Abstract
Skeptron's history cases reach back to a copy of an Egyptian treatise on injuries made about 1600 BC and forward to a statement issued in Geneva in March 2020. Among them are public acts on disease: gates shut, the dead tallied, remedies tried, notices sent, declarations made. Of each such act this study asks what was already settled at the moment the body acted, and whether the record lets a later reader hold three moments apart: when the body acted, when the evidence arrived, and when the count was made. It reads 23 cases closely, beside the texts their actors wrote where those survive, and codes 89 cases from Skeptron's histories of medicine, disaster and industry on the same fields. Its answer is about records and the time they keep. In most of these records, the body that acted did so while the cause, or the act's own effect, was still unsettled. Dates of report and tally often stand in them beside the date of the event; the people who bore the cost of an act still pending seldom appear; and only one of the 36 cases with a public act holds everything a later reader would need to judge that act as of its own day.
What this study is
The study was made by RISCON Labs AI. It belongs to a series of studies that put the same question to different archives of decisions: what had to be decided, and what change should make that decision be revised. The archive read here is Skeptron's, a studio in the RISCON family that publishes history cases, each a short scenario with five questions about its sources, its facts, the limits of judgment, a date and the conditions of a later comparison. Skeptron's cases are its own readings, written in 2026, of documents written at the time or long after.
The study is a reading of those cases, not a new history of any epidemic. Every sentence in a block quote is Skeptron's record, word for word. The rest is writing by AI, made to rules set before the writing began: the works it may cite and what it may say of each, the passages it may quote, the counts it may give, and the cases it reads. Its numbers come from one place, the ledger's count of the coded cases, which the study calls "The ledger's numbers"; its claims about other authors come from reading notes checked against bibliographic databases before any chapter was written.
How the study was made
The study was made in steps that kept each kind of work apart. A design came first, written before any chapter: the question and the fields to be coded, the hypotheses and the predictions about them, the plan of chapters and sections, the list of works with a note on what each argues, and the passages that may be quoted. The hypotheses and predictions were committed to the record before the coding began, so that the study could not fit its expectations to its results.
The cases were then coded. A model read each of the 89 cases twice on the study's fields, and a model of another family read them once more; where the readings differ, the cell is counted as unsettled rather than decided. Each chapter was written by a model from the records, the reading notes, the ledger's counts and the surviving texts of the actors, and then checked, by machine and by other readings, for quotations that are not in the records, numbers that are not in the ledger, claims about works that their notes do not bear out, and sentences that repeat other chapters. What the checks found was corrected before the study was assembled. This preface, the introduction and the conclusion were written last, from the chapters as they stand.
None of this makes the study certain. A model can misread a record, and two models can misread it in the same way. The checks catch what they are built to catch: a quotation that is not word for word, a count that has no source, a claim that a note does not hold. They do not catch a reading that is plausible and wrong. For that reason the study shows its counts with their denominators, reports where its readers disagreed, and says plainly where its own predictions failed. A reader who doubts a reading can follow it to the record, which each chapter names.
What the records are
Each of Skeptron's history cases is short. A scenario sets out what happened and when, with the sources the case relies on; five questions follow, each with an answer. The first asks when the sources were made, the second asks for a concrete fact, the third marks where judgment must stop, the fourth checks a date by calendar arithmetic, and the fifth names what a later comparison would need. Many sentences recur from case to case with only names and dates changed; the study calls them the frame and keeps them apart from what a case itself shows.
Why read these records at all
Acts on disease are a hard test of the series' question. A city that waits for proof of what spreads a plague waits while the dead are counted; a city that acts first may be wrong about the cause and still be right to act, or right about the cause and still act too late. Readers who meet such acts later usually meet them through their outcome: the epidemic ended, or did not; the measure is remembered as wise, cruel or futile. The records themselves often say something less settled. They give a date for the order, a date for a report, a count with a qualification, a notice that names no cause. Reading them for the time they keep is a way to see what the actors could have known, and what a later reader can still tell.
The study does not ask whether any act was right. It asks what the record allows a reader to say about the act in its own time, and where the record falls silent. That restraint is a choice, not a modesty: a judgment of rightness needs evidence these records do not hold, and the study prefers to show the edges of the evidence than to cross them.
What the reader will find
Part I sets out four bodies of writing on acting before the evidence and the method that turns Skeptron's cases into comparable rows. Part II takes the three times one by one (when the body acted, when proof arrived, when the dead or the sick were counted) and then asks who waited while those times ran apart. Part III compares the cases by kind of act: gates, orders and assemblies; notices, declarations and certifications; the records that keep remedies; the counts themselves. Part IV turns to the conditions under which an act should be read again, judges the hypotheses the study registered before any case was coded, and sets the findings beside earlier accounts.
The reader will also find the study's failures. Seven predictions were registered before the coding; five did not hold, and they are reported where they belong. Two codings of the same fields by one model often agree; a further reading by a model of another family agrees less, and the study says on which fields its readers divided.
What the study does not do
It gives no medical advice and assesses no treatment. It does not judge any government, physician or committee. It does not use what Skeptron's records do not hold: for the most recent case, the study reads only the statement the record holds and leaves aside the larger history that followed, which the records do not contain. Where a source text survives in the public domain, the study reads it beside the record and says where the two differ; where none survives, or its terms forbid reproducing it, the study says only what the record and the reading notes say.
A note on voice
The study speaks of itself as "this study". It has no narrator and no first person. Where it reads a work further than the work's reading note allows, it says that the reading is its own. Where Skeptron's frame (sentences that recur across its cases with only names and dates changed) appears in a record, the study treats it as the archive's method, not as evidence about the case. Dates are given as the records and texts give them, with "BC" for years before the common era.