6.1Two ways of knowing an epidemic in its first weeks
Both reports were federal, weekly and written while an epidemic was arriving, and they know in different ways. The reports of 1918 know by solicited hearsay: what state officers telegraphed to Washington, summarised alphabetically, with the date of going to press as the only common clock. They mark the incompleteness of what they hear and do not sort it further; a description of cases that might be influenza, an estimate, and a count of registered deaths stand on the page with equal weight. The report of 2020 knows by attribution and sorting. Each figure is assigned to its source and dated; the known is separated from the new, the evident from the unclear and the believed; the assumption is named.
The difference is partly a century's change in method, and partly one of purpose. The 1918 item is a digest of prevalence for a national readership, containing no action and no advice. The 2020 report accounts for an agency's own response and attaches interim guidance; it has to justify what it does. Purpose shapes what each says about change. The 1918 reports say nothing about what would change any action, because they record no decision of their own. The 2020 report says twice that new knowledge would change its guidance, without saying which. Neither builds in the planned points of reconsideration that Neustadt and Fineberg recommend: the one had nothing to reconsider, the other a great deal, and left the points open.
6.2What the timing debate needs, and where it comes from
The comparisons of cities by Hatchett and by Markel and their colleagues need, for each city, the start of its local epidemic, the date of its measures, and a mortality series on a fixed denominator. The federal reports of autumn 1918 supply none of the three for Philadelphia: a quiet entry followed by an epidemic one instead of a start date; measures reported clearly only for Washington; and a statewide running total of registered deaths to one day, split by cause, instead of a city's rate over a fixed period.
None of this says the timing thesis is wrong. Nothing in the weekly reports bears on whether early measures went with lower peaks. It says that the dates on which the thesis rests were not part of the federal running account of the hour, and that the platform's summary of Philadelphia takes them from the later study and says so. The record is honest about its source; its title, with the word "delayed", is that study's verdict on timing placed at the head of a summary of the event.
Turner's account of how disasters incubate ends with an inquiry that reconstructs what should have been seen (Turner 1978). The comparisons of 1918 cities are not inquiries in his sense, but the structure is close enough to be useful: the reconstruction supplies a moment, the start of the local epidemic, from which a delay can be measured, and the record of the hour does not contain that moment. Neither the later study nor the record claims that Philadelphia's officials knew on 28 September what the comparison later showed. A reader who wants to say they should have known has to say from what, and the federal journal of 27 September is not a promising answer.
6.3What the precaution debate can take from the two reports
The two reports do not settle the disagreement between Sunstein and the Late Lessons editors, and this article does not try to. They show two things that bear on it.
First, the report of 2020 acted on signs while saying, in its central paragraph, that its measures might not prevent what they aimed to slow. It invoked no principle. It named three aims, two of which, preparation and characterisation, would be served even if transmission could not be stopped: act now because the action buys time and knowledge whatever happens. That justification sits near Sunstein's weighing with margins, and the same report acknowledges ignorance and commits to monitoring, as the Late Lessons editors ask. It fits neither side cleanly. The two works argue about principles of decision; the report argues from purposes.
Second, neither report states the costs of the measures or who bore them. The 1918 reports record measures without comment. The 2020 report weighs a familiar disease against the new one in a single sentence about influenza vaccination and is otherwise silent about what screening, restriction of entry and quarantine cost and whom they affected. Sunstein's demand that risks on all sides be weighed finds one side stated in detail and the other not stated. The parent study's ledger, on its lenient reading, finds the same gap common in the platform's records (see the Appendix). It is the most consistent gap in both texts read here.
6.4The time of the act, the time of the evidence, the time of the count
The parent study asks of every record whether it lets a later reader keep apart three times: of the act, of the evidence and of the count. The reports of 1918 keep one clock, the press, and within it dates of telegrams and counts given sometimes and sometimes not. The time of any act in Philadelphia is not kept at all; a later reader can recover when the federal journal recognised the city's epidemic, and nothing about the city's own acts.
The report of 2020 keeps the three apart with care and, in one sentence, failed. Its counts are dated "As of" a day and attributed, its acts are dated, and its evidence is marked as preliminary, evident or unclear. Yet the emergency authorisation of the test was printed as 4 January, before the genome on which the test was based had been posted, and only an erratum in another issue restored it. Even a report built to keep the times apart could merge them by a slip.
The platform records keep the times apart by rule, which is their strength and their limit. The record of Philadelphia separates the first civilian reports, the parade, the order and the source; the record of 11 March separates the statement from the start of infection and the announced count from the later one. But each takes its dates from a single later source, and the separation it enforces is among those dates, not between the record and the documents of the hour. The record of Philadelphia does not know that the federal journal of 27 September did not name the city; the record of March does not mention the emergency declared on 30 January.
6.5What the evidence cannot carry
Four limits should be stated plainly. The reports of 1918 are read in paraphrase from an uncorrected machine reading, without their figures; a reader with the page images could quote them and give their counts. The reading rests on the order and kind of their entries and on their silences, which no transcription error could create. The weekly reports are a federal digest, not the city's records, and nothing here says what Philadelphia's own offices knew on 28 September or 3 October. The report of 2020 covers the weeks from 31 December 2019 to 4 February 2020 and one country's federal response; it is not evidence about later measures or about the effect of those it describes. And the platform records are modern summaries, each built on one later source: the Philadelphia record's dates are Hatchett and colleagues' dates, and the March record's subject is a statement this article has not read.