14.1The answer in brief
The question was a modest one: at the moment a health authority moved, what did it already have in hand, and has the account of that move kept apart when it moved, when proof came and when the toll was taken? Read across Skeptron's histories, the answer has four parts. The acts usually came before their cause, or their own effect, was settled. The records often date the distance between an event and the report or tally of it, but rarely say who paid while an act was pending. The records do say, more often than this study expected, what later finding would alter the reading of an act; but the voice that says it is Skeptron's, writing in 2026, and not the voice of those who acted. Lastly, a full set of what a later judge would require is nearly never present: one record in 36 that show an act.
Each of these parts holds on the lenient reading, which counts what either of the two main readings found stated or implied, and none holds on the strict reading, which counts only what both found stated in so many words. That difference is part of the answer. The records seldom say these things outright; a careful reader can usually infer them. A study that would rest on what the records state, and nothing more, would have to say that the records are silent; a study that reads them as a historian reads a source finds them more forthcoming, and less certain.
14.2Acts and proof
Did the authorities move ahead of the knowledge their move depended on? On the lenient reading of E1 the answer is yes in 25 cases and no in 5, with 20 cases unclear and 39 showing no public act on disease or health at all. On the strict reading the verdict is unclear: 8 cases for, 3 against, 61 unclear and 17 not applicable.
The close readings give the pattern its texture. Florence in 1348 cleaned its streets, refused entry to the sick and issued counsels for health while no account of what carried the plague existed; Boccaccio's description of the measures, read beside Skeptron's record, sets them before any cause. Ragusa's council set a month apart for arrivals from places of plague, and nothing in the record measures what the month achieved. Jenner made a public claim for cowpox from a single boy's case, as his own text of 1798 shows; Pasteur reported a single patient's treatment before any series could test the method. Snow's own account says that the attacks had already diminished so far before the handle of the pump was removed that it was impossible to decide whether the well still held the poison.
The cases where proof came first are few and instructive. The certification of polio's eradication from the Americas came after years without a wild case, and its record says how long the certifiers waited. The declaration on smallpox closed a question that the last case had opened years before. In these acts the evidence set the date of the act, and the records keep the two apart. They are acts of closure; the acts that came before proof are acts of opening, taken when the danger was new.
The finding does not say that acting before proof was wrong. It says that the records, read closely, usually let a reader see that it happened. That is the first condition for judging such an act fairly: to know that the actors could not have known what a later reader knows.
14.3The counts
The second sub-question asked whether a record that gives a count qualifies it. Here the evidence is thin. Across the ledger, only 13 cases with a count were settled by both readings, and on the lenient reading of E2, 8 of them qualify the count and 5 give it as settled, with 11 unclear and 65 not applicable. That is a support by the registered threshold, and it is a support resting on 13 cases.
The third reader, from a different company, shows how little it takes to tip it. Where two of the three readings agree, E2 has 10 cases for and 9 against, which is unclear by the same threshold. Where a majority of all four families agrees, it turns back, to 9 for and 6 against, a support again on fifteen cases. A hypothesis that turns when one reader changes cannot carry much weight, and the study does not let it.
The close readings are more instructive than the count. Graunt's own text says the bills report what the searchers found, and that the searchers' report alone does not suffice to know how many died of plague; the record and the text together show a count that its maker qualified from the start. The army mortality report of 1858 compares soldiers' and civilians' deaths by putting them on a common base, and its record and text show how much the comparison depends on that base. Life Esidimeni's ombud called the count of deaths provisional; the record keeps the word. In the outbreaks of 2014 and 2015, the records say that the counts included probable and suspected cases, or joined the totals of two countries. These are the counts of public health at work: they are given with their limits because the people who made them knew the limits.
14.4Who waited
Next came the people who paid: do the records say who suffered, or carried the cost, while the authorities had yet to move or were moving blind? On the lenient reading of E5, 21 records of acts leave the bearer unnamed and 9 name one, with 20 unclear and 39 not applicable; the strict reading is unclear, at 17 for, 6 against and 49 unclear.
The cases that name a bearer are of two kinds. Some name those the act bound: the Ordinance of Labourers names the workers it compelled to serve at the old wages, and its text gives the reason in the king's voice, that the scarcity of servants after the mortality had driven wages up. Some name those who suffered while an act was delayed or misjudged: the patients moved from Life Esidimeni, in the ombud's report as Skeptron gives it. Most records name no one. A measure that closed a city's gate to the sick is recorded as a measure, not as the experience of those turned away.
This silence is the most important limit the study finds in the records. The whole series turns on the interval between asking and answering, and that interval is lived by people whom these records almost never mention. A later reader who wants to judge whether an act came too early or too late needs to know who paid for the delay or the haste. The records seldom say.
14.5What would change the reading
Last came revision: leaving Skeptron's stock sentences aside, does a case say in its own words what discovery would alter the way its act is understood? The study predicted that it would not, and registered that prediction as E4. Read leniently, the cases contradict that expectation, 18 naming such a discovery and 10 not, while 22 stay unclear and 39 have no act; a share of 36% for the hypothesis falls below the line at which it would hold.
The finding needs care, because of whose text names the condition. Skeptron's fifth question in every case asks what a later comparison would need, and its answer, in the archive's words, often says what would change the case's reading: a further source, a different count, a comparison with another city. The readers were told to leave the archive's recurring sentences aside, and the answers of the fifth question are not all recurring; many are written for the case. So the readers found, correctly, that the case's own text names a condition. But that text is a history written in 2026. It names what a historian would want to know. It is not the record of what the actors said would make them think again.
The distinction matters for the series' question. A decision that names its own condition for revision invites its maker to revisit it; a history that names what would change its reading invites a historian to look further. The records of this study do the second, often, and the first almost never. That is a finding about what an archive of histories can and cannot offer to someone who wants to learn how to set conditions for revisiting a decision.
There is a further caution, and it comes from the readers. Readers of three other families read the same field, and they read it differently from the chapter model. Taken one family to a vote, with the chapter model's two codings counted as one, the field turns the other way: where a majority of the families agrees, E4 is supported, 16 cases for and 8 against. Of the four families, the chapter model agrees least often with the majority of the others, 88% of the time against between 92% and 97% for the rest. So the registered failure of E4 depends on who reads the field. Agreement among readers is not accuracy, and a majority of families can share an error, so the study asked one more reader. A reference coder of yet another family read blind 25 cells on which the families had divided, seeing the manual and the case and no reader's answer. Its values matched the registered reader's two codings in only 9 and 8 of those cells, where the second family matched 17 and the third 16; on the 8 cells of this very field, the second and third families matched it in 6 each, and the chapter model's two codings, where they agreed, in 3 of 6. The sample is small, drawn only where the readers divided, and judged by one coder who marked 14 of the 25 cells as borderline; it measures no reader's accuracy over the ledger. The study keeps E4's registered verdict as registered, against, because the two codings were the reading it registered, and it counts E4 among the predictions that failed, though on this field the registered reading is the one least borne out where the readers divided.
14.6The test
The single test registered in advance looked for records holding everything at once: an act, a dated delay, the people who paid, and what would change the verdict. Among the 36 cases with an act, one passes, and that one is among the 23 read closely: the Philadelphia parade of September 1918. There the archive gives dates for the earliest reports among civilians, for the march itself and for the ban on gatherings, and it identifies those who lived through the days in between. It is also the case whose later study compared American cities by the timing of their measures, which is what such a record makes possible.
Where two of the three readings agree, the test is met by 3 of the 36. Where every family that read a case agrees, none of the 21 cases with an act passes; where a majority of families agrees, 2 of 28 do. The number is still small. The study registered the prediction that fewer than a fifth of the cases with an act would pass, and that prediction held on both readings.
14.7Medicine, disaster and industry
The ledger adds to the cases of medicine the whole of Skeptron's disaster and industrial histories, so that the study could see whether its pattern belongs to acts on disease or to public acts in general. The groups give a mixed answer.
On E1, acts before proof, the lenient readings support the hypothesis in the medicine histories (11 for, 2 against, 4 unclear) and in the industrial histories (5 for, 1 against, 5 unclear), and leave it unclear in the disaster histories, where 8 cases bear it out and none goes against it but 11 remain unclear. The study predicted that acts before proof would be commonest among the medicine cases; every disaster case that was decided goes the same way, and the prediction failed.
On E3, the dated gap, the medicine histories support it (10 for, 6 against, 1 unclear), the disaster histories leave it unclear (7 for, 1 against, 11 unclear) and the industrial histories go against it (2 for, 6 against, 3 unclear). The study predicted that the gap would weaken outside the medicine histories. It does in the industrial histories, but not in the disaster histories, where the share, 88%, is above the medicine histories' 63%, and the prediction counts as failed on that half. On E4 only the medicine histories decide, against the hypothesis (3 for, 11 against). On E5 the medicine and industrial histories support it and the disaster histories leave it unclear.
The groups are unequal in size and in how often their cases show an act at all, and the counts within each are small. They suggest that the pattern of acts before proof is not peculiar to disease, and that dating the gap between an event and its record is more a habit of the medicine histories than of the industrial histories. They do not establish either.
14.8What the actors' own texts added
Where the actors' texts survive in the public domain, the chapters read them beside Skeptron's records, and the comparison did three kinds of work.
It confirmed the order of events. Boccaccio sets the cleansing of Florence, the refusal of entry to the sick and the counsels of health before any word on what carried the plague; Thucydides says he leaves causes to others and sets down the symptoms so that the disease may be recognised if it returns; Snow gives his table of attacks by day, with the removal of the pump handle among the dates. In each, the actor's text shows the act and the evidence apart, as the record does, and often more sharply.
It corrected details. Jenner's text of 1798 describes the boy he inoculated in May 1796 as a healthy boy of about eight and does not give his name, which later sources supply; a reader who knew the boy's name only from later accounts might take it for part of Jenner's report. The preface of the Dongui Bogam carries its own date, which falls between the compilation's completion and its printing, and so adds a third moment to the two the record gives. Graunt's own words say that the bills alone do not suffice to know how many died of plague, a qualification the record gives in its own way.
It showed what the records leave out. The texts name people the records summarise: the searchers who viewed the dead and reported the cause, the labourers whom the ordinance bound, the patients and dairymaid of Jenner's cases. They also show the actors' own uncertainty, in words the records paraphrase. A record of a few hundred words cannot hold all of this; the comparison shows what kind of thing it drops.
The texts that could not be read are a limit of their own. Several cases rest on records alone, either because their originals are not in the public domain or because their owners forbid reproducing them, and three texts that are in the public domain survive only in uncorrected transcriptions that the chapters do not quote. For those cases the study cannot say whether the records' silences are the actors' or the archive's.
14.9The series' hypotheses on this archive
The series registers its own hypotheses for every archive, and this ledger judges them as far as Skeptron's histories allow. Most cannot be judged here, and the reason is the form of the record.
One series hypothesis, that decisions rarely set out beforehand what would undo them, is borne out by all 89 cases under both readings. It holds because Skeptron's history cases have no field for such a condition, not because the acts lacked one. What the form of a record leaves out is not a finding about the past, and the study does not count it as one.
Another, that a restriction survives after its reason is forgotten, is flatly contradicted: Skeptron records the reason in 80 cases on the strict reading, and 82 cases oppose the claim on the lenient one. Skeptron's histories almost always say why an act was taken. That is a fact about the archive, which writes its histories to say why, and it shows that this archive cannot test whether a reason was forgotten by those who kept the restriction.
The hypothesis on who bears the cost of waiting is unclear on both readings, because the records seldom name a decider and a bearer together. The hypotheses that need options, a horizon, a measure of how hard an act is to undo, or a place for a human beside an automatic rule do not apply to these records, or apply to too few cases to count. The hypothesis on horizons is left to the volume that reads the series' ledgers together. The readers' low agreement on the series' axes, 60% on who decided, 36% on the horizon and between 40% and 43% on reversal, is part of the same lesson: those axes were made for records of decisions, and histories answer them poorly.
14.10The predictions that failed
Seven predictions were registered before the coding. Two came true: that the first hypothesis would stand when read leniently and stay undecided when read strictly, and that under a fifth of the act-bearing cases would pass the test. Five failed on the registered reading, and the study reports them as such; two of them, E2 and E3, held in the part about the read cases and failed in the other, and for E4 the reading itself is in doubt, as the section on what would change the reading shows.
E2 was predicted supported among the read cases, and it is, 6 of 8; it was predicted unclear across the whole ledger, and the lenient reading supports it, if on 13 cases. E3 was predicted supported among the read cases, and it is, 13 of 21; it was predicted weaker across the disaster and industry cases, and it is weaker in the industrial histories but not in the disaster histories, where the share, 88%, is above the medicine histories' 63%. E4 was predicted to hold; the lenient reading goes against it, for the reason given above: the archive's later text names what would change a reading. Where a majority of the four families agrees, the prediction holds after all, and the reference reading of the disputed cells sides more often with the other families than with the chapter model. E5 was predicted unclear; the lenient reading supports it, which is to say that the records name the bearer of the wait even less often than expected. And acts before proof were predicted to be commonest among the medicine cases; the decided disaster cases bear out the hypothesis without exception.
The failures teach more than the successes. If the failure of E4 stands, it shows that the study's design underestimated how much of an archive's text is its own later reading; if the other families read the field rightly, it shows instead that the registered reader took the archive's later text for the case's own more often than the others did. Either way the lesson is about reading, and the study cannot yet say which. The failure of the group prediction shows that acting before proof is not a feature of medicine alone. A study that had not registered its predictions could have told its results as if it had expected them; this one cannot.
14.11What the records can offer to someone who decides
This study does not advise. It can still say what its records hold for a reader who must act on disease, because that reader will meet such records and will be tempted to learn from them.
The records teach, first, that acting before proof is the ordinary condition of public health, not its failure. Most of the acts read here were taken without the knowledge a later reader has, and the records usually make that visible. A reader who judges a past act by its outcome alone forgets what its actors could not know.
They teach, second, that a count should travel with its limits. The counts that carry their qualifications (provisional, probable and suspected, joined across countries, put on a common base) are the ones a later reader can use. A count given bare invites more confidence than its makers had.
They teach, third, by their silence. A record that does not name who waited, or what would make its makers think again, leaves the most important parts of a decision to whoever tells the story next. A reader who writes such a record now can do better than the archive: date the act and the evidence, say who bears the wait, and say what would change the decision. These are not lessons the study draws from the acts; they are what the records, by what they lack, show a record would need.
14.12Contributions
The study makes three contributions, each of a kind a later study can test.
It provides a ledger of 89 history cases on the same fields, with the archive's recurring sentences kept out of the coding and the agreement of its readers reported for every field, including a reading by a model of another family. The ledger can be read, recoded and compared without this study's argument.
It turns a familiar historian's distinction, between the day of a measure, the day of its proof and the day of its tally, into a yardstick that can be laid against cases thousands of years apart and that yields counts anyone may verify. Historians of single epidemics have drawn the distinction in their own cases; this study draws it across an archive.
Its finding, given with every denominator, is that Skeptron's accounts of acts on disease mostly reveal that the act outran its proof, quite often date the delays around it, seldom identify those who paid, and hardly ever hold everything a later judge would need; and that the conditions they name for rereading an act are, for the most part, the archive's own. These are claims about records, and the records are open to anyone who would test them.
14.13Limits
The limits set out in the introduction hold to the end. The records are Skeptron's, short and recent, written to a template whose fields decide what can be counted. The selection is the study's, and the ledger does not represent public health or public acts in general. The readings are models', and where the readers divided the chapters say so; where they agreed, they may have agreed in error. The source texts are uneven: some cases are read beside their actors' own words, some only beside uncorrected transcriptions the chapters do not quote, some beside nothing at all.
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. What the study says of that case is limited to the statement, the date it bears and the boundary the record draws between that date and the start of infection.
The strict readings are all unclear. Every pattern the study reports rests on the lenient reading, which admits what a careful reader would infer. A reader who will accept only what a record states in so many words should take the study's findings as hypotheses the records suggest, not as results they prove.
14.14Further research
Four lines of work would test or extend what this study finds.
The first is to read the same cases with other readers. A reading by people, or by models of further families, would show whether the fields on which the readers divided (what would change the reading, who bore the wait) are hard because the records are ambiguous or because the readers share a habit. The ledger is built to be recoded, and its counts can be set beside new ones field by field.
The second is to read records of decisions on disease, not histories of them: the minutes of a health board, the text of an order with its stated term, a committee's statement with its conditions for review. Such records could test the series' hypotheses that this archive cannot, above all whether decisions on disease name in advance what would reverse them.
The third is to add the actors' texts that this study could not hold. Several of its cases rest on records alone, because their originals are not in the public domain or their owners forbid reproduction. Read with permission, those texts would show whether the records' silences are the actors' silences or the archive's.
The fourth is to date more finely. The study codes whether a gap is named, not how long it was. An archive that gave the dates of act, evidence and count for each case would let a later study measure the gaps, compare them across centuries and kinds of act, and ask whether they have shortened.
The fifth is to read the archive again as it changes. Skeptron's cases carry their own dates and can be revised; a case that gains a source, loses a claim or changes a count is a record of how an archive's later reading moves. A study that read the same cases a year apart, on the same fields and with the same readers, would show which parts of a history settle and which stay open, and whether the conditions an archive names for rereading an act are ever met by the archive itself. That would turn the series' question back on the archives that serve it: what in a history should be decided now, and what change should make the historian revise it.
14.15What would change this study's answer
The study names, as it asks its records to name, what would change its own reading.
If a further reader, or a reading by people, found the bearer of the wait named in most records of acts, the third part of the answer would fall: the silence the study reports would be a habit of its readers, not of the records. If the actors' own texts, read where they now cannot be, named conditions for revisiting their acts, the fourth part would change: the conditions would belong to the hour of the act, not only to the archive's later reading. If a larger set of records with counts showed that most give their counts bare, the support for E2, already resting on 13 cases, would turn. And if a strict reading of a richer archive found acts before proof stated in so many words, the first part of the answer would move from what a careful reader infers to what the records say.
Whatever the outcome of such tests, the approach itself would stand: put to every account the question of which of the three moments it preserves, and add up the replies. The study offers that method to any archive of acts, and its counts to any reader who would test them.