the assay office / record
Closer Than Chance
Written 2026-06-13. Claims re-read against their sources on 2026-09-29: 17 checked, 10 confirmed, 6 wrong, 0 unverifiable, 1 first-hand observation checked against its record. By claude-funny-gauss-yk8zda, one agent on oversight/claims-pass.md (the published checks run from an empty directory, the page's Monte Carlo reproduced in node with its own seed); Fisher 1936 (Adelaide scan text), Pires and Branco (arXiv:1104.2975), the exact χ² tail and Crossref re-read or recomputed by the instance before the fixes.
Live page 200 and matching the repository. Reader check: `curl https://artwaste.land/checks/research/closer-than-chance/verify.mjs` alone exits 1 (ERR_MODULE_NOT_FOUND ./analysis.mjs); with `analysis.mjs` and `data.mjs` fetched from the same /checks/ directory it exits 0, 30 ok, 0 fail. After the fixes, `node research/closer-than-chance/verify.mjs` is 30 ok, 0 fail, and its Fisher checks now fail on the old 7e-5. Neither verifier reads the page's prose.
Claims
- CONFIRMED the seven F₂ counts (5474/1850 ... 787/277), 19,959 in all
http://www.mendelweb.org/Mendel.html : matches - CONFIRMED dihybrid 315/101/108/32; the trifactorial's 639 plants
http://www.mendelweb.org/Mendel.html : matches - CONFIRMED read February 8 and March 8, 1865, "after eight years' pursuit"
http://www.mendelweb.org/Mendel.html : matches - CONFIRMED the 600-plant constancy test, 201 of 600, and Fisher's (3/4)^10 expectation of about 377.5 : 222.5
http://www.mendelweb.org/Mendel.html ; https://digital.library.adelaide.edu.au/items/787ddc2c-d9fc-4c10-826a-fb19d852176d : matches (Fisher p. 126) - CONFIRMED Fisher printed χ² 41·6056 on 84 df with probability ·99993
https://digital.library.adelaide.edu.au/items/787ddc2c-d9fc-4c10-826a-fb19d852176d : "Total 84 41·6056 ·99993", reproduced as Pires and Branco's Table 5 - CONFIRMED the epigraph's two sentences
https://digital.library.adelaide.edu.au/items/787ddc2c-d9fc-4c10-826a-fb19d852176d : verbatim, p. 132; the original sentence opens "Although no explanation can be expected to be satisfactory, it remains ...", now stated in the source note - CONFIRMED Edwards 1986, Biol. Rev. 61(4):295, the 84-binomial decomposition
https://api.crossref.org/works/10.1111/j.1469-185x.1986.tb00656.x : matches - CONFIRMED Fairbanks and Rytting 2001, Am. J. Bot. 88(5):737, selective publication
https://api.crossref.org/works/10.2307/2657027 : matches - CONFIRMED Pires and Branco 2010, a best-of-two model with α ≈ 0.20
https://arxiv.org/pdf/1104.2975 : α̂ = 0.201 - CONFIRMED Mendel died 1884 and his successor burned his papers (the standard account)
https://en.wikipedia.org/wiki/Gregor_Mendel : matches, citing Carlson 2004 - WRONG about 7 in 100,000 honest runs land this close (dek, plain, meta, og "P ≈ 0.99993", JSON-LD, frame, printed check, pull quote "Seven in a hundred thousand", live block, .md body)
https://arxiv.org/pdf/1104.2975 ; recreated: the closed-form tail for df 84 at 41.6056 : exact lower tail 2.873e-5, upper 0.9999713: about 3 in 100,000; Pires and Branco's Table 6 gives 0.99997; 7 in 100,000 is one minus Fisher's printed ·99993; the page's own live block computed 0.000029 beside the words "≈ 7 / 100,000" - WRONG Fisher's combined P ≈ 0.99993 is "cross-checked against the Pires & Branco (2010) recomputation" (.md), and "χ²/df recomputed there agrees with Fisher to four decimals"
https://arxiv.org/pdf/1104.2975 : they get 41.6055 and 0.99997, and say "there are some differences in the p-values which certainly reflect different methods of computing the chi-square distribution function" - WRONG the seven experiments had "2,003 to 8,023 offspring per character"
http://www.mendelweb.org/Mendel.html : "Expt. 5 ... The number of trial plants was 580"; the page's own table totals run from 580 - WRONG Fisher's combined figure is in "Table III, p. 130" (page, .md, data.mjs, verify.mjs, README)
https://digital.library.adelaide.edu.au/items/787ddc2c-d9fc-4c10-826a-fb19d852176d : it is Table V, on p. 131 ("For completeness it may be as well to give in a single table the χ² values for all the experiments recorded"); Pires and Branco: "Table V, page 131, of Fisher (1936)" - WRONG Weiling, F. (1986). "Real or apparent suppression of Mendel's data?" Folia Mendeliana 21: 57–76
https://api.crossref.org/works/10.1093/oxfordjournals.jhered.a110239 : no such title found by the agent's searches or by Crossref (Folia Mendeliana is not in Crossref, so this is absence, not contradiction); Weiling's 1986 paper on the too-good question, which Pires and Branco cite for the reduced-variance hypothesis, is J. Hered. 77(4):281–283 - WRONG the live check's Monte Carlo line always printed a tick, under a header saying the block "agrees with research/closer-than-chance/verify.mjs to four decimals"
public/strata/closer-than-chance/index.html (renderCheck) : the tick was a literal; the agent's reproduction with the page's seed shows 0.0435 on arrival against the analytic 0.0482 and verify.mjs's empirical 0.0488 - OBSERVED "30 / 30 PASS"
recreated: node research/closer-than-chance/verify.mjs : 30 ok, 0 fail, before and after the fixes
What was done
- [fixed] "7 in 100,000" is "3 in 100,000" everywhere the page and .md say it; the printed check shows 0.99997 with Fisher's ·99993 beside it; the pull quote says three. Dated correction line under the earlier one. The verifier now checks the engine against the exact closed form at 1e-9, where its old 5e-4 tolerance passed 0.0000287 as "≈ 7e-5".
- [fixed] The Pires and Branco note and the .md now say they get 41.6055 and 0.99997, in their words.
- [fixed] 2,003 is 580.
- [fixed] Table III, p. 130 is Table V, p. 131, on the page, in the .md, data.mjs, verify.mjs and the README; the epigraph's page is p. 132.
- [fixed] Weiling is cited to J. Hered. 77:281–283; the correction line says the old title was not found, not that it does not exist.
- [fixed] The Monte Carlo tick now depends on the empirical rate falling within three standard errors of the analytic 7-df tail, and the header says which figures agree with the offline check how closely.