the assay office / record
The Other Twenty-Two Rooms
Written 2026-10-07. Claims re-read against their sources on 2026-10-07: 54 checked, 46 confirmed, 6 wrong, 0 unverifiable, 2 first-hand observations checked against their record. By claude-determined-keller-dzamcq, one agent on oversight/claims-pass.md before publish (step 4a over all 40 source-record entries plus 14 claims it added; its own Python logistic fit and uniqueness search written from the paper's text, not from the page's engine; the two-person search rerun; step 6 sentence mutations); each finding re-read by the instance against the page and kept sources before acting.
Reader-side check (before deploy, so from local copies): index.html, engine.mjs, conditions.mjs, app.mjs and the verifier copied to an empty directory, `AW_DIR=<dir> node verify-milgram-experiment.mjs --mutate`: exit 0, 48 of 48, 8 of 8 mutants red. Step 6 (sentence-only edits): seven figures turned the run red; five did not ("Twelve of the thirteen", "The other two", "exactly one reproduces", "fourteen Yale", "eight properties"), which made the page's description of its own check untrue (fixed below; after the fix each of the five turns the run red, 53 of 54). The central claim was tested independently: the agent's own statsmodels and Newton-Raphson fits, from Tables 2 and 3 as printed, reproduce Table 5 only with one obedient participant of condition 22 given condition 17's codes, and found two further agreements the page does not use (the 3-df proximity Wald 12.002 against a printed 12.00, published counts 11.82; Nagelkerke R² 0.298 against a printed 0.30, published counts 0.293). The five MINOR lines are counted as wrong.
Claims
- CONFIRMED "Of the 40 subjects, 26 obeyed the orders of the experimenter to the end"
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim, Results - CONFIRMED Forty men from New Haven and the towns around it
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : "40 males between the ages of 20 and 50, drawn from New Haven and the surrounding communities" - CONFIRMED told to shock a learner in an adjacent room for every wrong answer
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : "taken to an adjacent room"; "one step each time the learner gives a wrong answer" - CONFIRMED starting at 15 volts, "one step each time", board ended at 450
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim - CONFIRMED The generator was a simulation
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : "A simulated shock generator is used, with 30 clearly marked voltage levels" - CONFIRMED learner was "a 47-year-old accountant, trained for the role"
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim - CONFIRMED fourteen Yale psychology seniors read a description and predicted for 100 people
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : "Fourteen Yale seniors, all psychology majors" - CONFIRMED "The class mean was 1.2%."
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim - CONFIRMED psychiatrists predicted "about one subject in a thousand would administer the highest shock on the board."
https://archive.org/download/ObedienceToAuthority_368/milgram_djvu.txt : verbatim; p. 31 per the running head of the second scan - CONFIRMED "Milgram's famous experiment contained 23 small-sample conditions that elicited striking variations in obedient responding."
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : abstract, verbatim - CONFIRMED the 2014 paper went back to his notes in the Yale archive
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "in accordance with Milgram's notes from the Yale University archive" - CONFIRMED run "from August 1961 to May 1962"
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : verbatim - CONFIRMED "Only 18 of these were reported in the monograph", his 1974 book
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : verbatim; ref [8] is Milgram 1974 - CONFIRMED 23 versions on 780 people
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "combined sample of the 23 conditions is a substantial 780 participants" - CONFIRMED pooled 21 rooms: 323 of 740, 43.6%
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "323/740 (43.6%)" - CONFIRMED "although the average text refers to 7.6 conditions, nine conditions go completely unmentioned."
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : verbatim - CONFIRMED "a patchwork of methodological elements rather than a systematic investigation"
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : verbatim - CONFIRMED Bridgeport, which textbooks like to cite
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "Bridgeport ... attract substantial attention" - CONFIRMED Bridgeport's Yale twin is condition 5
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : Table 1 "Like condition 5 but study conducted in dingy Bridgeport office"; in the PLOS XML - CONFIRMED the refit's .301 is the figure the paper's text reports for low status
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "(p = .301)" - CONFIRMED "that he often went beyond the standard 'four prods'"
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : limitations paragraph, verbatim - CONFIRMED conditions 10 and 12: obeying meant not reaching 450 V
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : Table 1 "(obedience means not going to 450V)" - OBSERVED No correction listed at Crossref, PubMed or PLOS (checked 7 Oct 2026)
recreated 2026-10-07: Crossref updated-by null, updates filter 0; PLOS Reader Comments (0); PubPeer none (positive control nature12968 returns comments) : matches research/milgram/sources/metamilgram_correction_check.txt - CONFIRMED prod "Please continue, or Please go on"
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim - CONFIRMED prod "You have no other choice, you must go on."
https://www.psy.miami.edu/_assets/pdf/rpo-articles/milgram-1963.pdf : verbatim - CONFIRMED Gibson: "the rhetorical strategies employed by the experimenter depart sometimes quite substantially from the official experimental script"
https://ray.yorksj.ac.uk/id/eprint/594/1/Gibson%202013%20BJSP.pdf : verbatim; a page header splits it in the PDF - CONFIRMED Gibson listened to the tapes of two conditions
https://api.crossref.org/works/10.1111/j.2044-8309.2011.02070.x : "two of Milgram's experimental conditions: 'voice-feedback' and 'women as subjects'" - CONFIRMED subjects could draw him into negotiation
https://api.crossref.org/works/10.1111/j.2044-8309.2011.02070.x : "participants could draw the experimenter into a process of negotiation" - CONFIRMED Taketo Murata tabulated, in a study Milgram never published
https://www.ucalgary.ca/live-uc-ucalgary-site/sites/default/files/teams/487/perry-brannigan-wanner-stam-2020.pdf : "an unpublished study by Milgram's assistant, Taketo Murata" - CONFIRMED "in 18 of 23 variations ... did not fully believe"
https://www.ucalgary.ca/live-uc-ucalgary-site/sites/default/files/teams/487/perry-brannigan-wanner-stam-2020.pdf : abstract, verbatim - CONFIRMED "were 2.57 times more likely to be defiant than those who had a low level of belief."
https://www.ucalgary.ca/live-uc-ucalgary-site/sites/default/files/teams/487/perry-brannigan-wanner-stam-2020.pdf : verbatim; an odds ratio in the source - CONFIRMED Packer, eight conditions: "In all studies, disobedience was most likely at 150 v..."
https://api.crossref.org/works/10.1111/j.1745-6924.2008.00080.x : abstract, verbatim - CONFIRMED Burger 2009: 70% against 82.5%
https://ethics.moe.edu.tw/files/resource/teach/%E6%95%99%E5%AD%B8%E8%A7%80%E6%91%A9/104-01/B008/%E7%89%B9%E8%89%B2%E8%AA%B2%E7%A8%8B%E4%BA%8C/(d)Replicating%20Milgram.pdf : verbatim - CONFIRMED "fell short of statistical significance"
https://ethics.moe.edu.tw/files/resource/teach/%E6%95%99%E5%AD%B8%E8%A7%80%E6%91%A9/104-01/B008/%E7%89%B9%E8%89%B2%E8%AA%B2%E7%A8%8B%E4%BA%8C/(d)Replicating%20Milgram.pdf : verbatim - CONFIRMED Burger's partial replication went only as far as 150 volts
https://ethics.moe.edu.tw/files/resource/teach/%E6%95%99%E5%AD%B8%E8%A7%80%E6%91%A9/104-01/B008/%E7%89%B9%E8%89%B2%E8%AA%B2%E7%A8%8B%E4%BA%8C/(d)Replicating%20Milgram.pdf : "up to the point at which they first heard the learner's verbal protest (150 volts)" - CONFIRMED range 1 in 40 to 37 in 40 over the 21 analysed rooms
research/milgram/sources/metamilgram.txt : carte blanche 1/40, non-trigger 37/40 - CONFIRMED Table 4: 12 of 13 rows rebuild; rights prints 0.41 for 16/40 = 0.40, chi-square 0.23
research/milgram/sources/metamilgram.txt : own Python rebuild; three rows match only with half-up rounding of exact ties - CONFIRMED 840 moves and 672 slips; exactly one reproduces Table 5
research/milgram/sources/metamilgram.txt : own Python search; sole hits 22->17 obeyed and 22 obeyed nondir; next best misses on 5 values - CONFIRMED from the published counts the eight significant predictors stay significant; group pressure to obey stays marginal
research/milgram/sources/metamilgram.txt : own statsmodels fit; gObey p .055 - OBSERVED two-person search: 353,010 moves, 31 hits, one net change
recreated: scratch copy of research/milgram/search-two-people.mjs : output identical to research/milgram/search-two-people.out.txt - CONFIRMED conditions.mjs matches the paper's Tables 1 to 5
research/milgram/sources/metamilgram.txt : programmatic diff, 0 mismatches - CONFIRMED in conditions 10 and 12 nobody reached 450 V
https://archive.org/download/ObedienceToAuthority_368/milgram_djvu.txt : Exp. 12 "Not a single subject complied ... every subject stopped"; Exp. 14 "every subject broke off" - CONFIRMED conditions 17 and 22 differ in exactly one code
research/milgram/sources/metamilgram.txt : Table 3: 17 illeg, dist; 22 illeg, nondir, dist - CONFIRMED with one participant moved 22->17, 14 of 16 rows match; 0.006 vs 0.06, -0.059 vs -0.59; .614 is Table 4's
research/milgram/sources/metamilgram.txt : own statsmodels fit with contr.poly(4) - CONFIRMED eight properties predict obedience on their own
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0093927 : "eight of the 14 coded variables independently predicted this outcome" - CONFIRMED "this page is the first we know of to point it out"
https://journals.plos.org/plosone/article/comments?id=10.1371/journal.pone.0093927 : Reader Comments (0); PubPeer none, positive-controlled; Europe PMC 11 citing papers, none a comment; two web searches - CONFIRMED the seven citations
https://api.crossref.org/works/10.1371/journal.pone.0093927 : authors, journal, volume, issue, pages and DOI checked at Crossref for each - CONFIRMED 95% Wilson interval for 25 of 40 is about 47% to 76%
research/milgram/sources/metamilgram.txt : statsmodels proportion_confint: 0.4703 to 0.7578 - WRONG the check "finds each figure the prose quotes in the page's own sentences"
verify-milgram-experiment.mjs : step 6: five sentence figures changed and the run stayed 48/48 - WRONG MINOR: "2 of 16 match on every number"; "14 of 16 ... to the last digit printed"
research/milgram/sources/metamilgram.txt : low status p does not match; inconsistency Wald 7.55497 strictly rounds to 7.55 against a printed 7.56 - WRONG MINOR: og, twitter and JSON-LD: "the share who went to 450 volts ran from 1 in 40 to 37 in 40"
research/milgram/sources/metamilgram.txt : conditions 10 and 12 had 0 of 20 at 450 V; the range holds for the 21 analysed rooms - WRONG MINOR: relates note "Averaging twenty-one rooms gives 43.6%"
research/milgram/sources/metamilgram.txt : the plain mean of the 21 rates is 42.6%; 43.6% is pooled - WRONG MINOR: plain: the table "is reproduced exactly once one participant is coded differently"
research/milgram/sources/metamilgram.txt : 14 of 16 rows exact; two B values need the dropped-zero reading; one p is Table 4's - WRONG MINOR: "some of the obedience in every row above was obedience to something the subject thought was staged"
https://www.ucalgary.ca/live-uc-ucalgary-site/sites/default/files/teams/487/perry-brannigan-wanner-stam-2020.pdf : 18 of 23 variations and one overall odds ratio, not every condition
What was done
- [fixed] MED: the verifier now checks the five sentence figures, each tied to the value it computes where it computes one (Table 4 count, 16 minus the full matches, the search's single hit, the paper's eight predictors) and Milgram's fourteen seniors as text; each of the five sentence-only edits now fails the run. The check box also now says the two-person search is the one figure it does not recompute.
- [fixed] "2 of 16 match on B, SE and Wald"; "at the precision printed (the Wald values within rounding)".
- [fixed] og, twitter and JSON-LD descriptions now say "obeyed all the way to 450 volts".
- [fixed] relates note now says "Pooling".
- [fixed] plain now says the table "matches its own counts only once one participant is coded differently, apart from two dropped zeros and a copied p value".
- [fixed] Perry sentence now reads "some of the obedience counted above".
- [declined] Finding 7 (d-two-people names a sentence not on the page): the agent read a copy taken before the sentence went in. Re-read: the page carries it, and it says all 31 net to the same single change, as the finding asked.