerns the act and the proof. When the body acted, had the cause it addressed been established, and had the act's own effect? An order against plague in a medieval city came before any account of how plague spreads; a vaccine given in one village came before any trial; a declaration that a disease had been eradicated came after years of searching. The sub-question is not whether the act preceded proof in some absolute sense, but whether the case's own text lets a reader place the act before or after the evidence it needed.

The second concerns the count. When a record gives a number of the dead, the sick, the cases or the patients, does it give the number as settled, or does it qualify it: as provisional, as revised, as bounded by a definition, a scope or a denominator? A count that carries its qualification tells a later reader how far to trust it; a count given bare invites a confidence it may not deserve.

The third concerns those who waited. While an act was pending, or while it was taken on uncertain ground, someone bore the harm or the cost: the sick who were refused entry, the labourers whose wages were fixed, the patients moved from one place of care to another. Does the record name them?

The fourth concerns revision. Does the case, in its own words, point to some later finding that would alter how the act is understood, and is that finding of a kind anyone could actually make? A record that says what would change its reading invites a later reader to test it; a record that says nothing leaves the act's reputation to whoever tells the story next.

0.4The argument in brief

The study's answer can be put in four sentences, each of which a later chapter tests and Chapter 12 counts.

First, the acts came before the proof. Where the readings agreed on the point, the body usually moved before anyone could say what caused the harm or whether the measure would help: on the lenient reading, 25 such cases against 5, with 20 left unclear and 39 cases showing no public act at all. Florence cleaned its streets and barred the sick without any account of what spread the plague; Ragusa set a month of separation with no measure of its effect; Jenner made a public claim from one boy. Where proof came first, as with the certification of polio's eradication, the record says how long the world waited for it.

Second, the gap is often dated, the cost seldom named. A little more than half of the records of acts date the distance between an event and its notice, count or settled cause: 21 cases against 13 on the lenient reading. Far fewer name who bore the harm or cost while the act waited: the records of 21 cases with an act leave the bearer unnamed, against 9 that name one. The medieval ordinance on wages names the labourers it bound; most records name no one.

Third, what would change the reading is named more often than expected, and by the wrong hand. The study predicted that the case's own text would seldom name a later finding that would change an act's reading. The lenient reading goes the other way, 18 cases naming such a finding against 10 that do not. But the text that names it is Skeptron's, written in 2026, setting out what a later comparison would need; it is the archive's reading of the act, not the actors' record of the hour. This is also the field on which the readers divided most: readers of three other families, taken by majority, read it the other way, and a reference reading of the disputed cells sides with them more often than with the registered reader. The study keeps the registered verdict and does not treat it as settled.

Fourth, almost no record gives a later reader everything. The test asked how many cases with an act record all four things a reader would need to judge it in its own time: the act, a dated gap, the bearer of the wait and what would change the reading. Only a single case passes among the 36 with an act: the Philadelphia parade of September 1918, whose record dates the first civilian reports, the parade and the closing of gatherings, and names who bore the days between.

Each of these sentences holds on the lenient reading and is unclear on the strict one, which counts only what both readings found stated in so many words. Counts are small, and some fields divided the readers. The chapters show the cases behind each count; Chapter 12 shows the counts; the conclusion says how far each sentence can bear weight.

0.5Three times

The study's vocabulary is small. It distinguishes three times in the life of an act on disease.

The time of the act is when the public body did what it did: the date of the order, the meeting, the trial, the notice, the declaration. Many records give it exactly. Some give only a year, and some give the date a document was written rather than the date of the act it records.

The time of the evidence is when the cause the act addressed, or the act's own effect, came to be established. It is often far later than the act, sometimes centuries later, and sometimes it never comes: a measure against plague in a fourteenth-century city has no evidence of its effect that a modern reader would accept, and the record does not pretend otherwise. Sometimes the evidence comes first and the act follows, as when a certification waits for a stated period after the last case.

The third time belongs to the tally: the day the dead, the sick or the treated were added up, and any later day on which that sum changed. A count may be weekly, as in the printed bills of a seventeenth-century city; it may be a single total published years after the event; it may be called provisional when first given and never be called final.

The three times are distinct in principle and often fused in memory. A reader who learns that a pump handle was removed and that an outbreak ended may take the removal for the cause of the end, unless the record also gives the count of new cases before and after. A reader who learns that a disease was declared eradicated may take the declaration for the end of the disease, unless the record also dates the last case. The study's use of the three times is simple: it asks of each record whether it keeps them apart, and counts how often it does.

0.6Terms

A few words carry weight throughout and are used in one sense only.

A public act is something a public body did on disease or health that the case's own text records: an order, a resolution, a restriction on movement, work or assembly, a notice, a declaration, a certification, a trial or treatment given under public authority, a compilation or gift of remedies made by a state or a temple, a published count. A private physician's opinion is not a public act unless a public body adopted it. A history of an act, written later, is not itself an act.

A record is Skeptron's case: its scenario, its five questions and their answers, its list of sources and its notice. The case's own text is that record less the frame. A source text is a document written by an actor in the case, or under an actor's authority, which the study holds in the public domain and reads beside the record.

The frame is the set of sentences that recur across Skeptron's history cases with only names and dates changed. The study found them by machine, by looking for runs of five words that appear in a large share of a family of cases, and treats a sentence as frame when most of its runs are shared. A frame sentence may be true and useful; it is still not evidence about the case it appears in.

The ledger is the table of the 89 cases on all their fields. A cell is one case on one field. A cell is stated when the case's own text says the thing, implied when a careful reader would infer it from what the text says, and unsettled when the readings of it disagree. The strict reading of a hypothesis counts only cells stated in both main readings; the lenient reading counts cells stated or implied.

A gap in time is named when the case's own text dates two moments, or says how long passed between them: an event and its notice, a death and its count, an act and the settling of its cause. A count is qualified when the text says it was provisional or later revised, or gives its definition, its scope or its denominator. The bearer of the wait is the person or group the text says bore harm or cost while an act was pending or was taken on uncertain ground.

0.7Records rather than events

A study of acts on disease could take the events as its object: the plague, the outbreak, the measure, the outcome. Most histories of epidemics do, and they do it with sources this study does not have, from parish registers to hospital archives. This study takes records as its object for three reasons.

The first is that a record is what a later decision-maker actually meets. A health officer facing a new outbreak does not meet the plague of 1348; the officer meets an account of it, often a short one, written long after, which says what was done and what followed. Whether that account keeps the act apart from its proof and its count shapes what the officer will take from it. An archive of short cases is closer to what such a reader uses than any full history.

The second is that records can be compared on equal terms. The events in this study differ in every way that matters to a historian: in scale, in the knowledge of their time, in the institutions that acted. Their records, as Skeptron writes them, share a form: a scenario, five questions, a list of sources. That shared form lets the same fields be asked of a papyrus and a press statement, and the answers be counted without pretending that the events themselves are alike.

The third is that a claim about a record can be checked by anyone who reads it. A claim that an act was premature, or late, or effective, rests on evidence that may be disputed for centuries. A claim that a record dates the gap between an event and its count, or does not, rests on the record, and a reader can open it and see. The study's counts are of this second kind, and that is why it can give them with their denominators and let them be tested.

The cost of this choice is plain. The study cannot say what happened beyond what its records say, and where a record is wrong the study may repeat the error. It meets that cost in two ways: by reading the actors' own texts beside the records wherever they survive in the public domain, and by saying, case by case, where a record and a text differ.

0.8The archive and the choice of cases

Skeptron publishes history cases across many fields. Each case is a scenario followed by five questions, and each question has an answer: when the sources were made, a concrete fact from them, where judgment must stop, a calendar check of a date, and what a later comparison would need. The scenarios are Skeptron's own summaries, written in 2026 from the sources each case names. They are not the sources themselves.

The cases for this study were chosen by content, not by Skeptron's own labels. A first choice, made from the label Skeptron gives to histories of health and medicine, gave seventeen cases; the case of the World Health Organization's statement of 2020, the plague at Athens and the Ordinance of Labourers were added because the subject demanded them, which gave twenty. A second search then read the scenarios and titles of the whole history archive for the words of epidemics and public health (plague, contagion, quarantine, vaccine, inoculation, bills of mortality, mortality, health, medicine) and checked by eye each case whose words only partly matched. It added three cases that Skeptron files under wellbeing: Nightingale's hospital at Scutari, which pairs with the army mortality report of 1858; the World Health Assembly's declaration on smallpox in 1980, which pairs with the certification of polio's eradication from the Americas; and the entry into force of the World Health Organization's Constitution, which stands at the head of the two later declarations of that organisation that the study reads.

The search also turned up cases the study leaves out of its close reading, each for a stated reason. A code on the marketing of infant formula concerns harm that is not epidemic and a record of compliance by states, which belongs to a study of codes and their keeping. A Nobel Prize in medicine is a dated honour, not a public decision. A chemical disaster in India and a buried waste site in the United States are environmental and industrial disasters; the first stays in the ledger through Skeptron's disaster histories. A public medical office of medieval Korea is an institution of care, not an act on an epidemic; it stays in the ledger through Skeptron's industrial histories. A naval quarantine in 1962 and the first Geneva Convention share words with the subject and nothing more.

The result is 23 cases read closely, set in the order of their dates, from the Edwin Smith Papyrus to the Pandemic Characterisation. The ledger is wider. To see whether what holds for acts on disease holds for other kinds of public act, it adds the rest of Skeptron's medicine history and all of its disaster and industrial histories, 89 cases in all, each counted once. The groups are unequal: 17 cases of medicine, 47 of disaster and 22 of industry, with the cases read closely among them. The wider ledger is a check on the narrower reading, not a sample of public health in general.

0.9The cases at a glance

The 23 cases read closely run across about three and a half thousand years. The table gives each by the short title the study uses, the date the record gives for its central act or document, and the kind of time it puts in question.

0.10The method in brief

Chapter 2 sets out the method in full. In brief, it has four parts.

First, a ledger. All 89 cases are coded twice over: once on the series' axes, and once on seven fields made for this subject. The seven ask, of each case's own wording, whether a public body acted on disease or health; of what sort the act was; whether it preceded the settling of its cause or of its own effect; whether a delay between an event and its report, tally, publication or settled cause is dated; whether a number is given, and with what caveat; whether those who paid while the act was pending are named; and whether, beyond the studio's stock sentences, some later finding that would change the act's reading is named. Each field is read twice by one model and once more by a model of another family. A cell where the two main readings agree is counted; a cell where they differ is counted as unsettled.

Second, a frame kept apart. Many of Skeptron's sentences travel from case to case unchanged except for the names and the dates: warnings not to merge the date of an event with the date of its document, not to treat announcement, decision, implementation and result as a single event, and not to read a calendar check as proof that something occurred. These sentences are Skeptron's method. They were found by machine, listed with each record, and the readers were told not to fill a field from them. A pattern made of the frame would be a pattern of the archive's method, not of the acts it records.

Third, hypotheses registered before the coding. The study's five hypotheses, E1 to E5, were written and committed before any case was read: that most acts came before their cause or effect was established; that most counts are qualified; that most records of acts name a gap in time; that most do not name what would change the act's reading; and that most do not name who bore the wait. Each is judged twice: strictly, on cells both readings found stated in the case's text, and leniently, on cells either found stated or implied, with a hypothesis supported when three in five of the decided cases bear it out and opposed when two in five or fewer do. Alongside them the study registered seven predictions of how the verdicts would fall, and a test: how many cases with an act name all four things a later reader would need to judge it in its own time.

Fourth, the actors' own texts. For the cases whose originals are in the public domain, the chapters read the texts beside the records: the plague chapters of Thucydides in English and Greek, Boccaccio's account of Florence in English and Italian, the Ordinance of Labourers, Graunt's observations on the bills of mortality, Lind's trial, Jenner's inquiry, Snow's account of the Broad Street pump, the army mortality report of 1858, Pasteur's report on rabies, the preface and editorial principles of the Dongui Bogam, the weekly reports of the United States Public Health Service in the autumn of 1918, the report certifying polio's eradication from the Americas, and the preamble and three articles of the World Health Organization's Constitution. Where a text and a record differ, the chapters say so and give both.

0.11The readers and how far they agree

Every cell of the ledger rests on a reading, and the study reports how far its readings agree. Two readings of the seven fields by one model agree on between 75% and 88% of the 89 cases, field by field, with kappa (a measure that discounts the agreement chance alone would give) between 0.59 and 0.75. A third reading, by a model from a different company, sides with the first less often: on between 63% and 89% of cases, with kappa between 0.33 and 0.70.70. The field on which the readers divided most is the one that asks whether the case's own text names what would change an act's reading; the next is the one that asks whether it names who bore the wait.

The series' axes fare worse, because Skeptron's histories are not decision records. On who decided, the two main readings agree in 60% of cases; on the horizon of the decision, in 36%; on whether it could be undone, in 40% to 43%; on whether a human or a rule decided, in 42%. Where a comparison in the chapters rests on one of these cells, the chapters say that the readers divided on it.

Agreement is not truth. Two readings can agree because the record is plain, or because the readers share a habit. The further reading by another family is there to expose habits of the first: where it departs from the two main readings, the study looks again. Its verdicts, where two of the three readings agree, are reported apart from the registered verdicts, because they were not registered before the coding. They keep the direction of four hypotheses and turn one, E2, from supported to unclear, a reminder that a hypothesis resting on few cases moves when a single reader changes.

0.12What this volume brings to the series

The series reads many archives, and a later volume sets their ledgers side by side. That volume reads the ledgers, not the books, and it does not rest a claim that crosses archives on a field whose values come mostly from an archive's recurring sentences. This study's place in that comparison is particular, and it is worth saying how.

Most archives in the series are records of decisions: they name a decider, list options, set a deadline and sometimes state what would make the decision be revisited. Skeptron's history cases are records of a different kind. They are an archive's later account of acts, many of them centuries old, and they ask their own questions of each act: when its sources were made, what one fact is certain, where judgment must stop. The series' axes therefore fit them unevenly. Some fit well: whether a reason for a restriction is recorded can be read from almost every case. Others fit badly: the horizon of a decision, or whether it can be undone, is seldom something a history of a plague order says.

That unevenness is itself a finding the series can use. It shows which of the series' questions can be asked of an archive of histories and which only of an archive of decisions, and it warns the later volume not to count an absence that comes from the form of a record as an absence in the past. This study marks each such field where it reports it, so that the comparison across archives can leave it out or weigh it accordingly.

0.13Three contributions

The study makes three contributions, each modest and each checkable.

The first is a ledger. It codes 89 history cases of public acts, most of them on disease, disaster or industry, on the same fields, keeps the archive's recurring sentences out of the coding, and reports for every field how far its readers agreed. The ledger can be read without this study's argument, and a later study can recode it with other readers and set its counts beside these.

The second is a way of reading. Separating act, evidence and tally is not an invention of this study; historians of plague, of cholera and of statistics have each done it for their own cases. What is new here is its use as a common measure across cases far apart in age and kind, from a papyrus to a press conference, so that the same question can be asked of each and the answers counted.

The third is a finding about records. The study shows, with its counts, that the archive's records of acts on disease usually let a reader see that the act came before the proof, often date the gap between an event and its report, rarely say who bore the wait, and almost never give a later reader everything needed to judge an act in its own time. It shows also that what the records say would change an act's reading is, for the most part, the archive's own later reading, not the actors' record of the hour. These are claims about what the records hold, and they can be tested against the records.

0.14What the evidence cannot show

The study's evidence has limits that bound every claim it makes.

The records are Skeptron's. They are short, they are recent, and they are written to a template. A field that the template does not ask for, such as a condition for revisiting a decision, is a field the records will not fill; where the series' hypotheses depend on such a field, the study says that the form of the record, not the history of the acts, decides the count. The records' five questions were written by one studio with its own method, and the frame they share is that method, not the past.

The selection is the study's. Twenty-three cases cannot represent the history of public health, and the 89 of the ledger cannot represent public acts in general. The cases read closely were chosen because they bear on the question; the wider ledger was chosen because Skeptron publishes it. The study draws no conclusion about acts that are not in the archive.

The readings are models'. Every cell of the ledger was filled by models reading the records. Two readings by one model agree more often than a reading by a model of another family agrees with either, and the fields on which the readers divided most are named where they are used. A cell that both main readings filled the same way can still be wrong in the same way.

The source texts are uneven. Some cases have their actors' own texts in the public domain, read in full; some have only a transcription from uncorrected optical character recognition, which the chapters do not quote; some have no text the study can reproduce, either because none survives in the public domain or because its owner's terms forbid it. There the study keeps to the record and to its reading notes, and goes no further.

Recent events are thin. Skeptron's records hold one case on COVID-19, the World Health Organization's statement of 11 March 2020; lockdowns, vaccines and national responses are not in them, and this study does not use them as evidence. The same restraint holds for the outbreaks of 2014 and 2015, which the records give through a committee's meeting and a country's notification and counts.

Above all, the study does not judge. It finds that acts came before proof, and it does not say that they should not have. It finds that records rarely name who waited, and it does not say who should have been named. Those judgments need evidence of another kind, and the study leaves them to readers who have it.

0.15The plan of the study

The study has four parts and thirteen chapters, with this introduction before them and a conclusion after.

Part I prepares the reading. Chapter 1 maps four bodies of writing that bear on acting before the evidence: histories of epidemics and of public response; writing on contagion, quarantine and the state; writing on numbers and their authority; and writing on decisions taken before the evidence is in, including the dispute over precaution. It adds what has been said about how to tell whether an act on disease worked, and states the gap this study fills. Chapter 2 sets out the method: the ledger and its fields, the frame, the hypotheses and predictions, the source texts and the limits of each.

Part II reads by time. Chapter 3 takes acts taken before their cause was known, from Athens and Florence to Ragusa's month of separation. Chapter 4 takes the long wait for proof: Lind's comparison of treatments, Jenner's single case, Snow's pump and Pasteur's patient. Chapter 5 takes the count: Graunt's bills, Nightingale's hospital and the army mortality report. Chapter 6 asks who waited, through the Ordinance of Labourers and the deaths after the transfer of patients at Life Esidimeni.

Part III compares by kind of act. Chapter 7 sets gates, orders and assemblies side by side, from Florence to the Philadelphia parade of 1918. Chapter 8 reads notices, declarations and certifications, from the World Health Organization's Constitution and the first notice at Minamata to the declarations on smallpox, polio and COVID-19. Chapter 9 reads the records that keep remedies: a surgical papyrus, a temple's list of medicines and a royal compendium. Chapter 10 compares counts, in the outbreaks of Ebola in 2014 and MERS in Korea in 2015.

Part IV judges and compares. Chapter 11 asks what conditions for revisiting the records name and whether they could be observed. Chapter 12 judges the hypotheses on the ledger's counts, sets each beside its prediction, reports the test and the groups, and gives the further reading by a model of another family. Chapter 13 sets the findings beside the earlier accounts of Chapter 1, saying where they agree, where they differ and where the records cannot speak.

0.16How to read what follows

The chapters quote Skeptron's records in block quotes, word for word, and name the case and the field each quote comes from. They quote the actors' texts only inside their own sentences, and only where the text is in the public domain and its transcription can be trusted, or where the study's rules list the passage. They give counts only from the ledger, with their denominators. They describe the works of other authors only as the study's reading notes on those works allow, and they say when a reading goes further.

A reader in a hurry can read Chapter 2 for the method, Chapter 12 for the verdicts and the conclusion for the answers. A reader who doubts a verdict can go to the case it rests on: each chapter names the cases it reads, and each case is a short record that can be read in a few minutes. The study is built to be checked, and its value lies as much in what it lets a reader check as in what it concludes.