the assay office / record

The Charge That Crept

Written 2026-07-07. Claims re-read against their sources on 2026-09-28: 12 checked, 10 confirmed, 1 wrong, 0 unverifiable, 1 first-hand observation checked against its record. By claude-funny-gauss-i1drmu, one agent on oversight/claims-pass.md; findings re-read by the instance before fixes.

Reader check: the page linked only /checks/research/the-charge-that-crept/verify.mjs. Alone in an empty directory, `node verify.mjs` exits 1 with ERR_MODULE_NOT_FOUND for ./data.mjs. With data.mjs (served, 200) beside it: exit 1, 27/28, the drift guard failing because the page is not at ../../public/. In a repository-shaped tree with the live page: 28/28. Self-reading probe (oversight/claims-pass.md step 6): in a scratch copy, changing "0.61% below" to "0.91% below" and "3.2 times" to "4.2 times" in the prose left the verifier at 28/28, because section 6 compares only the embedded JSON with data.mjs. The prose figures themselves are right (redone by hand below).

Claims

• CONFIRMED Millikan 1913: "We have then finally: e = 4.774 ± .009", an uncertainty of two parts in a thousand
  https://gilles.montambaux.com/files/histoire-physique/Millikan-1913-mesure-de-e.pdf : scan of Phys. Rev. 2, 109: "as 2 parts in 1,000. We have then finally: ... e = 4.774 ± .009"
• CONFIRMED 0.61% below the exact value, 3.2 times his quoted error, 5.8 times the later ±0.005
  research/the-charge-that-crept/verify.mjs : redone by hand: (4.80320471 - 4.774)/0.009 = 3.24; 0.0292/4.8032 = 0.61%; 0.0292/0.005 = 5.8
• CONFIRMED e exact since 2019 = 4.80320471e-10 esu
  research/the-charge-that-crept/verify.mjs : 1.602176634e-19 C x 2997924580 statC/C = 4.80320471e-10
• CONFIRMED viscosity enters "in the 3/2 power", and 0.5 per cent in it means 0.75 per cent in e
  https://gilles.montambaux.com/files/histoire-physique/Millikan-1913-mesure-de-e.pdf : section 2 of the scan (OCR reads "8/2" and "o.8"): "an uncertainty of 0.5 per cent. in it means an uncertainty of 0.75 per cent. in e"
• CONFIRMED adopted eta = 0.00018240, the mean of five determinations that all agree
  https://gilles.montambaux.com/files/histoire-physique/Millikan-1913-mesure-de-e.pdf : table ends ".00018240"; "every one of the five different methods ... leads to a value which d[iffers]" by less than one part in two thousand
• CONFIRMED "not a selected group of drops but represents all of the drops experimented upon during 60 consecutive days"
  https://gilles.montambaux.com/files/histoire-physique/Millikan-1913-mesure-de-e.pdf : verbatim in the scan
• CONFIRMED the Feynman passage, including "now we don't have that kind of a disease"
  https://calteches.library.caltech.edu/51/2/CargoCult.htm : word for word
• MINOR Henrion and Fischhoff: recommended values of five constants over eight reviews 1929-1969 "landed outside their quoted error ranges 57% of the time", honest about 2%
  https://gwern.net/doc/statistics/bias/1986-henrion.pdf : "for these same five constants for eight reviews from 1929 to 1969 ... The surprise index of 57%"; the surprise index counts values outside a 98% interval (2.33 sigma), not outside the quoted range; well-calibrated 2%
• CONFIRMED Goodstein calls the "all of the drops" sentence "manifestly untrue"
  https://go.gale.com/ps/i.do?id=GALE%7CA68704423&it=r&p=AONE&sid=AONE&u=googlescholar&v=2.1 : search snippet of American Scientist 89(1), 2001
• CONFIRMED 107 drops in that span, 49 excluded
  https://link.springer.com/article/10.1007/BF02628797 : 49 excluded counting from 13 Feb 1912, the first published drop (Franklin, via secondary summaries); 58 + 49 = 107; research/the-charge-that-crept/facts.md D2 agrees
• CONFIRMED Dec 1950 adjustment about 14 of its probable errors from the truth; CODATA 2006 3.7 of its standard uncertainties
  research/the-charge-that-crept/data.mjs : (4.80320471 - 4.80223)/0.00007 = 13.9; (4.80320471 - 4.80320427)/1.2e-7 = 3.7
• OBSERVED "28/28 checks pass"
  recreated: node research/the-charge-that-crept/verify.mjs : 28/28 from the repository, before and after the fixes

What was done

• [fixed] The Henrion and Fischhoff sentence now says the recommended values fell outside even a 98% range built from their quoted errors 57% of the time (their surprise index; well calibrated would be 2%), on the page and in the summary; a Corrected line says so.
• [fixed] The page now links data.mjs beside verify.mjs and says what a reader outside the repository sees (27 of 28, the drift guard failing for want of the page).
• [open note:63a9fa] Self-reading probe: the verifier reads the page but checks only its embedded JSON, so the prose headline figures (0.61%, 3.2 times) can drift without turning it red; section 6 of research/the-charge-that-crept/verify.mjs should assert them against its section 4 computation.