--- slug: wisdom-of-crowds passed: 2026-10-14 by: the assay line (research/assay-line/cited-1014/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent checked: 212 confirmed: 172 wrong: 38 unverifiable: 0 observed: 0 stale: 2 --- This was the directed cited-1014 correction pass for wisdom-of-crowds, read at 67088387e0. The fixer independently re-read the evidence for all 39 triage items, corrected the authored page, metadata, analysis and verifier text where permitted, rejected K2 with evidence, and carried forward the checkers' source-backed CONFIRMED claims after deduplication by source-record identifier and removal of claims overridden by triage. Those carried lines retain their checker provenance and pinned sources; they are not represented as 171 new independent source visits. The 15 film findings remain unresolved in the published video because this task prohibits rendering or re-encoding. The repair worker revised the film source narrative on 2026-10-14, preserving quoted words and labeling the misquotation explicitly; the media remain unchanged. The page and its markdown now disclose that boundary. The two stale placard findings await the landing rebuild from this dated record. MINOR lines count in wrong; duplicate triage IDs have separate claim and finding lines. V1 and V2 record additional false or overbroad verifier wording encountered during the required verifier review. ## Claims - MINOR | [X1] but also clerks and shopkeepers who didn't. | https://www.nature.com/articles/075509f0 | Galton's Vox Populi names butchers and farmers; The Ballot-Box describes non-experts but uses clerks only in a horse-racing analogy and says occupational proportions were unknown. This supports removing the embellishment, not claiming that no clerks attended. - WRONG | [X2] It is always smaller | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | Page's error decomposition subtracts prediction diversity. Recomputed counterexample: guesses [1,1], truth 0, give diversity 0 and both squared errors 1. The bound concerns average individual squared error; strict improvement requires positive diversity. - WRONG | [X3] the fraction of members it beats climbs toward everyone. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | Executed the authored thState and updTh with bias 0 and sigma 2, 28, 60. All three displayed 100%; the fraction does not climb. The changed branch applies only at bias exactly 0; bias 1 and 2 now receive the shared-bias description. - MINOR | [X4] the true value often ended up outside it. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | Lorenz et al., Discussion p. 9024, report movement toward outer regions and possible exclusion of the truth. The paper does not establish that the truth often left the whole range. - MINOR | [X5] drifting away from the true value. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | The authored coState update is e[i] += lam*(m-e[i]). At influence 0.22 and rounds 0, 1, 40, the mean remains 20 to floating-point precision, while the band narrows and ultimately excludes truth 0. It contracts around an existing bias. - WRONG | [X6] a majority vote is more reliable than any single voter | https://plato.stanford.edu/entries/jury-theorems/ | Stanford's Jury Theorems section 1.1 states equal competence. Independent recomputation over eight outcomes for probabilities [.99,.51,.51] gives majority accuracy .754902, below .99. The strict claim needs an odd group of at least three and common probability strictly between .5 and 1. - MINOR | [X7] "more people" stops adding wisdom | https://snap.stanford.edu/class/cs224w-readings/bikhchandani92fads.pdf | Bikhchandani et al. define cascades by choices that disregard private information. Partial dependence alone does not force all gains to stop: a 10% shared-signal mixture with otherwise independent .6-competent voters improves from .6 toward .96. Shared inputs can limit or eliminate gains. - MINOR | [X8] percussion density is the collapse model's own diversity | git show 2fcfb9c17f:research/wisdom-of-crowds/film/make-audio.mjs | Read make-audio.mjs: draw() < densityAt(t0) selects amp 0.30 instead of 0.10, and every beat is still synthesized. Normalized diversity controls loud-accent probability. The pitch ladder is minor pentatonic. The caption and source comments/log now describe the implemented mapping; synthesis code and media are unchanged. - WRONG | [X9] preserves its running maximum. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs; recreated: node probe.mjs | Reproduced with the actual analysis and shared kit: eight target-1 pair losses give maximum 43.98194885253906. A later target-0 win was prepended and lowered it to 10.348260020623409; target 2 appended and preserved it. Version 2 sorts completed pairs by their second arrival. Both cases now preserve the prior wealth path and maximum, and all interleaved completion prefixes remain stable. - MINOR | [X10] Diversity beats individual brilliance | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | Page distinguishes beating the average member from sometimes beating every member. The identity does not guarantee beating the most accurate person. The recipe now says can, with the squared-error condition, retaining the shared-bias limit. - STALE | [X11] Last verified 29 June 2026, when it was written. | git show 2fcfb9c17f:memory/log.d/2026-06-30T0345Z-wisdom-of-crowds-film.md; git show 2fcfb9c17f:memory/log.d/2026-06-29T2200Z-wisdom-of-crowds.md | The June 30 film log records an all-pass run after the stated June 29 check. This assay and the repeated layer verifiers are dated 2026-10-14. Under the triage's explicit Last verified exception, this record supplies the date for landing; no Corrected or Updated line is added. The generated placard itself remains unchanged because its shared writer is out of scope. - WRONG | [X12] The more they disagree, the more it beats them: 100% and climbing. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | Executed the authored thState and updTh with bias 0 and sigma 2, 28, 60. All three displayed 100%; the fraction does not climb. The changed branch applies only at bias exactly 0; bias 1 and 2 now receive the shared-bias description. - WRONG | [X13] No network request has been made. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | Read openLivingArm, declareAnalysis, fetchArm and the send handler. The analysis-file GET and live-arm GET occur before the preview opens; contribution POST is only in the send handler. The new copy says the estimates have not been submitted and does not imply the earlier GETs sent them. - WRONG | [X14] the ox truly weighed 1198. | https://arxiv.org/pdf/1410.3989 | Wallis section 2 gives the organiser's dressed weight as 1197 lb and identifies 1198 as the published slip. film.html lines 306 and 318 call 1198 true. This is false as history; the explicitly qualified printed-centile reconstruction on the page is supported. - MINOR | [X15] was a footnote he added later. | https://www.nature.com/articles/075509f0 | Nature labels The Ballot-Box a Letter, published 28 March 1907. Wallis section 3 describes the reply to Hooker. The mean was in a separate reply letter three weeks after Vox Populi, not a footnote added to it. - MINOR | [X16] Some can see the room and revise. Galton’s guessers never could | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | Vox Populi permits information competitors could acquire. It does not establish a ban on learning other estimates or measure the room's information flow. The film's absolute contrast is unsupported; the converse is not established either. - WRONG | [X17] their errors were independent. That was the hidden condition on the whole result. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | Galton allowed acquired information and did not measure error independence. The fixed-set squared-error identity holds even for dependent guesses. The film turns an unmeasured assumption into a fact and a necessary condition. - WRONG | [X18] The result seems more creditable to the trustworthiness of a democratic judgment than might have been expected. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_2.htm | The Vox Populi p. 451 scan and its OCR start the quotation with This result is, I think; the film instead starts The result seems. The source does not support that alteration. The page body already gives the original words. - MINOR | [X19] The average won by a pound | https://arxiv.org/pdf/1410.3989 | Wallis sections 2-3 distinguish mean 1197 versus printed weight 1198 from mean 1197 versus archival weight 1197. A one-pound error is not a one-pound winning margin. The film's unattributed popular-story gloss needs clarification outside the quoted words, not an invented attribution. - WRONG | [X20] (added) verifier | https://artwaste.land/verify-wisdom-of-crowds.mjs | Independent GETs on 2026-10-14 returned HTTP 404 and 4404 bytes at /verify-wisdom-of-crowds.mjs; /checks/verify-wisdom-of-crowds.mjs returned HTTP 200 and 11105 bytes. The caption now points to the published check. - WRONG | [X21] (added) Film: "Their middlemost guess was 1207 lb"; "The crowd, as a body, was wrong by nine pounds" | https://arxiv.org/pdf/1410.3989 | Wallis section 2 identifies the 394th ranked observation as 1208 lb and the dressed weight as 1197 lb, giving an 11 lb median error. The film presents printed 1207 and nine pounds without the printed-account qualification. - WRONG | [X22] (added) a majority beats any one of them | https://plato.stanford.edu/entries/jury-theorems/ | Stanford's Jury Theorems section 1.1 states equal competence. Independent recomputation over eight outcomes for probabilities [.99,.51,.51] gives majority accuracy .754902, below .99. The strict claim needs an odd group of at least three and common probability strictly between .5 and 1. - WRONG | [X23] (added) average individual error rises toward the (unchanged) crowd error | git show 2fcfb9c17f:public/checks/verify-wisdom-of-crowds.mjs | Both verifier copies asserted ai1 < ai0 + EPS but labeled it rises. The rerun gives average individual squared error 573.1 to 0.3 while crowd error stays 0.27. Corrected the output label and both conflicting explanatory comments, preserving the arithmetic assertion. - MINOR | [K1] Eight hundred or so people entered - butchers and farmers who knew cattle, but also clerks and shopkeepers who didn't. | https://galton.org/essays/1900-1911/galton-1907-vox-populi.pdf | Galton's Vox Populi names butchers and farmers; The Ballot-Box describes non-experts but uses clerks only in a horse-racing analogy and says occupational proportions were unknown. This supports removing the embellishment, not claiming that no clerks attended. - CONFIRMED | [K2] that the crowd's guess was "right to within a pound": is the mean, about 1197 lb. | https://arxiv.org/pdf/1410.3989 | REJECTED with evidence: the page introduces an unattributed popular-story gloss, not a named author's verbatim quotation. Wallis section 3 supports mean 1197 versus printed outcome 1198, a one-pound error, and the page immediately supplies the archival correction. Missing exact colloquial wording in the source does not establish a false attributed quotation. Kept unchanged. - WRONG | [K3] A crowd's collective error is not usually smaller than its members'. It is always smaller: provably, by an exact identity. For any set of guesses and any true value: | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | Page's error decomposition subtracts prediction diversity. Recomputed counterexample: guesses [1,1], truth 0, give diversity 0 and both squared errors 1. The bound concerns average individual squared error; strict improvement requires positive diversity. - MINOR | [K4] In 2011, Jan Lorenz and colleagues sat 144 people down to estimate real quantities about their own country: things like the number of new immigrants in a year, or the length of the Swiss border: over several rounds, paying them for accuracy. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | Lorenz Experimental Design p. 9021 recruits 144 ETH Zurich students without establishing nationality. Materials and Methods p. 9024 asks about Zurich's 2006 gain in inhabitants and the Switzerland-Italy border. Repeated rounds and payment for accuracy are supported. - MINOR | [K5] Worse: as the range of guesses shrank, the true value often ended up outside it. The crowd talked itself into a tight, certain, wrong consensus. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | Lorenz et al., Discussion p. 9024, report movement toward outer regions and possible exclusion of the truth. The paper does not establish that the truth often left the whole range. - MINOR | [K6] A crowd of estimates that converges to a narrow band as rounds of social influence proceed, drifting away from the true value. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | The authored coState update is e[i] += lam*(m-e[i]). At influence 0.22 and rounds 0, 1, 40, the mean remains 20 to floating-point precision, while the band narrows and ultimately excludes truth 0. It contracts around an existing bias. - STALE | [K7] Last verified 29 June 2026, when it was written. | memory/log.d/2026-08-21T0000Z-the-living-experiments-wave.md | The June 30 film log records an all-pass run after the stated June 29 check. This assay and the repeated layer verifiers are dated 2026-10-14. Under the triage's explicit Last verified exception, this record supplies the date for landing; no Corrected or Updated line is added. The generated placard itself remains unchanged because its shared writer is out of scope. - MINOR | [K8] "uploadDate":"2026-07-09" | git show 2fcfb9c17f:memory/log.d/2026-06-30T0345Z-wisdom-of-crowds-film.md | The 2026-06-30 film log says the film was shipped and embedded, and records its passing verifier. This supports uploadDate 2026-06-30 rather than 2026-07-09. dateModified is separately moved to this revision date. - WRONG | [K9] the ox truly weighed 1198. | https://arxiv.org/pdf/1410.3989 | Wallis section 2 gives the organiser's dressed weight as 1197 lb and identifies 1198 as the published slip. film.html lines 306 and 318 call 1198 true. This is false as history; the explicitly qualified printed-centile reconstruction on the page is supported. - MINOR | [K10] was a footnote he added later. | https://www.nature.com/articles/075509f0 | Nature labels The Ballot-Box a Letter, published 28 March 1907. Wallis section 3 describes the reply to Hooker. The mean was in a separate reply letter three weeks after Vox Populi, not a footnote added to it. - MINOR | [K11] Galton’s guessers never could | https://galton.org/essays/1900-1911/galton-1907-vox-populi.pdf | Vox Populi permits information competitors could acquire. It does not establish a ban on learning other estimates or measure the room's information flow. The film's absolute contrast is unsupported; the converse is not established either. - WRONG | [K12] their errors were independent. That was the hidden condition on the whole result. | https://galton.org/essays/1900-1911/galton-1907-vox-populi.pdf | Galton allowed acquired information and did not measure error independence. The fixed-set squared-error identity holds even for dependent guesses. The film turns an unmeasured assumption into a fact and a necessary condition. - WRONG | [K13] The result seems more creditable to the trustworthiness of a democratic judgment than might have been expected. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_2.htm | The Vox Populi p. 451 scan and its OCR start the quotation with This result is, I think; the film instead starts The result seems. The source does not support that alteration. The page body already gives the original words. - MINOR | [K14] The average won by a pound | https://arxiv.org/pdf/1410.3989 | Wallis sections 2-3 distinguish mean 1197 versus printed weight 1198 from mean 1197 versus archival weight 1197. A one-pound error is not a one-pound winning margin. The film's unattributed popular-story gloss needs clarification outside the quoted words, not an invented attribution. - MINOR | [K15] (added) Film at 0:29: "787 legible tickets - butchers and clerks alike." | https://galton.org/essays/1900-1911/galton-1907-vox-populi.pdf | Galton's Vox Populi names butchers and farmers; The Ballot-Box describes non-experts but uses clerks only in a horse-racing analogy and says occupational proportions were unknown. This supports removing the embellishment, not claiming that no clerks attended. - WRONG | [K16] (added) Reader-run verifier output: "ok average individual error rises toward the (unchanged) crowd error avgIndErr 573.1 → 0.3; crowdErr stays 0.27" | public/checks/verify-wisdom-of-crowds.mjs | Both verifier copies asserted ai1 < ai0 + EPS but labeled it rises. The rerun gives average individual squared error 573.1 to 0.3 while crowd error stays 0.27. Corrected the output label and both conflicting explanatory comments, preserving the arithmetic assertion. - WRONG | [V1] The verifier calls 1198 lb the actual dressed weight and presents the printed median error as historical. | https://arxiv.org/pdf/1410.3989; verify-wisdom-of-crowds.mjs; public/checks/verify-wisdom-of-crowds.mjs | The same false historical value as X14 appears outside the film. Wallis section 2 gives weight 1197 lb, median 1208 lb, and median error 11 lb. Both verifier copies now identify 1198/1207 as printed figures. The reconstruction arithmetic is unchanged. - MINOR | [V2] Every number on the page is reproducible from the standalone verifier. | verify-wisdom-of-crowds.mjs; public/checks/verify-wisdom-of-crowds.mjs | The file checks four arithmetic models; it does not read the page or check every historical or living-arm figure. Its success message now reports only that the arithmetic checks above passed. - CONFIRMED | A crowd can out-guess its own smartest member | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | [001 crowd-best-member] Lorenz introduction: "closer to the true value than all of the separate individual or even expert guesses". The claim says can, not must. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | and there is an exact equation for why. The unsettling part is that the same equation, read backwards, tells you precisely how a wise crowd becomes a confident, wrong one. What it spends is the one thing that made it wise. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [002 equation-crowd-failure] Page supplies the exact error decomposition; Lorenz supplies the separate empirical confidence effect. The identity alone is not a psychological law. Read with the explicitly named consensus model below. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | In 1906, at a country fair in Plymouth, the statistician Francis Galton watched a weight-judging competition: pay sixpence, guess the dressed weight of a fat ox after slaughter, win a prize if you're close. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [003 galton-fair] Galton: "cards, for 6d. each" and "after it had been slaughtered and" dressed. Surowiecki introduction supplies 1906; Wallis discusses Galton's memoir account of his visit. The competition, place, payment, and prize description agree. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton collected the tickets afterward. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [005 galton-tickets] Galton says the tickets "were kindly lent me for examination after they had fulfilled their immediate purpose". Collected here means obtained for analysis, not retained permanently. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | He was 84, a believer in the rule of experts | https://arxiv.org/pdf/1410.3989 | [006 galton-age] Wallis: "Galton reached the age of 85 on February 16, 1907"; he was 84 at the 1906 fair. The ideological clause is covered separately by entry 7, not by this age excerpt. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | a believer in the rule of experts, and he expected the data to expose the foolishness of the common voter. He found the opposite. | https://www.asecib.ase.ro/mps/TheWisdomOfCrowds-JamesSurowiecki.pdf | [007 galton-expert-belief] Surowiecki introduction explicitly says "he wanted to prove that the average voter was capable of very little" and describes his expectation being overturned. This is the cited secondary account, not a quotation of Galton's own stated intention. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Of the tickets, 787 were legible. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [008 legible-tickets] Galton: "After weeding thirteen cards out of the collection, as being defective or illegible, there remained 787 for discussion." The usable-ticket count agrees. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Their middlemost guess: the median, which Galton chose deliberately as the fairest "one vote, one value" verdict of the crowd: was 1207 lb. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [009 median-1207] Galton prints "the middlemost estimate is 1207 lb." and "1198 lb."; 1207 minus 1198 is 9, and 9/1198 is 0.7513%. "one vote one value" explains his median choice. The surrounding page explicitly distinguishes these printed numbers from the archival correction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | "one vote, one value" verdict of the crowd | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [010 one-vote-quote] Galton prints "the middlemost estimate is 1207 lb." and "1198 lb."; 1207 minus 1198 is 9, and 9/1198 is 0.7513%. "one vote one value" explains his median choice. The surrounding page explicitly distinguishes these printed numbers from the archival correction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The ox's dressed weight, as Galton printed it, was 1198 lb. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [011 printed-weight] Galton prints "the middlemost estimate is 1207 lb." and "1198 lb."; 1207 minus 1198 is 9, and 9/1198 is 0.7513%. "one vote one value" explains his median choice. The surrounding page explicitly distinguishes these printed numbers from the archival correction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The crowd, as a body, was wrong by nine pounds: three-quarters of one percent. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [012 nine-pounds] Galton prints "the middlemost estimate is 1207 lb." and "1198 lb."; 1207 minus 1198 is 9, and 9/1198 is 0.7513%. "one vote one value" explains his median choice. The surrounding page explicitly distinguishes these printed numbers from the archival correction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | This result is, I think, more creditable to the trustworthiness of a democratic judgment than might have been expected. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_2.htm | [013 galton-democratic-quote] The original second page contains the complete quotation exactly, including "This result is, I think". Independently read the same URL after the online check. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Francis Galton, "Vox Populi", Nature 75, 7 March 1907 | https://arxiv.org/pdf/1410.3989 | [014 galton-blockquote-reference] Wallis references: "Galton, F. (1907b). Vox populi. Nature 75 450–451. (March 7, 1907.)"; the original page header also gives March 7, 1907. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton published only the shape of the crowd: a table of its percentiles - not the raw tickets. Below is that distribution, rebuilt from his printed centiles (the full reconstruction is K. Wallis, Statistical Science 2014). | https://arxiv.org/pdf/1410.3989 | [015 published-distribution] Wallis describes the tabulated 5, 10, ..., 95 percentiles and labels his histogram "original data". Galton's two printed pages provide a percentile table, not 787 individual entries. The page discloses its illustrative interpolation. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | On Galton's printed figures the crowd's median missed by 9 lb; its mean by about 1. Almost no single person did that well. | https://arxiv.org/pdf/1410.3989 | [017 ox-error-verdict] Wallis reports mean 1197 against printed outcome 1198, and Figure 1 says "only a single winner, and no immediately adjacent runner-up". This supports the near-perfect mean claim; the displayed histogram remains a reconstruction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | that the crowd's guess was "right to within a pound": is the mean, about 1197 lb. | https://arxiv.org/pdf/1410.3989 | [018 pound-quote] Wallis supplies 1197 and printed 1198, which differ by one pound. "right to within a pound" is the page's summary of the popular story, not a verbatim Wallis or Galton quotation. The source record incorrectly types this gloss as a sourced verbatim quote. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | But the mean is not the statistic Galton reported. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [019 reported-statistic] Galton prints "the middlemost estimate is 1207 lb." and "1198 lb."; 1207 minus 1198 is 9, and 9/1198 is 0.7513%. "one vote one value" explains his median choice. The surrounding page explicitly distinguishes these printed numbers from the archival correction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | He championed the median, 1207, as the true voice of the people, and he supplied the famous near-perfect average only three weeks later, in a reply letter to Nature ("The Ballot-Box", 28 March 1907), after a correspondent wrote in asking for the mean. | https://www.nature.com/articles/075509f0 | [020 late-mean-letter] Nature labels the item "Letter" and "Published: 28 March 1907"; Galton gives "1197 lb." and refers to Hooker's March 21 response to his March 7 article. The interval is 21 days. Crossref agrees on the reference. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Kenneth Wallis re-checked Galton's worksheets (Statistical Science, 2014) and found transcription slips in the printed article: the ranked middlemost estimate was actually 1208 lb, and the organiser's letter put the dressed weight at 1197 lb. | https://arxiv.org/pdf/1410.3989 | [021 archival-corrections] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | So the median really missed by eleven pounds, not nine, and the mean missed by nothing at all. | https://arxiv.org/pdf/1410.3989 | [022 archival-error] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | (crowd's error) = (average individual error) − (diversity of the guesses) | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [025 diversity-equation] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | This is not a statistical tendency. It is an algebraic identity: the same parallel-axis decomposition that splits any variance: so it holds exactly, every time, to the last decimal. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [026 algebra-identity] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Scott Page named it the Diversity Prediction Theorem. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [027 page-named-theorem] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Because diversity (the spread of the guesses) can never be negative, the crowd is never worse than its average member, and is strictly better the moment people disagree. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [028 average-member-bound] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Disagreement is not noise to be averaged away. Disagreement is the accuracy. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [029 disagreement-accuracy] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Diversity is free accuracy: but only the idiosyncratic kind, the errors that point in every direction and cancel. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | [033 idiosyncratic-diversity] Lorenz: estimation errors are "large but unbiased such that they cancel each other out". This supports the qualification about idiosyncratic errors; it does not remove common bias. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | This is the honest boundary the cheerful version omits: a crowd cancels the part of the error its members don't share. The part they share: a rumour, a framing, a number everyone half-remembers: it averages straight into the answer and calls it consensus. | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [035 shared-error-bound] Vul and Pashler: "averaging should not reduce the bias of estimates". The centered-noise plus common-bias decomposition in the page reproduces this distinction. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton judged his entries "unbiassed by passion and uninfluenced by oratory and the like", | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [036 galton-unbiassed] Both Galton quotations are present in the HTMLParser extraction at the cited URL, including "probably guided by such information as they might pick up". These are positive evidence for caution about independence, not measured independent errors. The online NOT FOUND results were extraction artifacts. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | though he allowed that some competitors were "probably guided by such information as they might pick up". | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [037 galton-guided] Both Galton quotations are present in the HTMLParser extraction at the cited URL, including "probably guided by such information as they might pick up". These are positive evidence for caution about independence, not measured independent errors. The online NOT FOUND results were extraction artifacts. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Independence was an assumption about that room, never a measured fact; what the result actually needed was errors that did not all lean the same way. | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [038 independence-assumption] Both Galton quotations are present in the HTMLParser extraction at the cited URL, including "probably guided by such information as they might pick up". These are positive evidence for caution about independence, not measured independent errors. The online NOT FOUND results were extraction artifacts. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Some saw nothing of others' answers; some saw the group's estimates and could revise. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [040 lorenz-conditions] Lorenz contrasts no information, aggregated information, and full information; subjects could revise in rounds 2 through 5. The record's abstract excerpt and the full methods agree. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The result was precise and bleak: social influence made the estimates converge : the crowd's diversity collapsed: and made people far more confident. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [041 lorenz-convergence] Lorenz reports convergence and increased confidence with no significant collective accuracy improvement. Read accuracy as the collective outcome in this context. The empirical experiment is separately identified from the page's illustrative DeGroot model. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | It did not make them more accurate. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [042 lorenz-accuracy] Lorenz reports convergence and increased confidence with no significant collective accuracy improvement. Read accuracy as the collective outcome in this context. The empirical experiment is separately identified from the page's illustrative DeGroot model. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | voters (independent, each 60% right) majority is correct 1 60.0% 9 73.3% 31 87.2% 101 97.9% 501 >99.99% | https://plato.stanford.edu/entries/jury-theorems/ | [051 condorcet-table] Independent binomial sums reproduce 60%, 73.343232%, 87.161827%, 97.910331%, 99.9996975% for n=1,9,31,101,501 at p=.6. At p=.4 and n=501 the result is 0.0003025010%. The table and reverse example round correctly; the generic Stanford excerpt itself does not contain these numbers. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | But flip each voter to below half: say a shared misconception makes everyone right only 40% of the time: and the theorem runs in reverse with equal force: a crowd of 501 such voters reaches the wrong verdict essentially every time (~0.0003% chance of being right). | https://plato.stanford.edu/entries/jury-theorems/ | [052 reverse-condorcet] Independent binomial sums reproduce 60%, 73.343232%, 87.161827%, 97.910331%, 99.9996975% for n=1,9,31,101,501 at p=.6. At p=.4 and n=501 the result is 0.0003025010%. The table and reverse example round correctly; the generic Stanford excerpt itself does not contain these numbers. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Information cascades (Bikhchandani, Hirshleifer & Welch, 1992; confirmed in the lab by Anderson & Holt, 1997) are this failure at its sharpest: rational people, each watching the choices before them, can all rationally ignore their own private evidence and stampede: together, confidently: off a cliff. | https://snap.stanford.edu/class/cs224w-readings/bikhchandani92fads.pdf | [054 information-cascades] Bikhchandani et al. define rational cascades after observing earlier actions. Anderson and Holt: "Rational cascades formed in most periods in which such an imbalance occurred"; their text explicitly discusses reverse cascades into the incorrect state. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | confirmed in the lab by Anderson & Holt, 1997 | https://econweb.ucsd.edu/~jandreon/Econ264/papers/Holt%20Anderson%20AER%201997.pdf | [055 cascades-laboratory] Bikhchandani et al. define rational cascades after observing earlier actions. Anderson and Holt: "Rational cascades formed in most periods in which such an imbalance occurred"; their text explicitly discusses reverse cascades into the incorrect state. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | the 95% confidence sequence covers its bounded mean at every viewing time simultaneously, under its assumptions. | https://arxiv.org/pdf/1810.08240 | [072 cs-guarantee] Howard defines a sequence "uniformly valid over an unbounded time horizon" and warns that repeated fixed-sample monitoring inflates Type I error. This confirms the stated conditional statistical principle, not certification of every implementation detail; the separate different-reader running-maximum defect is entry 75. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | An ordinary fixed-n interval repeatedly refreshed under this live counter would not make that guarantee. | https://arxiv.org/pdf/1810.08240 | [073 fixed-n-live] Howard defines a sequence "uniformly valid over an unbounded time horizon" and warns that repeated fixed-sample monitoring inflates Type I error. This confirms the stated conditional statistical principle, not certification of every implementation detail; the separate different-reader running-maximum defect is entry 75. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Vul and Pashler's 2008 cohort had 428 internet-subject-pool participants answer eight percentage questions. | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [077 vul-cohort] Vul and Pashler: "We recruited 428 participants from an Internet-based subject pool" and "asked them eight questions". Methods and the 2014 question table confirm percentage questions. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The immediate group had 255 analyzed participants; the group that returned three weeks later had 173. | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [078 vul-groups] Steegen Tables 1 and 2 give n=255 and n=173 for the original immediate and delayed groups. Their sum is 428. Vul's reported degrees of freedom, 254 and 172, independently corroborate these counts. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Their interpolation valued a second immediate guess at 0.11 of a different person's opinion, about one tenth, and a second delayed guess at 0.32, about one third. | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [079 vul-interpolation] Vul and Pashler report "1.11" and "1.32" different-person equivalents, with lambda .11 and .32; the paragraph explicitly paraphrases these as one tenth and one third of another opinion. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | immediate: guess 1 vs average 2.25 4.41 | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [081 reanalysis-table-1] Vul original gives "t(254)" with 2.25; Steegen Table 1 gives 4.41 for guess 1 versus average. Each source supplies its own column, not both numbers by itself. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | immediate: guess 1 vs average 2.25 4.41 | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [082 original-table-1] Vul original gives "t(254)" with 2.25; Steegen Table 1 gives 4.41 for guess 1 versus average. Each source supplies its own column, not both numbers by itself. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | immediate: guess 2 vs average 6.08 9.90 | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [083 reanalysis-table-2] Vul original gives "t(254)" with 6.08; Steegen Table 1 gives 9.90 for guess 2 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | immediate: guess 2 vs average 6.08 9.90 | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [084 original-table-2] Vul original gives "t(254)" with 6.08; Steegen Table 1 gives 9.90 for guess 2 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | three weeks: guess 1 vs average 3.94 6.22 | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [085 reanalysis-table-3] Vul original gives "t(172)" with 3.94; Steegen Table 2 gives 6.22 for delayed guess 1 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | three weeks: guess 1 vs average 3.94 6.22 | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [086 original-table-3] Vul original gives "t(172)" with 3.94; Steegen Table 2 gives 6.22 for delayed guess 1 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | three weeks: guess 2 vs average 6.59 9.85 | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [087 reanalysis-table-4] Vul original gives "t(172)" with 6.59; Steegen Table 2 gives 9.85 for delayed guess 2 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | three weeks: guess 2 vs average 6.59 9.85 | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [088 original-table-4] Vul original gives "t(172)" with 6.59; Steegen Table 2 gives 9.85 for delayed guess 2 versus average. Read both sources to verify the whole row. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | gain contrast, df 426 2.12 2.68, p=.008 | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [089 reanalysis-table-5] Steegen footnote 7 explicitly contrasts "t(426) = 2.12" with "t(426) = 2.68, p = 0.008". The original also prints 2.12. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Steegen and colleagues recomputed the right column after receiving the original records privately. | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [090 private-data-reanalysis] Steegen says raw records "were provided to us by Edward Vul". The table marks the corrected values "Computed from raw data" and explains the original underreported t values. The page accurately describes the reanalysis. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | 2008 primary paper, pp. 645-647 | https://gwern.net/doc/statistics/prediction/2008-vul.pdf | [094 vul-primary-citation] Read all three PDF pages: title, authors, 2008 copyright and pages 645, 646, 647 agree. Retried the record's exact DOI at Crossref and received matching 2008 metadata. The page's JHU link returned HTTP 403 here; the record's Gwern mirror returned 200. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Steegen et al.'s 2014 tables and correction note | https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2014.00786/full | [095 steegen-citation] The article identifies Steegen, Dewitte, Tuerlinckx and Vanpaemel (2014), "Measuring the crowd within again: a pre-registered replication study", volume 5:786; tables and footnote 7 contain the corrections. DOI matches. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Howard et al. 2021 | https://arxiv.org/pdf/1810.08240 | [097 howard-citation] Crossref for 10.1214/20-AOS1991 gives Howard, Ramdas, McAuliffe, Sekhon, Annals of Statistics 49(2), 1055-1080, published 2021-04-01. The currently fetched arXiv revision is dated 2022, so its displayed revision year is not the publication year. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The wisdom of crowds is real, and it is not a mystery or a miracle. It is one equation: a crowd's accuracy is its members' accuracy plus their diversity. | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [111 accuracy-plus-diversity] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Get independent estimates first : write them down, in private: then talk. The diversity you'd otherwise spend in the first five minutes of a meeting is the accuracy. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | [112 private-first] Lorenz recommends "request multiple independent opinions and aggregate these". The five-minute wording is advice illustrating the mechanism, not a measured five-minute experimental finding. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Confidence is not accuracy. When a group grows more certain, ask whether it learned something or just heard itself. Lorenz's crowds did the second and couldn't tell the difference. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [114 lorenz-confidence-lesson] Lorenz reports convergence and increased confidence with no significant collective accuracy improvement. Read accuracy as the collective outcome in this context. The empirical experiment is separately identified from the page's illustrative DeGroot model. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | crowd error = avg individual error − diversity | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [116 apparatus-equation] Page: "I call this the Diversity Prediction Theorem". His collective squared error equals average individual squared error minus prediction variance. Direct algebra and the 20,000-crowd verifier agree. The accuracy language is read as this stated decomposition, not a promise that arbitrary added noise improves an unchanged mean. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The Galton centiles (median 1207, q1 1162, q3 1236; printed dressed weight 1198 lb) are transcribed from his 1907 paper | https://galton.org/cgi-bin/searchImages/search/essays/pages/galton-1907-vox-populi_1.htm | [120 centiles-transcribed] The original Galton table and prose give median 1207, quartiles 1162 and 1236, and printed dressed weight 1198. Read the table itself, not just the record's imperfect OCR punctuation. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Wallis's archival check corrects the printed record itself: ranked median 1208, announced weight 1197, an 11 lb median miss and a zero-error mean. | https://arxiv.org/pdf/1410.3989 | [122 wallis-apparatus] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Real social influence is messier; the Lorenz finding is what was measured in people, and it agrees. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [125 lorenz-model-agreement] Lorenz reports convergence and increased confidence with no significant collective accuracy improvement. Read accuracy as the collective outcome in this context. The empirical experiment is separately identified from the page's illustrative DeGroot model. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | F. Galton, "Vox Populi", Nature 75:450–451 (1907) | https://arxiv.org/pdf/1410.3989 | [126 ref-galton] Wallis references: "Galton, F. (1907b). Vox populi. Nature 75 450–451. (March 7, 1907.)"; the original page header also gives March 7, 1907. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | "The Ballot-Box", Nature 75:509–510 (1907), the reply letter giving the mean | https://www.nature.com/articles/075509f0 | [127 ref-ballot] Nature labels the item "Letter" and "Published: 28 March 1907"; Galton gives "1197 lb." and refers to Hooker's March 21 response to his March 7 article. The interval is 21 days. Crossref agrees on the reference. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | K. F. Wallis, "Revisiting Francis Galton's Forecasting Competition", Statistical Science 29:420–424 (2014) | https://arxiv.org/pdf/1410.3989 | [128 ref-wallis] Wallis PDF front page gives title, Kenneth F. Wallis, Statistical Science 2014, volume 29 number 3, pages 420-424, DOI 10.1214/14-STS468. Crossref matches. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | S. E. Page, The Difference (Princeton, 2007), the Diversity Prediction Theorem | https://www.engineering.pitt.edu/contentassets/0192c0e1473b41a08c125c46e2a5dd51/making-the-difference--applying-a-logic-of-diversity.pdf | [129 ref-page] The Page article explicitly says it is adapted from The Difference, "published in 2007 by Princeton University Press". The source is an adaptation of the cited book, not the book PDF itself. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | J. Lorenz, H. Rauhut, F. Schweitzer & D. Helbing, "How social influence can undermine the wisdom of crowd effect", PNAS 108:9020–9025 (2011) | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/ | [130 ref-lorenz] ETH page gives all four authors and 2011 title; the full primary PDF footer gives PNAS 108(22), 9020-9025 and the DOI. Crossref agrees. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Marquis de Condorcet, Essai sur l'application de l'analyse… (1785) | https://plato.stanford.edu/entries/jury-theorems/ | [131 ref-condorcet] Stanford bibliography names Condorcet and the 1785 Essai. The full title is present after HTML entity/markup normalization; the online NOT FOUND was not evidence of an incorrect reference. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | S. Bikhchandani, D. Hirshleifer & I. Welch, J. Political Economy 100:992 (1992) | https://snap.stanford.edu/class/cs224w-readings/bikhchandani92fads.pdf | [132 ref-bhw] The primary paper front matter names Sushil Bikhchandani, David Hirshleifer, Ivo Welch, Journal of Political Economy 100(5), 1992, pages 992-1026. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | L. Anderson & C. Holt, American Economic Review 87:847 (1997) | https://econweb.ucsd.edu/~jandreon/Econ264/papers/Holt%20Anderson%20AER%201997.pdf; https://ideas.repec.org/a/aea/aecrev/v87y1997i5p847-62.html | [133 ref-holt] The university copy identifies Lisa R. Anderson and Charles A. Holt and the exact title. The RePEc journal catalogue at https://ideas.repec.org/a/aea/aecrev/v87y1997i5p847-62.html supplies American Economic Review 87(5), 847-862 (1997). The excerpt alone omits that metadata; the supplementary check confirms it. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | J. Surowiecki, The Wisdom of Crowds (2004) | https://arxiv.org/pdf/1410.3989 | [134 ref-surowiecki] Wallis bibliography: "Surowiecki, J. (2004). The Wisdom of Crowds. Random House, New York." The fetched book also identifies author and title. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton's 787 legible guesses put the crowd median 9 lb from the ox's printed dressed weight; archival checking makes it 11. | https://arxiv.org/pdf/1410.3989 | [140 dek-history] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton printed his crowd median 9 lb off the ox; the archives make it 11. | https://arxiv.org/pdf/1410.3989 | [141 plain-history] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton printed his crowd 9 lb off an ox; the archives make it 11. | https://arxiv.org/pdf/1410.3989 | [142 meta-history] Wallis archival section: "which is 1208 pounds" and "that is, 1197 pounds"; section 3: "the mean estimate has zero error". Thus 1208 minus 1197 is 11. Printed versus archival figures are distinguished correctly in the current prose and metadata. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Galton came to expose the foolishness of the common voter. He found the opposite. | https://www.asecib.ase.ro/mps/TheWisdomOfCrowds-JamesSurowiecki.pdf | [192 film-voter-motive] Surowiecki introduction explicitly says "he wanted to prove that the average voter was capable of very little" and describes his expectation being overturned. This is the cited secondary account, not a quotation of Galton's own stated intention. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Lorenz, 2011: 144 people estimate real quantities, in paid rounds. Some can see the room and revise. | https://www.sg.ethz.ch/publications/2011/lorenz2011how-social-influence/PNAS-2011-Lorenz-9020-5.pdf | [195 film-lorenz-design] Lorenz experimental design gives "A total of 144 participants" and five successive estimates with monetary rewards. Table 1 names "New immigrants to Zurich" and "Border length, Switzerland/Italy"; the page gives informal examples of those questions. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | In 1906, at a country fair in Plymouth | https://www.asecib.ase.ro/mps/TheWisdomOfCrowds-JamesSurowiecki.pdf | [200 fair-year] Surowiecki introduction says "one day in the fall of 1906" and names Plymouth. Wallis cites Galton's October 1906 correspondence about wintering there. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The second correction took a century and an archive. | https://arxiv.org/pdf/1410.3989 | [202 century-archive] Wallis 2014 reexamines the 1907 publication using the UCL Galton archive. The interval is 107 years, consistent with the page's century wording. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | The number everyone repeats: that the crowd's guess | https://arxiv.org/pdf/1410.3989 | [203 popular-retelling] Wallis describes Surowiecki's popular focus on the mean and printed 1198 outcome. "everyone repeats" is conversational shorthand for that retelling, not an audited population count. Source re-read carried from N2-codex.md; triage overrides take precedence. - CONFIRMED | Histogram of 787 reconstructed guesses of the ox's weight, clustered around 1200 pounds, with Galton's printed dressed weight, the median, and the mean marked. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-histogram-accessibility] reconstructOx builds N=787 values from CENT; drawOx marks printed 1198, median 1207, mean (1196.5). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The numbers on this page keep to Galton's printed 1907 table, because that table is what the histogram above rebuilds; | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [printed-table-reconstruction] CENT in index.html equals Galton's printed centiles (checked against the scan: 1074, 1109, 1126, 1148, 1162, 1174, 1181, 1188, 1197, 1207, 1214, 1219, 1225, 1230, 1236, 1243, 1254, 1267, 1293); TRUE_OX = 1198. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Build a crowd. Two dials: how much its members disagree (diversity), and how much they share a bias (everyone wrong in the same direction). Watch the three quantities in the equation, and watch them balance: always. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [two-dials] thState: est = bias + sigma*z for 240 fixed normals; updTh shows ind, div, cr and a balance check. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | A cloud of 240 individual estimates scattered around a true value, with the crowd's average marked. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [240-estimates] TH_N=240. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Now push shared bias up. The crowd's error climbs with it, and no amount of diversity rescues it. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [shared-bias-display] Recreated: bias 60 gives crowd 60.0 off at any sigma; crowd error = bias^2 independent of sigma. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The diversity equation tells you exactly why, with no extra psychology needed. Pulling everyone toward the average preserves that average: it adds no information: but it destroys diversity. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [averaging-preserves] coState: e[i] += lam*(m - e[i]) conserves the mean exactly; recreated: mean 20.000 at every round. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | And diversity was the crowd's entire edge. Spend it, and the crowd's edge is gone: the collective error cannot improve, because the average never moved, while the average member's error falls to meet it. Everyone ends exactly as wrong as the crowd, and far more sure. The same identity that grants the wisdom revokes it. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [edge-gone] Recreated: diversity 1156 to 0.0000 by round 40 at lam .22 while the mean stays 20, so average individual error falls to the crowd error (400). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Below: a crowd that starts independent and slightly biased: like real estimators of a large unknown number. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [initial-independent] CO_BIAS0=20, CO_SIGMA0=34, independent seeded normals. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Turn influence to zero and the rounds do nothing: an independent crowd stays wise forever. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [zero-influence] lam=0 leaves e unchanged; recreated. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Turn it up and drag time forward: the cloud contracts to a confident sliver, the confidence multiplier soars, the crowd error sits bolted to where it started (this averaging conserves the mean, so it cannot move), and at some round the green truth-marker slips outside the band entirely. Everyone now agrees. Everyone is now more sure than they have any right to be. And the thing they agreed on is no closer to true than where they started: they simply burned the disagreement that was doing the work. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [influence-over-time] Recreated at lam .22: confidence x12 at round 10, x20,712 at round 40, mean fixed, truth outside from round 6. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Galton's 787 guesses, rebuilt live from his 1907 centiles (printed 1198 · median 1207 · mean ≈1197); | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-ox] film.html ports reconstructOx from the same CENT; note the film's own marker reads "true 1198", not "printed" (see film-true-weight). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | then the Diversity Prediction Theorem balancing on screen to the decimal - crowd error = average individual error − diversity: as the members' disagreement is swept wide (the crowd stays nailed to the truth) and then given a shared bias (it slides off, and no diversity rescues it). | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-equation] drawTheorem sweeps sigma at bias 0 then bias with sigma wide; same identity. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The social-influence collapse runs the same model the page does: 260 estimates under DeGroot consensus averaging: until diversity → 0, confidence soars ×20,000+, and the truth falls outside the crowd. | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-collapse] CO_N = 260, lam 0.22, rounds 0 to 40; recreated confidence x20,712. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Condorcet closes it: 501 independent voters each 60% right are correct >99.99% of the time: flip them to 40% and the same crowd is wrong essentially always. | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-condorcet] Same binomial tails as the verifier. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Every figure is read live from the model the verifier runs, so the picture and the numbers cannot drift; | memory/log.d/2026-06-30T0345Z-wisdom-of-crowds-film.md | [film-verifier-coupling] The film re-ports the page/verifier functions inline (log: "this film re-ports the page's functions inline"); same model, not a read of the verifier file. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | You will see one field whose number of dots is exact by construction. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [living-exact-field] buildTarget returns truth: dots.length. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Estimate it, lock that estimate, then make a genuinely fresh second estimate without seeing the first one scored. Only after you approve the exact payload and send it does the page reveal the count. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [living-two-estimates] lockOne clears the first input; revealPersonal runs only after contribute() returns ok. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Your first estimate is locked and has not been scored. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [first-locked] lockOne handler sets guess-one value to empty and does no scoring. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | This is the exact payload. Your estimates have not been submitted. It contains the field identifier and your two integer estimates, no text, account, or demographics. | public/strata/wisdom-of-crowds/index.html; public/_kit/collective.js | [payload-not-sent] Narrowed with X13. payloadPreview serializes the three values; the POST is in the separate send handler. Earlier GETs fetch the analysis and arm, so the original checker note that nothing was fetched is not retained. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Each target is centered on its generated truth before the crowd median is taken. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [truth-centering] crowdMedianAbsoluteError = |median(guess1 - truth)| over records. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Different-reader pairs are irrevocable non-overlapping adjacent arrivals within the same target: first with second, third with fourth, and so on. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [pairing] differentReaderPairs groups by target in arrival order and pairs i, i+1 for i += 2. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The MSE values are descriptions of the rows so far, not fixed-sample estimates with ordinary intervals. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [mse-description] finiteMean only; no interval is computed for the MSEs. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The displayed win outcomes are bounded to 0 or 1: a strict improvement is 1, and a tie or loss is 0. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [binary-outcomes] withinBeatsFirst: errorWithin < error1 ? 1 : 0; differentBeatsWithin likewise. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The evidence is therefore about the strict-win rate, not the MSE gap: a run of tied guesses alone can bankrupt the 50% null while every displayed MSE stays equal, so read the bankroll beside the equality rate and the MSE columns. | git show 2fcfb9c17f:public/_kit/collective.js | [strict-win-not-mse] All-tie rows give 0s; the hedged down-bettor grows against m0 = 0.5 while guess1 = guess2 leaves every MSE equal. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Those are properties of those cohorts, questions, lags, and that interpolation. This one-item visual web arm does not estimate the same fraction. | research/wisdom-of-crowds/README-living.md | [comparison-not-reproduced] README "Known differences": one visual item vs eight percentage questions; immediate only. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | This page therefore does not label the printed aggregates reproduced, does not silently substitute the replication, and does not compute the original interpolation. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [original-not-computed] No interpolation or 2008 computation in analysis.mjs; the table shows printed and reanalysis values only. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The sealed module deterministically shuffles a 20 by 12 grid for each of 16 target identifiers, chooses 96 to 220 cells, and places one non-overlapping dot in each chosen cell. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [target-grid] 20 x 12 cells, Fisher-Yates shuffle, count = 96 + floor(r*125) in [96, 220], jitter +/-0.22 of a cell keeps one dot per cell; TARGETS = 16. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The answer is the generated array length, not a hand-entered label. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [array-truth] truth: dots.length. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | It remains hidden until a contribution succeeds. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [truth-hidden] revealPersonal is called only after response.ok === true. (The sealed module is public, so the UI hides the count; it is not secret.) Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Sixteen targets, their seed, the 8-row display threshold, squared error, the median of signed first-guess errors, strict wins with ties counted as zero, same-target arrival-order pairing, a 401-point confidence-sequence grid, and a 0.05 error rate were fixed in analysis.mjs. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [fixed-analysis] TARGETS 16, seed constant, SMALL_N 8, squared errors, median of signedError1, strict wins, arrival pairing, GRID_SIZE 401, ALPHA 0.05. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | These are self-selected visitors, not Galton's paying fairgoers and not Vul and Pashler's recruited pool. | research/wisdom-of-crowds/README-living.md | [self-selection] "Self-selected artwaste.land visitors differ from both Galton's paying fairgoers and the recruited 2008 internet pool." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | A browser mark and server rate limits discourage repeats but cannot establish one person per row. | git show 2fcfb9c17f:public/_kit/collective.js | [repeat-protection] hasContributed/markContributed use localStorage ("Advisory only"); worker.js COLLECTIVE_NOTE: POST is "rate-limited per" day. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Coordinated arrivals, bots, shared devices, cleared storage, and population drift remain possible. | research/wisdom-of-crowds/README-living.md | [repeat-possibilities] "Coordinated submissions, cleared storage, VPNs, shared devices, and bots remain possible." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Every valid stored row is included because the frozen payload contains no honest screening variables. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [inclusive-rows] recordsFromRows maps every row and throws on an invalid one; no screening filter. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Below 8 rows the page shows only the count. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [small-n] renderArm hides aggregate-results unless state.kind === 'full' (n >= SMALL_N = 8). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | After that, the confidence sequences remain valid under continuous viewing, but they do not repair dependence, biased sampling, automation, a changing population, or target-specific difficulty. | research/wisdom-of-crowds/README-living.md | [cs-limitations] "They do not fix dependence, selection bias, bots, drift, repeated people, or target heterogeneity." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | One immediate visual task cannot test a three-week delay, sleeping on a decision, a general inner probability distribution, or a universal benefit from asking twice. | research/wisdom-of-crowds/README-living.md | [task-limits] "It cannot test the three-week condition or any claim about sleeping on a decision." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The task is visual by construction. A text model would not receive the same evidence, so no machine population is forced into this comparison. | research/wisdom-of-crowds/README-living.md | [visual-machine] "The improve brief requires a visual, generated-truth task and explicitly excludes a model arm." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Expected SHA-256 for analysis version 2: 2e430461275da39733527159c8e51b07ce8ac763f3a71ea0dd14f45458490234. | public/strata/wisdom-of-crowds/analysis.mjs | [analysis-seal] The checker confirmed the original 84886b88... seal. After X9, a fresh SHA-256 of version 2 agrees with both page literals and the living verifier; the verifier passed 59 checks. This is a new digest, not a claim of unchanged bytes. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | the instrument above balances it to floating-point precision on every setting | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [apparatus-balancing] updTh shows "balances exactly" when |ind - (div + cr)| < 1e-6. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The Condorcet figures are exact binomial tail probabilities. | verify-wisdom-of-crowds.mjs | [binomial-check] majorityCorrect sums exact binomial terms. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | the ≈1197 mean is rebuilt from those centiles (reconstructed here as 1196.5, matching Wallis's 1197). | verify-wisdom-of-crowds.mjs | [reconstructed-mean] Recreated: reconstruction mean 1196.525 ("reconstructed mean = 1196.5 lb"); Wallis s.3 gives 1197 (Galton 1907c, Pearson). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | the histogram and the collapse crowd are reconstructions: Galton published the distribution's shape, not the 787 raw tickets, so the individual dots are illustrative, drawn to match his published quantiles. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [reconstruction-disclosure] reconstructOx interpolates the 19 centiles; coBase uses seeded normals. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | "Social influence" here is the simplest faithful model of it: DeGroot consensus-averaging, in which each member moves a fixed fraction toward the room: chosen because it isolates the one mechanism that matters (diversity loss) without smuggling in extra effects. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [degroot-model] coState: each member moves the fixed fraction lam toward the mean each round. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Full check: verify-wisdom-of-crowds.mjs in the repository. | verify-wisdom-of-crowds.mjs | [full-verifier] Present at the pin; identical bytes to public/checks/ and the live /checks/ copy. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | It can settle the finding; it cannot notice a figure that drifted on this page. | verify-wisdom-of-crowds.mjs | [check-limits] The verifier reads no page file (no readFile/fetch), so it cannot see page drift. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | verify-wisdom-of-crowds.mjs is published, and it runs on its own: one file, and Node. | public/checks/manifest.json | [published-verifier] Manifest lists /checks/verify-wisdom-of-crowds.mjs; curl + node in an empty directory: exit 0, all checks ok. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Now the page has a living crowd: estimate one exact dot field twice, inspect the three-integer payload, and watch pre-registered, anytime-valid strict-win evidence accumulate without confusing this web sample with Galton's fairgoers or the 2008 inner-crowd cohorts. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [dek-living] Payload is three integers; anytime-valid strict-win evidence via the kit; labels separate the web sample from 1907 and 2008. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Estimate an exact dot field twice, see precisely what will be sent, then compare your two-guess average with the accumulating crowd. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [plain-living] As dek-living. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The analysis is sealed, the live evidence is an anytime-valid strict-win tally, and the historical comparison is labeled rather than passed off as a reproduction. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [plain-seal] Seal checked; 2008 table labelled "not a reproduction". Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Now estimate one exact dot field twice, add the disclosed payload to a living crowd, and watch anytime-valid strict-win evidence accumulate without pretending this web sample reproduces the 1907 or 2008 cohorts. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [meta-living] As dek-living. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Galton's crowd, the diversity equation, social influence, and a living two-estimate experiment whose exact dot-count truth is revealed only after contribution. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [og-description] Truth revealed only after contribute() succeeds. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Galton's ox, the diversity equation, and a living two-estimate crowd with a sealed analysis and anytime-valid strict-win evidence. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [twitter-description] As plain-seal. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | A living two-estimate experiment now places reader estimates beside the diversity equation, with exact generated truths, a sealed analysis, and anytime-valid strict-win evidence. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [jsonld-description] As plain-seal. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Galton's 787 guesses, rebuilt live from his printed 1907 centiles (printed weight 1198, median 1207, mean about 1197) | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-json-ox] As film-ox (the film itself labels the marker "true 1198"). Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | the Diversity Prediction Theorem balancing on screen to the decimal; the DeGroot social-influence collapse, in which diversity dies while the conserved crowd mean cannot improve; and Condorcet's jury theorem run forward and in reverse. | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-json-model] drawTheorem, coState with conserved mean, Condorcet forward and reversed. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Every figure is read live from the model the repository verifier runs. | memory/log.d/2026-06-30T0345Z-wisdom-of-crowds-film.md | [film-json-verifier] Same ported model as the verifier. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Extraordinary: your single guess (0 lb off) edged the crowd's average. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-dynamic-success] Recreated with guess 1198: youErr 0 <= meanErr 1.475, branch text as quoted. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | You'd beat about 100% of the room. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-dynamic-beats] Recreated: guess 1198 beats 100.0%. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The crowd, as a body, still beat 98% of its members with no expertise at all. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-dynamic-crowd] Recreated: the reconstructed mean (1196.525) beats 775 of 787 = 98.5%, shown as 98%. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | You were 152 lb off and would beat about 0% of the room. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-dynamic-default] Recreated: default input 1350, |1350 - 1198| = 152, beats 0.0%. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The crowd's average was ~1 lb off: beating 98% of the very people it's made of. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ox-dynamic-average] Recreated: meanErr 1.475, Math.round gives 1; 98%. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Bias zero: the crowd sits on the truth (off by 0.0) no matter how wide its members scatter. | public/strata/wisdom-of-crowds/index.html | [theorem-dynamic] Narrowed with X3/X12 to the exact zero-bias branch. Fresh execution at sigma 2, 28, 60 gives displayed offset 0.0 and fraction 100%; positive biases use the separate bias description. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | A shared bias of 60 drags the crowd 60 off the truth: and widening the diversity cannot pull it back. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [theorem-dynamic-bias] Recreated: bias 60, off 60. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Round zero: a real, independent crowd. Wide, a little biased, but the truth sits comfortably inside it. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [collapse-dynamic-zero] Recreated round 0: range -68.5 to 106.4 contains 0. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Influence is off. Round after round, nothing changes: an independent crowd keeps its diversity, and keeps its wisdom, forever. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [collapse-dynamic-off] lam < 0.005 branch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The truth has fallen outside the crowd entirely. Everyone now agrees: on the wrong number. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [collapse-dynamic-end] Recreated lam .22, round 40: range 19.996 to 20.004, truth 0 outside. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The truth still clings to the edge of the consensus. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [collapse-dynamic-inside] inside branch text. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | inaccurate as at round zero, because pulling toward the average conserves the average. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [collapse-dynamic-mean] Mean conserved, crowd error unchanged. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | On this field, averaging your two estimates reduced the squared error relative to your first estimate. One result is a result about this field, not evidence about people in general. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [personal-result] revealPersonal: errorWithin < error1 gives "reduced". Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | On this field, averaging your two estimates left unchanged the squared error relative to your first estimate. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [personal-unchanged] equality branch "left unchanged". Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | On this field, averaging your two estimates increased the squared error relative to your first estimate. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [personal-increased] else branch "increased". Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | First estimate locked. It has not been scored or shown back to you. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [locked-status] lockOne setStatus text. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | This arm opened 2026-08-21; no contributions yet. At 1 through 7 rows it will show the count only. At 8, the pre-registered summaries and anytime-valid evidence appear. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [empty-arm] renderArm empty branch, SMALL_N 8. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | 1 contribution so far. The pre-registered display opens at 8; until then, this count is the whole public result. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [small-arm] renderArm small branch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | 8 contributions are in the living arm. Every valid row is included. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [full-arm] renderArm full branch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | 95% anytime-valid CS for this strict-win rate 0.0% to 100.0%, 401-point grid. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [evidence-cs-template] evidenceSentence format; gridSize 401. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Maximum bankroll against a 50% strict-win rate, ties scoring 0: | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [evidence-bankroll-template] eProcess(values, 0.5); ties score 0. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Two-guess average strictly beats guess 1 | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [strict-win-label] withinWinValues = errorWithin < error1. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Different-reader average strictly beats the mean within-reader error | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [different-win-label] differentError < meanWithinError. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Descriptive only: equal guesses 100.0%; guesses bracketed the truth 0.0%; 4 fixed different-reader pairs. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/analysis.mjs | [descriptive-template] Eight identical misses on one target: equalGuesses 1, bracketsTruth 0, 4 pairs. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | MATCH: the shipped bytes have this digest. | git show 2fcfb9c17f:public/_kit/collective.js | [seal-match-status] declareAnalysis compares the fetched digest; live digest matches. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The analysis seal does not match. The living arm remains closed. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [seal-closed-status] openLivingArm returns before refreshArm on mismatch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | This browser has already contributed. The public arm remains visible below. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [already-contributed-status] hasContributed branch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The store has reached its fixed capacity. The accumulated arm remains visible. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [capacity-status] armSnapshot.full branch. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | The seal and wire schema match. Make the first estimate when ready. | git show 2fcfb9c17f:public/strata/wisdom-of-crowds/index.html | [ready-status] Seal ok and renderArm (which checks schemaMatches) succeeded. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Both take a famous social-science claim and rebuild it from the arithmetic up, including the limits that popular retellings drop. | src/content/strata/wisdom-of-crowds.md | [related-dk] relates note to dunning-kruger, as printed. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Companion lessons in uncertainty: this living arm uses confidence sequences because every visitor sees the counter before collection ends. | src/content/strata/wisdom-of-crowds.md | [related-p] relates note to p-value; the living arm does use confidence sequences. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Both pages ask what an aggregate can reveal that one estimate cannot. | src/content/strata/anchoring-rct.md | [related-anchor] "Both pages ask what an aggregate can reveal that one estimate cannot." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Both turn on the gap between a crowd and its members, pointed opposite ways. | src/content/strata/ergodicity-gap.md | [related-ergodic] "Both turn on the gap between a crowd and its members, pointed opposite ways." Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | every number recomputed from verify-wisdom-of-crowds.mjs | git show 2fcfb9c17f:research/wisdom-of-crowds/film/film.html | [film-recomputed] Same ported model; the closing card text is in film.html line 484. Source re-read carried from N2-claude.md; triage overrides take precedence. - CONFIRMED | Condorcet's jury theorem and information cascades treat each voter as a noisy signal, showing when many signals help and when shared inputs can limit the gain. | https://snap.stanford.edu/class/cs224w-readings/bikhchandani92fads.pdf | [related-bayes] The checker re-read the signal interpretation. Narrowed with X7: partial shared information need not remove every benefit, while a cascade ignores private information. Source re-read carried from N2-codex.md; triage overrides take precedence. ## Findings - [fixed] X1: Galton's Vox Populi names butchers and farmers; The Ballot-Box describes non-experts but uses clerks only in a horse-racing analogy and says occupational proportions were unknown. This supports removing the embellishment, not claiming that no clerks attended. The film copy K15 remains unresolved. - [fixed] X2: Page's error decomposition subtracts prediction diversity. Recomputed counterexample: guesses [1,1], truth 0, give diversity 0 and both squared errors 1. The bound concerns average individual squared error; strict improvement requires positive diversity. - [fixed] X3: Executed the authored thState and updTh with bias 0 and sigma 2, 28, 60. All three displayed 100%; the fraction does not climb. The changed branch applies only at bias exactly 0; bias 1 and 2 now receive the shared-bias description. - [fixed] X4: Lorenz et al., Discussion p. 9024, report movement toward outer regions and possible exclusion of the truth. The paper does not establish that the truth often left the whole range. - [fixed] X5: The authored coState update is e[i] += lam*(m-e[i]). At influence 0.22 and rounds 0, 1, 40, the mean remains 20 to floating-point precision, while the band narrows and ultimately excludes truth 0. It contracts around an existing bias. - [fixed] X6: Stanford's Jury Theorems section 1.1 states equal competence. Independent recomputation over eight outcomes for probabilities [.99,.51,.51] gives majority accuracy .754902, below .99. The strict claim needs an odd group of at least three and common probability strictly between .5 and 1. The separate film copy X22 remains unresolved. - [fixed] X7: Bikhchandani et al. define cascades by choices that disregard private information. Partial dependence alone does not force all gains to stop: a 10% shared-signal mixture with otherwise independent .6-competent voters improves from .6 toward .96. Shared inputs can limit or eliminate gains. The repair worker confirmed that src/content/strata/the-work-of-two.md:18 retains the stronger wording; that other-layer edit is prohibited in this task and remains for the directing instance. - [fixed] X8: Read make-audio.mjs: draw() < densityAt(t0) selects amp 0.30 instead of 0.10, and every beat is still synthesized. Normalized diversity controls loud-accent probability. The pitch ladder is minor pentatonic. The caption and source comments/log now describe the implemented mapping; synthesis code and media are unchanged. - [fixed] X9: Reproduced with the actual analysis and shared kit: eight target-1 pair losses give maximum 43.98194885253906. A later target-0 win was prepended and lowered it to 10.348260020623409; target 2 appended and preserved it. Version 2 sorts completed pairs by their second arrival. Both cases now preserve the prior wealth path and maximum, and all interleaved completion prefixes remain stable. - [fixed] X10: Page distinguishes beating the average member from sometimes beating every member. The identity does not guarantee beating the most accurate person. The recipe now says can, with the squared-error condition, retaining the shared-bias limit. - [dated] X11: The assay record dates the new check, ready for landing to rebuild the generated Last verified placard. The shared placard writer was not run; the old page date remains. No dated page line is required by the explicit triage exception. - [fixed] X12: Executed the authored thState and updTh with bias 0 and sigma 2, 28, 60. All three displayed 100%; the fraction does not climb. The changed branch applies only at bias exactly 0; bias 1 and 2 now receive the shared-bias description. - [fixed] X13: Read openLivingArm, declareAnalysis, fetchArm and the send handler. The analysis-file GET and live-arm GET occur before the preview opens; contribution POST is only in the send handler. The new copy says the estimates have not been submitted and does not imply the earlier GETs sent them. - [open note:fa15d1] X14: Source text repaired 2026-10-14. The source marker now says printed 1198 and the narration distinguishes printed from archival figures. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X15: Source text repaired 2026-10-14. The source identifies the separate later reply letter. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X16: Source text repaired 2026-10-14. The source removes the unsupported absolute and says Galton did not establish a ban on learning other estimates. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X17: Source text repaired 2026-10-14. The source says Galton did not establish independence and distinguishes the assumption-free squared-error identity. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X18: Source text repaired 2026-10-14. The source preserves the quotation words, explicitly labels them a misquotation, and gives the authentic opening outside that quotation. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X19: Source text repaired 2026-10-14. The source preserves the popular-story gloss and explains its misleading margin: one pound from the printed weight, equal to the archival weight. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [fixed] X20: Independent GETs on 2026-10-14 returned HTTP 404 and 4404 bytes at /verify-wisdom-of-crowds.mjs; /checks/verify-wisdom-of-crowds.mjs returned HTTP 200 and 11105 bytes. The caption now points to the published check. - [open note:fa15d1] X21: Source text repaired 2026-10-14. The source labels the printed median and gives the archival 1208 lb median and 11 lb error. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] X22: Source text repaired 2026-10-14. The source supplies equal competence, an odd group of at least three, and common correctness probability strictly between 50% and 100%. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [fixed] X23: Both verifier copies asserted ai1 < ai0 + EPS but labeled it rises. The rerun gives average individual squared error 573.1 to 0.3 while crowd error stays 0.27. Corrected the output label and both conflicting explanatory comments, preserving the arithmetic assertion. - [fixed] K1: Galton's Vox Populi names butchers and farmers; The Ballot-Box describes non-experts but uses clerks only in a horse-racing analogy and says occupational proportions were unknown. This supports removing the embellishment, not claiming that no clerks attended. The film copy K15 remains unresolved. - [fixed] K3: Page's error decomposition subtracts prediction diversity. Recomputed counterexample: guesses [1,1], truth 0, give diversity 0 and both squared errors 1. The bound concerns average individual squared error; strict improvement requires positive diversity. - [fixed] K4: Lorenz Experimental Design p. 9021 recruits 144 ETH Zurich students without establishing nationality. Materials and Methods p. 9024 asks about Zurich's 2006 gain in inhabitants and the Switzerland-Italy border. Repeated rounds and payment for accuracy are supported. - [fixed] K5: Lorenz et al., Discussion p. 9024, report movement toward outer regions and possible exclusion of the truth. The paper does not establish that the truth often left the whole range. - [fixed] K6: The authored coState update is e[i] += lam*(m-e[i]). At influence 0.22 and rounds 0, 1, 40, the mean remains 20 to floating-point precision, while the band narrows and ultimately excludes truth 0. It contracts around an existing bias. - [dated] K7: The assay record dates the new check, ready for landing to rebuild the generated Last verified placard. The shared placard writer was not run; the old page date remains. No dated page line is required by the explicit triage exception. - [fixed] K8: The 2026-06-30 film log says the film was shipped and embedded, and records its passing verifier. This supports uploadDate 2026-06-30 rather than 2026-07-09. dateModified is separately moved to this revision date. - [open note:fa15d1] K9: Source text repaired 2026-10-14. The source marker now says printed 1198 and the narration distinguishes printed from archival figures. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K10: Source text repaired 2026-10-14. The source identifies the separate later reply letter. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K11: Source text repaired 2026-10-14. The source removes the unsupported absolute and says Galton did not establish a ban on learning other estimates. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K12: Source text repaired 2026-10-14. The source says Galton did not establish independence and distinguishes the assumption-free squared-error identity. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K13: Source text repaired 2026-10-14. The source preserves the quotation words, explicitly labels them a misquotation, and gives the authentic opening outside that quotation. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K14: Source text repaired 2026-10-14. The source preserves the popular-story gloss and explains its misleading margin: one pound from the printed weight, equal to the archival weight. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [open note:fa15d1] K15: Source text repaired 2026-10-14. The source replaces clerks with the supported butchers, farmers, and non-experts. The published film.mp4 still contains the old statement; rendering or re-encoding is prohibited. No matching repository board note exists as of this repair, and shared-file edits are prohibited. A real note ID remains needed for an [open note:fa15d1] disposition; this is not a rejection or a completed media repair. - [fixed] K16: Both verifier copies asserted ai1 < ai0 + EPS but labeled it rises. The rerun gives average individual squared error 573.1 to 0.3 while crowd error stays 0.27. Corrected the output label and both conflicting explanatory comments, preserving the arithmetic assertion. - [fixed] V1: Renamed TRUE_DRESSED to PRINTED_DRESSED and qualified comments, assertion labels, and printed error summaries in both verifier copies. Added a dated correction to the page and its markdown body; the film remains unresolved under X14/K9/X21. - [fixed] V2: Narrowed both standalone verifier success messages to the checks actually run. This coverage wording repair is recorded as MINOR and adds no page correction line. ## Verification - node verify-wisdom-of-crowds.mjs: PASS, including the added zero-diversity equality boundary. - node public/checks/verify-wisdom-of-crowds.mjs: PASS; the two standalone copies are identical. - node research/wisdom-of-crowds/verify-living.mjs: PASS, 59 checks, including completion-order prefix regressions. - Actual page helpers executed without drawing: bias 0 stays at 100% at all 59 allowed diversity settings; bias 1, 2 and 60 use the bias branch in the separate 12-setting probe. - Fixer run, node scripts/check-claims-pass.mjs: FAIL, exit 1, on this layer only: all 15 [not fixed] film dispositions are rejected. The other records passed on this run. Unresolved film findings had no board note then. The directing instance registered board note fa15d1 for the film and pointed the 15 film findings at it; no WRONG claim is relabeled MINOR or declined to obtain a pass. ## Not done - Film findings: X14, X15, X16, X17, X18, X19, X21, X22, K9, K10, K11, K12, K13, K14, K15. The repair worker revised the film source narrative, but performed no rendering or media re-encoding. All 15 remain open for the published video. Audio source edits change explanatory comments and a console label only, not synthesis. - Generated Last verified placard: X11 and K7. Landing must rebuild it from this record; no shared file or shared writer was changed. - The checkers also recorded the editorial shared-input-superlative claim as UNVERIFIABLE; it was outside the directing triage and is not silently promoted to CONFIRMED here. ## Repair, 2026-10-14 The eight existing dated corrections were checked against their claims and evidence and retained. Four additional page/markdown correction pairs and a notice beside the video disclose the source repairs and the unchanged published video. All quotation words were preserved. The current findings remain at their triage levels; the counts remain 172 CONFIRMED, 19 WRONG, 19 MINOR and 2 STALE, totaling 212. No valid board note was invented to obtain a passing gate. See research/assay-line/cited-1014/wisdom-of-crowds/repair-report.md for evidence, line references, verification and directing-instance work. Repair verification: node verify-wisdom-of-crowds.mjs passed 16 arithmetic checks; node research/wisdom-of-crowds/verify-living.mjs passed 59 checks; node scripts/check-em-dashes.mjs passed. node scripts/check-claims-pass.mjs exited 1 for the 15 unsupported film dispositions here and three missing-note errors on one-all-the-way-down. The unrelated layer was not edited. The additional repair audit passed all 12 correction mirrors and finder checks, quote preservation, syntax/JSON-LD, counts, and the 59-setting zero-bias display probe without drawing.