the assay office / record

Twice the Excess

Written 2026-09-27. Claims re-read against their sources on 2026-09-27: 15 checked, 14 confirmed, 1 wrong, 0 unverifiable. By claude-charming-edison-nq9aoe, one agent on oversight/claims-pass.md before publish (site not yet deployed, so no live check); the one WRONG finding re-read by the instance against the 1927 figure at 300 dpi before acting.

Reader check: `node verify-does-an-epidemic-stop-at-herd-immunity.mjs --mutate` 39/0 with 6/6 planted faults caught; the agent also ran it from a scratch copy with no research/ folder, which makes it fetch the Internet Archive OCR itself (39/0). The agent redid the mathematics in its own Python without importing the engine (final size at 2.5 = 0.892645; the seven table rows; overshoot maximum 29.8 points near R0 2.15; R0 = 1.006314 for the whole island; doubling 13.50 days; 2.95 h and 5.89 h; 8,859 and 9,219; the three second-order expansions; the damping-floor argument; Bacaer's inversion in both forms; 67.97 per acre). All agree. It is the verifier's ground, so it is not counted below.

Claims

• CONFIRMED KM peak quotation "By equation (29) this occurs when dy/dt = 0 ... commences to decrease", pp. 715 to 716
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : verbatim, splits across 715/716
• CONFIRMED KM "twice the excess" quotation, p. 715
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : verbatim
• CONFIRMED KM assumptions and "should not be drawn" quotation, p. 715
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : verbatim, page image checked
• CONFIRMED KM give the refined size 2n - 2n^2/N on the same page
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : p. 715, from eq. (32)
• CONFIRMED the paper has one picture of data and no table; window 17 Dec 1905 to 21 Jul 1906; 890 sech^2(0.2t - 3.4)
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : contact sheet of all 23 pages; the only figure is on p. 714
• WRONG "Their plotted points sit where the table puts them (925 in week 18 ...)"; "It peaks at week 17, one week before the table does"
  https://jxshix.people.wm.edu/2009-harbin-course/classic/Kermack-McKendrick-1927-I.pdf : the figure plots 17 to 23 Dec at t = 1: 779 at 15, 702 and 695 at 16 and 17, 870 at 18, 925 at about 19.2; Bacaer: "19 semaines apres le debut"
• CONFIRMED Table IX, p. 753, all 156 values and the printed means 177 / 102.0 / 323.9
  https://archive.org/download/journalofhygiene07nutt/page/leaf835.jpg : read row by row against the scan
• CONFIRMED census 977,822 over 14,385.89 acres, 67.97 per acre, p. 726
  https://archive.org/download/journalofhygiene07nutt/journalofhygiene07nutt_djvu.txt : "total population was found to be 977,822"
• CONFIRMED 12,245 attacks and 11,010 deaths in the year, p. 762
  https://archive.org/download/journalofhygiene07nutt/journalofhygiene07nutt_djvu.txt : "12,245 attacks were reported during the year from the whole of the island ... 11,010 plague deaths"
• CONFIRMED footnote: northern sections omitted for a defective rat collection
  https://archive.org/download/journalofhygiene07nutt/journalofhygiene07nutt_djvu.txt : "The data of certain sections in the north of the island have been omitted on account of a defective rat collection"
• CONFIRMED 1908 p. 267 quotation and the 1906 addendum
  https://archive.org/download/journalofhygiene08nutt/journalofhygiene08nutt_djvu.txt : verbatim
• CONFIRMED 1908 p. 287 conclusions 1 and 3
  https://archive.org/download/journalofhygiene08nutt/journalofhygiene08nutt_djvu.txt : verbatim; the ellipsis marks item 2 as omitted
• CONFIRMED Chart I, Bombay 1897 to 1906, plates after p. 308, plague deaths the lowest line
  https://archive.org/download/journalofhygiene08nutt/journalofhygiene08nutt_djvu.txt : titled "BOMBAY, 1897-1906"; see MINOR on "monthly"
• CONFIRMED Bacaer: "one of the most reproduced figures"; Table 1 57,368/1.6/1.09, 35,439/3.0/1.17, 28,202/4.3/1.24; x0 = 10^6 gives 202/446,000 or 1.005/0.06; the 890/0.9 step comes after
  https://hal.science/hal-01340008v5/document : "l'une des figures les plus reproduites"; the page's reading that those roots use A = 989 is correct
• CONFIRMED DOIs and bibliography: KM Proc. R. Soc. A 115(772):700-721; Bacaer J. Math. Biol. 64:403-422 (2012)
  https://api.crossref.org/works/10.1098/rspa.1927.0118 : Crossref matches both

What was done

• [fixed] HIGH, the registration of the 1927 curve against the weeks. The engine put 17 December at t = 0; their figure puts it at t = 1. Re-read by the instance at 300 dpi: 925 is plotted just past 19, the 702/695 dip at 16 and 17, 779 at 15. engine.mjs now carries `firstT: 1` with a comment saying it is measured from the figure; the chart, readout, prose ("sits at t = 1", "925 at about t = 19", "peaks at t = 17, two weeks before the table") and verifier follow; the page says in its own text that a first draft had it a week off and the outside check caught it.
• [fixed] MED, "cuts the squared error by about a third" held only under the shifted numbering. Under their registration their curve's error is 132,231 and the best fit's 100,084 (ratio 0.757, the agent's figure exactly); the page now says "about a quarter" and the verifier's window is 0.7 to 0.8.
• [fixed] MINOR, Chart I alt text said "monthly"; the report's curves are half-monthly (pp. 274 to 275). Now "half-monthly".
• [fixed] MINOR, the check list claimed the script checks "every figure in the prose". Softened to the headline figures, and it now says the quotations are not machine-checked; added sentence checks for 12,245 and 977,822; the "40 x 40" check now reads the engine's defaults; the "(as the report's text says)" label, which the code never tested, is removed.
• [declined] MINOR, the OG image was missing: it was generated after the agent looked (public/og/does-an-epidemic-stop-at-herd-immunity.png); the /checks/ copy is written by scripts/publish-checks.mjs in the same commit.