{ "claims": [ { "claim": "On the pages of sitting U.S. senators, negative facts are removed faster than positive facts (pooled Studies 1, 2 and 4: hazard ratio $1.59$, $p=0.001$). The paper describes the pattern as consistent across studies and as evidence of an editorial bias toward positivity.", "claim_id": "P041-H1", "related_corrections": [ "P041-E003", "P041-E012", "P041-N011" ], "tied_to": "S6 Table A column 1 ($0.629$, standard error $0.091$); S6 Tables B, D, E; Fig 1; S11 Tables A, B" }, { "claim": "Negative facts are $36\\%$ more likely than positive facts to be removed within 12 hours and $29\\%$ more likely within 3 days.", "claim_id": "P041-H2", "related_corrections": [ "P041-E001", "P041-E002" ], "tied_to": "Abstract; replication do-file 'Main findings' block (lines 30-38); no table in the paper" }, { "claim": "Citations substantially lengthen survival: uncited facts are $64\\%$ more likely than cited facts to be removed within four hours (reported hazard ratio $2.33$, $p<0.001$).", "claim_id": "P041-H3", "related_corrections": [ "P041-E001", "P041-E003", "P041-E010", "P041-E011" ], "tied_to": "S6 Table A column 2 ($0.429$, standard error $0.069$); S6 Tables B, D, E; S11; Fig 1" }, { "claim": "The positivity bias remains among cited facts; the reported tests found no statistically detectable interaction between valence and citation.", "claim_id": "P041-H4", "related_corrections": [], "tied_to": "S7 Tables A, B; Study 2 column of S6 Table D" }, { "claim": "The paper reports no editorial bias among retired and deceased senators in Study 3 and says that, if anything, positive facts were removed more quickly. It suggests that the bias is limited to active politicians and is not systemic to Wikipedia.", "claim_id": "P041-H5", "related_corrections": [ "P041-E005", "P041-E006" ], "tied_to": "S6 Table C; Fig 1 third panel; S11 Study 3 columns" }, { "claim": "Facing reelection in 2014 moderates the effect of citations (interaction hazard ratio $0.32$, $p<0.001$), suggesting that the citation policy is applied most when stakes are high. The paper reports no systematic evidence that other covariates moderate either effect.", "claim_id": "P041-H6", "related_corrections": [ "P041-E008", "P041-E003", "P041-E004", "P041-E009", "P041-N005" ], "tied_to": "S8; S9 Tables A, B; S10" }, { "claim": "Independent Mechanical Turk raters saw the positive facts as more positive and more favorable to the senator than the negative facts, and as equally relevant (means $5.0$ and $5.0$ on a 7-point scale, standard error $0.04$ each).", "claim_id": "P041-H7", "related_corrections": [ "P041-N015", "P041-E018", "P041-E013", "P041-E014" ], "tied_to": "Methods p. 3; S12 Table A, Figures A-C" }, { "claim": "Including the partially aborted Study 5 does not change the results.", "claim_id": "P041-H8", "related_corrections": [], "tied_to": "Methods p. 3; S5; Study 5 columns of S6 Tables D, E; S7; S9; S10" }, { "claim": "The paper states that true, relevant and cited information about active senators was selectively removed from Wikipedia pages in the days and months leading up to the 2014 midterm elections.", "claim_id": "P041-H9", "related_corrections": [ "P041-E018", "P041-N015" ], "tied_to": "Discussion p. 5; fact-validation description p. 3; Fig 1; S6; S12 Figures A-C and Table A" }, { "claim": "Repeated editorial biases may accumulate and distort the political information available on crowd-sourced sites.", "claim_id": "P041-H10", "related_corrections": [], "tied_to": "Discussion pp. 5-6; Fig 1 and the repeated-study evidence in S6 Figure A and Tables D, E" } ], "date_published": "2026-10-25", "findings": [ { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "Crop of eight lines of the abstract. Red boxes mark '36\\%' at the start of the line 'more likely to be removed by Wikipedia editors than positive facts within 12 hours and' and '29\\%' at the start of the next line, 'more likely within 3 days'.", "caption": "Abstract, p. 1 (PDF p. 1). The marked figures are the 36% and 29% said to be the increases in the likelihood of removal within 12 hours and within 3 days.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E001.png" }, { "alt": "Crop of the last three lines of page 4. A red box marks '64\\%' in 'uncited (but true) facts are 64\\% more likely to be', where the page ends.", "caption": "Results, “Effects of Citations on Survival”, last lines of p. 4 (PDF p. 4). The marked figure is the 64%; the sentence continues on p. 5 with “removed than cited facts within four hours (hazard ratio = 2.33, $p<0.001$)”.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E001-b.png" }, { "alt": "Listing of lines 30 to 43 of the Stata program. Red boxes mark line 33, 'disp ((abs(75 - 52)/((75 + 52)/2)) * 100)', line 38, 'disp ((abs(92 - 69)/((92 + 69)/2)) * 100)', and line 43, 'disp ((abs(62 - 32)/((62 + 32)/2)) * 100)'. The comment lines above them read 'Negative facts are 36\\% more likely to be removed', 'and 29\\% more likely within 3 days' and 'Uncited facts are 43\\% more likely to be removed'.", "caption": "Replication program Kalla_Aronow_PLOS_2015.do, lines 30–43, typeset from the file. The marked lines compute the three percentages: each divides the difference of the two counts by their mean.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E001-c.png" }, { "alt": "Dot plot with three rows and a horizontal axis labeled from 0 to 100 per cent. Negative against positive facts within 12 hours (75 against 52 of 150 removed): circle labeled 36 per cent, relative increase 44.2 per cent. Negative against positive facts within 3 days (92 against 69 of 150): circle labeled 29 per cent, relative increase 33.3 per cent. Uncited against cited facts within 4 hours (62 against 32 of 100): circle labeled 64 per cent, relative increase 93.8 per cent. In each row the square for the relative increase lies to the right of the open circle for the symmetric difference.", "caption": "The three comparisons, from the recount of the replication data. Open circles: the difference of the two shares removed divided by their mean, labeled with the rounded printed figures. Squares: the increase relative to the share removed in the reference group.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E001-percentages.png" } ], "formal_release": null, "formal_statements": [ { "id": "P041-T3", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T3" }, { "id": "P041-T4", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T4" }, { "id": "P041-T12", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T12" } ], "headline": "The printed 36%, 29% and 64% are differences divided by the mean of the two shares of facts removed; relative to the positive facts and to the cited facts, the increases in the share removed are 44%, 33% and 94%.", "id": "P041-E001", "location": "Abstract, p. 1 (PDF p. 1); Results, “Effects of Citations on Survival”, pp. 4–5 (PDF pp. 4–5); replication program Kalla_Aronow_PLOS_2015.do, lines 33, 38 and 43", "pdf_page": 1, "printed_page": "1", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Moderate", "subject": "paper", "target_ids": [ "P041-T12", "P041-T3", "P041-T4" ], "text": { "as_printed": "Abstract, p.~1: Negative facts are \\hl{36\\%} more likely to be removed by Wikipedia editors than positive facts within 12 hours and \\hl{29\\%} more likely within 3 days.\n\nResults, pp.~4--5: Consistent with Wikipedia's policy, uncited (but true) facts are \\hl{64\\%} more likely to be removed than cited facts within four hours (hazard ratio = 2.33, $p<0.001$).\n\nReplication program, lines 33, 38 and 43:\n\n\\texttt{disp ((abs(75 - 52)/\\hl{((75 + 52)/2)}) * 100)}\n\n\\texttt{disp ((abs(92 - 69)/\\hl{((92 + 69)/2)}) * 100)}\n\n\\texttt{disp ((abs(62 - 32)/\\hl{((62 + 32)/2)}) * 100)}", "corrected": "Abstract, p.~1: Negative facts are \\hl{44\\%} more likely to be removed by Wikipedia editors than positive facts within 12 hours and \\hl{33\\%} more likely within 3 days.\n\nResults, pp.~4--5: Consistent with Wikipedia's policy, uncited (but true) facts are \\hl{94\\%} more likely to be removed than cited facts within four hours (hazard ratio = 2.33, $p<0.001$).\n\nHere the increase is computed relative to the reference group. Let $p_1$ be the share of facts removed by the stated time in the comparison group (negative facts, or uncited facts) and $p_0>0$ the share in the reference group (positive facts, or cited facts). The increase is\n\\[\n100\\Bigl(\\frac{p_1}{p_0}-1\\Bigr)=100\\,\\frac{p_1-p_0}{\\hl{p_0}} ,\n\\]\nwhich gives 44.23\\%, 33.33\\% and 93.75\\% with the stored survival times. In lines 33, 38 and 43 of the replication program, the divisor is the count for the reference group: $100\\,(75-52)/\\hl{52}$, $100\\,(92-69)/\\hl{69}$ and $100\\,(62-32)/\\hl{32}$. An alternative is to keep 36\\%, 29\\% and 64\\% and to describe them as differences relative to the mean of the two shares, which reports a different measure.", "evidence": "The evidence is a recount from the replication data and exact arithmetic. I count a fact as removed by a horizon when its removal is recorded and its stored survival time is at most the horizon, as the program does. For negative against positive facts, Studies 1, 2 and 4 are pooled, with 150 facts in each group. For uncited against cited facts, Studies 1 and 4 are pooled, with 100 facts in each group. The counts agree with the numbers written into the program. Since the groups have equal sizes, I have $p_A/p_B=a/b$, and a direct computation gives\n\\[\n\\begin{aligned}\n100\\Bigl(\\frac{75}{52}-1\\Bigr)&=\\frac{575}{13}=44.23\\ldots,\\\\\n100\\Bigl(\\frac{92}{69}-1\\Bigr)&=\\frac{100}{3}=33.33\\ldots,\\\\\n100\\Bigl(\\frac{62}{32}-1\\Bigr)&=\\frac{375}{4}=93.75 .\n\\end{aligned}\n\\]\nEvaluating the three expressions in the program, I find\n\\[\n\\begin{aligned}\n\\frac{100\\cdot 23}{127/2}&=\\frac{4600}{127}=36.22\\ldots,\\\\\n\\frac{100\\cdot 23}{161/2}&=\\frac{200}{7}=28.57\\ldots,\\\\\n\\frac{100\\cdot 30}{94/2}&=\\frac{3000}{47}=63.83\\ldots .\n\\end{aligned}\n\\]\nEach printed integer is a rounded value from the second display, and none is a rounded value from the first. To compare the two measures in general, let $p_A>p_B>0$ and $d=p_A-p_B$, and write $R=100\\,d/p_B$ and $M=200\\,d/(p_A+p_B)$. Then\n\\[\nR-M=\\frac{100\\,d^{2}}{p_B\\,(p_A+p_B)}>0 .\n\\]\nThis implies that the symmetric difference is the smaller of the two measures. The Harvard and the Yale copies of the replication program contain the same three expressions.\n\n\\begin{audittable}{lrrrrr}{Facts removed in the comparison group and in the reference group (positive facts in the first two rows, cited facts in the third), the printed figure, the difference divided by the mean of the two shares, and the relative increase.}\nComparison & Removed & Reference & Printed & Over the mean & Relative \\\\\nNegative vs positive, 12 hours & 75 of 150 & 52 of 150 & 36\\% & 36.22\\% & 44.23\\% \\\\\nNegative vs positive, 3 days & 92 of 150 & 69 of 150 & 29\\% & 28.57\\% & 33.33\\% \\\\\nUncited vs cited, 4 hours & 62 of 100 & 32 of 100 & 64\\% & 63.83\\% & 93.75\\%\n\\end{audittable}", "problem": "The sentence ``$A$ is $x$\\% more likely to be removed than $B$ within a given time'' compares the shares of facts removed by that time, with $B$ as the reference group. If $p_A$ and $p_B$ are the two shares, it means that $p_A=(1+x/100)\\,p_B$, that is, $x=100\\,(p_A/p_B-1)$. The three printed figures were computed differently. Let $a$ and $b$ be the numbers of facts removed in the two groups. The replication program evaluates $100\\,|a-b|/\\{(a+b)/2\\}$, the difference divided by the mean of the two counts. Since the groups have equal sizes, this equals $100\\,|p_A-p_B|/\\{(p_A+p_B)/2\\}$, a symmetric percentage difference. When $p_A>p_B>0$, the symmetric difference is smaller than the relative increase. The counts are 75 negative against 52 positive facts of 150 at 12 hours, 92 against 69 of 150 at 72 hours, and 62 uncited against 32 cited facts of 100 at four hours. The printed 36\\%, 29\\% and 64\\% are the symmetric differences 36.22\\%, 28.57\\% and 63.83\\%, rounded. The relative increases are 44.23\\%, 33.33\\% and 93.75\\%, and rounding does not account for the gap. The article gives no convention under which ``$x$\\% more likely'' means a symmetric difference. Someone who quotes the three figures as relative increases quotes numbers that are too small.", "what_survives": "Correcting only these percentages leaves the three pairs of counts, the direction of the three comparisons, the Kaplan--Meier curves, and the Cox estimates with their $p$-values (hazard ratios 1.59 and 2.33) unchanged; I have not checked the correctness of those other results here. The corrected increases are larger than the printed ones. If the survival times are rebuilt from the recorded minutes (\\fid{P041-E002}), converted from milliseconds to hours by division by $3{,}600{,}000$, and counted with an inclusive 12-hour cutoff, then 76 negative facts are removed within 12 hours and the first increase is $600/13=46.15\\ldots$\\%." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E001", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "exact_counterexample", "exhibits": [ { "alt": "Crop of the heading 'Estimation of Experimental Effects' and the three lines below it. A red box marks 'survival time of the facts' in 'Our primary dependent variable is the survival time of the facts, the amount of time between when the fact is added to Wikipedia and when it is removed'.", "caption": "“Estimation of Experimental Effects”, p. 4 (PDF p. 4). The marked words name the quantity whose stored values are affected; the definition itself is correct.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E002.png" }, { "alt": "Crop of six lines of the codebook. Red boxes mark the variable names 'time\\_edit:', 'time\\_removed:' and 'time\\_alive:' at the start of their definitions, 'Date and time the edit was made, in Stata format', 'Date and time the edit was removed, in Stata format' and 'Difference between time\\_removed and time\\_edit'.", "caption": "Codebook of the replication files (Kalla_Aronow_PLOS_2015.pdf), p. 1. The marked names are the three variables that the data file stores in single precision.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E002-b.png" }, { "alt": "Listing of lines 11 to 17 of the Stata program. Red boxes mark line 12, 'gen curr\\_time = time\\_alive', and line 15, 'gen hours\\_alive=curr\\_time*2.77778e-7'.", "caption": "Replication program Kalla_Aronow_PLOS_2015.do, lines 11–17, typeset from the file. The marked lines take the stored survival time and convert it to hours with a rounded multiplier.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E002-c.png" } ], "formal_release": null, "formal_statements": [ { "id": "P041-T4", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T4" } ], "headline": "The replication data store the edit and removal times in single precision, so that the stored survival times differ from the recorded ones by up to 129 seconds; one 12-hour count changes from 75 to 76, and the pooled hazard ratios change in the third decimal.", "id": "P041-E002", "location": "“Estimation of Experimental Effects”, p. 4 (PDF p. 4); replication files: the three time variables of the data file, the codebook (p. 1), and lines 12 and 15 of the main Stata program", "pdf_page": 4, "printed_page": "4", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [ "P041-T4" ], "text": { "as_printed": "Article, p.~4: Our primary dependent variable is the survival time of the facts--the amount of time between when the fact is added to Wikipedia and when it is removed.\n\nCodebook, p.~1:\n\n\\hl{\\texttt{time\\_edit}}: Date and time the edit was made, in Stata format\n\n\\hl{\\texttt{time\\_removed}}: Date and time the edit was removed, in Stata format. Coded as missing if the edit was not removed.\n\n\\hl{\\texttt{time\\_alive}}: Difference between \\texttt{time\\_removed} and \\texttt{time\\_edit}. Amount of time the edit was alive on the Wikipedia page.\n\nReplication program, lines 12 and 15:\n\n\\texttt{gen curr\\_time = time\\_alive}\n\n\\texttt{gen hours\\_alive=curr\\_time*\\hl{2.77778e-7}}", "corrected": "Rebuild \\texttt{time\\_edit} and \\texttt{time\\_removed} from the recorded strings \\texttt{time} and \\texttt{removed} as millisecond datetimes \\hl{in double precision}, keeping a missing value where the fact was not removed. Set \\texttt{time\\_alive} to the difference of the rebuilt times, and compute the hours as this difference \\hl{divided by $3{,}600{,}000$}, in double precision. Converting the stored floats to doubles does not recover the recorded minutes. For the reported Cox recalculation, \\hl{retain the program's censoring time of $5\\times10^9$ ms and replace zero durations by $0.0001$ ms before converting to hours}. Equal rebuilt durations are ties at the recorded resolution of one minute. With these survival times, \\hl{76} of the 150 negative facts of Studies 1, 2 and 4 are removed within 12 hours (stored values: 75), and the pooled hazard ratios of Table A of S6 File are \\hl{$0.628$} and \\hl{$0.428$} (stored values: $0.629$ and $0.429$).", "evidence": "The evidence is an exact counterexample from the data, with a recalculation. Set $q=2^{17}=131{,}072$. For the fact on the page of Jon Tester in Study 5, the recorded times are 18:49 and 20:47 on 6 November 2014, which are 118 minutes apart. In milliseconds the two times are\n\\[\n\\begin{aligned}\nX_1&=1{,}730{,}918{,}940{,}000=13{,}205{,}863\\,q+64{,}864,\\\\\nX_2&=1{,}730{,}926{,}020{,}000=13{,}205{,}917\\,q+66{,}976 .\n\\end{aligned}\n\\]\nSince $64{,}864h\\}$. With the outcome $S_i(h)$, all twelve printed slopes of the treatment indicators are reproduced to three decimals. For example, the six-hour slope for positive facts in Studies 1, 2 and 4 is $17/150$, printed as 0.113. The removal indicator is $R_i(h)=1-S_i(h)$. Since each regression has an intercept, I have\n\\[\n\\widehat\\beta_R=e_0-\\widehat\\beta_S,\n\\]\nfor the two vectors of coefficients, where $e_0$ is the coordinate vector of the intercept. Every slope changes sign ($-17/150$ in the example), the residuals change sign, and the robust covariance matrix is the same.", "problem": "The two tables are titled ``OLS Estimates on Survival''. They regress an indicator that the fact was still present after 6, 12 or 24 hours or 7 days, and the printed slopes have the signs of that outcome. The sentence calls the outcome removal, for which every slope would have the opposite sign. The paragraph opens by naming survival, and the sentence that follows interprets the slopes correctly. The intended meaning can be recovered from them. The Harvard Dataverse copy (version 1.0) of the do-file codes the four outcomes as survival. The Yale Dataverse copy (version 2.0) codes them as removal by the stated time, and returns every slope with the sign opposite to the printed one.", "what_survives": "All twelve printed slopes and their interpretation stand, and no coefficient changes. The Cox models, the Kaplan--Meier curves and Fig~1 use the duration and the failure indicator, not these four outcomes. The Yale copy of the do-file reproduces the printed signs once the coding of lines 147--154 of the Harvard copy is restored. That coding sets each of \\texttt{hour6}, \\texttt{hour12}, \\texttt{hour24} and \\texttt{day7} to 1, and replaces it by 0 when \\texttt{hours\\_alive} is at or below the stated time." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E010", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "Table A, OLS Estimates on Survival at 6 Hours and 12 Hours, with six columns. In the second column the coefficient for Cited is 0.320 with three stars, and a red rectangle marks the standard error (0.645) printed below it. The corresponding 12-hour entry in the fifth column is 0.370 with three stars and standard error (0.065).", "caption": "S11 File, Table A, on the first page of the supplement as rendered from the published Word file. The standard error (0.645) of the six-hour coefficient for “Cited” is marked; the entry to its right, at 12 hours, is (0.065).", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E011.png" } ], "formal_release": null, "formal_statements": [ { "id": "P041-T7", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T7" } ], "headline": "The standard error of the citation coefficient at six hours in Table A of S11 File is 0.064, not 0.645.", "id": "P041-E011", "location": "S11 File, Table A, column 2 (6 hrs; $N=200$; Studies 1,4), standard error of “Cited”; cited from p. 4 (PDF p. 4)", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [ "P041-T7" ], "text": { "as_printed": "\\begin{audittable}{lr}{Table A: OLS Estimates on Survival at 6 Hours and 12 Hours}\nCoefficient: & 6 hrs \\\\\nCited & \\\\\nCoefficient: & 0.320*** \\\\\nStandard error: & (\\hl{0.645}) \\\\\n\\end{audittable}", "corrected": "\\begin{audittable}{lr}{Table A: OLS Estimates on Survival at 6 Hours and 12 Hours}\nCoefficient: & 6 hrs \\\\\nCited & \\\\\nCoefficient: & 0.320*** \\\\\nStandard error: & (\\hl{0.064}) \\\\\n\\end{audittable}", "evidence": "The evidence is a recalculation from the replication data in exact arithmetic. I regress the indicator of survival beyond six hours on an intercept, the citation indicator and an indicator for Study~4, in Studies 1 and 4. The regression has $n=200$ and $k=3$, and its citation coefficient is $8/25=0.320$. Each of the four cells of study by citation holds 50 facts, and the sum of squared residuals is $2047/50$. The HC1 variance of the citation coefficient is\n\\[\n\\frac{200}{197}\\cdot\\frac{1}{10000}\\cdot\\frac{2047}{50}=\\frac{2047}{492500},\n\\]\nand taking the square root, I find the standard error $\\sqrt{2047/492500}=0.0644697\\ldots$, which rounds to 0.064. The $t$ statistic is then 4.96, with $p\\approx1.5\\times10^{-6}$ on 197 degrees of freedom, in agreement with the three stars.", "problem": "The printed standard error is about ten times too large. A coefficient of 0.320 with a standard error of 0.645 has $t=64/129\\approx0.496$ and a two-sided $p$ of about 0.62, which contradicts the three stars printed beside the coefficient. The standard errors of the same coefficient at 12 hours, 24 hours and 7 days are printed as 0.065, 0.062 and 0.053.", "what_survives": "The coefficient 0.320 and its three stars stand. The recalculated standard errors at 12 hours, 24 hours and 7 days match their printed cells, and no other calculation uses this cell." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E011", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "The top of a regression table titled 'Table B: Covariate Adjusted Cox Regression Pooled Estimates'. Under 'Positive', the coefficient 0.670** with standard error (0.095); under 'Cited', the coefficient 0.352*** with standard error (0.055). A red box marks 0.670**.", "caption": "S6 File, Table B, in a rendering of the published Word file. The marked entry, for “Positive”, carries two stars.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E012.png" }, { "alt": "A regression table titled 'Table E: Study by Study Cox Regression Estimates, with Covariate Adjustment', with five columns for Studies 1, 2, 4, 5 (1st Half) and 5 (2nd Half). Row Positive: 0.532***, 0.574, 0.753, 0.387***, 0.274*, with standard errors (0.108), (0.220), (0.200), (0.139), (0.162). Row Cited: 0.455***, blank, 0.239***, 0.480*, blank, with standard errors (0.090), (0.068), (0.174). Red boxes mark 0.274* in the last column and 0.480* in the fourth column.", "caption": "S6 File, Table E, in a rendering of the published Word file. The marked entries are “Positive” for Study 5 (2nd Half) and “Cited” for Study 5 (1st Half), each with one star. The first line of the image is the end of the note to Table D.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E012-b.png" }, { "alt": "Five rows of a regression table with two columns of entries. Under 'Positive X NE Region', the coefficients 0.316*** and 0.472* with standard errors (0.149) and (0.192). A red box marks 0.316***.", "caption": "S9 File, Table A, rows around “Positive X NE Region”, in a rendering of the published Word file. The marked entry in the first column carries three stars.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E012-c.png" } ], "formal_release": null, "formal_statements": [ { "id": "P041-T8", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T8" } ], "headline": "Four significance markers are one level off: “Positive” in S6 Table B should carry ***, the two entries for Study 5 in S6 Table E should carry **, and “Positive X NE Region” in the first column of S9 Table A should carry ** in place of ***.", "id": "P041-E012", "location": "S6 File, Table B (row “Positive”) and Table E (columns for Study 5); S9 File, Table A (row “Positive X NE Region”, first column). The supplementary files are Word documents without page numbers; they are listed on p. 6 (PDF p. 6) of the article.", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [ "P041-T8" ], "text": { "as_printed": "S6 File, Table B, ``Positive'': 0.670\\hl{**} (0.095).\n\nS6 File, Table E, ``Cited'', Study 5 (1st Half): 0.480\\hl{*} (0.174); ``Positive'', Study 5 (2nd Half): 0.274\\hl{*} (0.162).\n\nS9 File, Table A, ``Positive X NE Region'', first column: 0.316\\hl{***} (0.149).\n\nNote to each table: ``*Significant at the 10\\% level; **Significant at the 5\\% level; ***Significant at the 1\\% level''.", "corrected": "S6 File, Table B, ``Positive'': 0.670\\hl{***} (0.095).\n\nS6 File, Table E, ``Cited'', Study 5 (1st Half): 0.480\\hl{**} (0.174); ``Positive'', Study 5 (2nd Half): 0.274\\hl{**} (0.162).\n\nS9 File, Table A, ``Positive X NE Region'', first column: 0.316\\hl{**} (0.149).", "evidence": "The evidence is a calculation from the printed numbers and a recalculation from the replication files. For a printed hazard ratio $\\widehat{H}$ with standard error $s$, the Wald statistic for the log hazard ratio and its two-sided $p$-value are\n\\[\nz=\\frac{\\log \\widehat{H}}{s/\\widehat{H}},\\qquad p=2\\,\\Phi(-|z|),\n\\]\nwhere $\\Phi$ is the standard normal distribution function. The table below lists $z$ and $p$ at the printed values, bounds on $p$ over all hazard ratios and standard errors that round to the printed values, and the $p$-value when the same model is fitted to the replication files. In every row these bounds lie on one side of the thresholds 0.01 and 0.05, so that rounding cannot account for the printed marker.\n\n\\begin{audittable}{lllllll}{The four entries: the printed hazard ratio with its marker and standard error, the Wald statistic and $p$-value that these give, and the marker that follows from the note of the tables. ``S6 B'' is Table B of the S6 File, and ``5 (1st Half)'' is the first half of Study 5; the entry of S9 Table A is in its first column.}\nEntry & Printed & $z$ & $p$ & Bounds on $p$ & Refit $p$ & Marker \\\\\nS6 B, Positive & 0.670** (0.095) & $-2.82$ & 0.0047 & 0.0044--0.0051 & 0.0045 & *** \\\\\nS6 E, Cited, 5 (1st Half) & 0.480* (0.174) & $-2.02$ & 0.0429 & 0.0418--0.0440 & 0.0434 & ** \\\\\nS6 E, Positive, 5 (2nd Half) & 0.274* (0.162) & $-2.19$ & 0.0285 & 0.0276--0.0296 & 0.0285 & ** \\\\\nS9 A, Positive X NE Region & 0.316*** (0.149) & $-2.44$ & 0.0146 & 0.0139--0.0152 & 0.0144 & ** \\\\\n\\end{audittable}", "problem": "The tables print hazard ratios with standard errors on the scale of the hazard ratio, and their notes assign one star to the 10\\% level, two to the 5\\% level and three to the 1\\% level. For the four entries above, the printed marker is not the one that the printed hazard ratio and standard error imply. Three markers understate the level of significance. The fourth, in S9 Table A, claims significance at 1\\% for an interaction that is significant at 5\\%.", "what_survives": "The four estimates and their standard errors are unchanged. The joint tests 0.321 and 0.221 of S9 Table A are also unchanged. The correction leaves in place the statement of the main text that there is no systematic evidence of moderation by covariates. The interaction with the Northeast region remains significant at 5\\%." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E012", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "exact_counterexample", "exhibits": [ { "alt": "Crop of the last four lines of the paragraph and the title of Table A. Red boxes mark 'using only the first' at the end of one line and 'randomly-assigned fact that each worker saw' at the start of the next.", "caption": "S12 File, paragraph introducing Table A, from a rendering of the published Word file (the supplement has no fixed page numbers). The marked words describe the restricted sample.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E013.png" }, { "alt": "Crop of Table A, 'Regression Analysis of Mechanical Turk Coding', and its note. Red boxes mark the three column heads 'First Fact', under Positive-Negative (3.26, standard error 0.23, N 112), Relevance (0.03, 0.25, 113) and Favorability (2.08, 0.24, 113), and the words 'column is only for the first fact seen' in the note.", "caption": "S12 File, Table A and its note, from the same rendering. The marked column heads and the marked words of the note label the restricted columns.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E013-b.png" }, { "alt": "Listing of lines 49 to 53 of the Stata program for the Mechanical Turk data. A red box marks 'drop if inputround==\"test\"' in line 49. Lines 51 to 53 read 'sort workerid submittime', 'bysort workerid: egen submitorder=seq()' and 'gen view\\_first=submitorder==1'.", "caption": "Replication program Kalla_Aronow_PLOS_2015_mturk.do, lines 49–53, typeset from the file. The marked command removes the quality-control items before the submissions of each worker are ranked.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E013-c.png" } ], "formal_release": null, "formal_statements": [], "headline": "The “First Fact” columns of S12 Table A use each worker’s first substantive fact after the quality-control items have been dropped; seven of the 113 workers had already completed a quality-control item.", "id": "P041-E013", "location": "S12 File, Results, paragraph introducing Table A; Table A, “First Fact” columns and note (Word supplement, no fixed pagination); replication program Kalla_Aronow_PLOS_2015_mturk.do, lines 49–53", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "As an additional robustness check designed to guard against any desirability bias and learning that may occur as Mechanical Turk workers rate up to 50 of these facts, we produce the same regression \\hl{using only the first randomly-assigned fact that each worker saw}. The results are substantively the same.\n\nTable A, ``\\hl{First Fact}'' columns: Positive-Negative, $3.26$ $(0.23)$, $N=112$; Relevance, $0.03$ $(0.25)$, $N=113$; Favorability, $2.08$ $(0.24)$, $N=113$.\n\nTable A, note: For each question, the first column reports the regression across all facts and the second \\hl{column is only for the first fact seen}.\n\nReplication program, lines 49 and 51--53:\n\n\\texttt{\\hl{drop if inputround==\"test\"}}\n\n\\texttt{sort workerid submittime}\n\n\\texttt{bysort workerid: egen submitorder=seq()}\n\n\\texttt{gen view\\_first=submitorder==1}", "corrected": "Either keep the printed numbers, head the restricted columns \\hl{``First substantive response''}, and replace the description of the sample and the note to the table by the following text.\n\nWe repeat the regressions using \\hl{each worker's first submitted substantive fact, after removing quality-control items}. Seven workers had previously completed a quality-control item; the restriction therefore avoids prior substantive responses but \\hl{does not exclude prior quality-control exposure}.\n\nOr restrict the sample to the workers whose first recorded assignment, quality-control items included, was substantive, and replace the three restricted columns by \\hl{$3.34$ $(0.24)$, $N=105$}; \\hl{$-0.02$ $(0.23)$, $N=106$}; and \\hl{$2.13$ $(0.25)$, $N=106$}. The text should then say that the order is that of accepted and submitted assignments, since no log of views exists.", "evidence": "The evidence is an exact counterexample from the replication data, with a recalculation. One worker submitted a quality-control item at 01:51:28 GMT on 9 December 2014 and accepted the selected fact, a Study 4 fact about Ed Markey, at 01:51:35, seven seconds later. For the seven workers the gaps between the submission of the quality-control item and the acceptance of the selected fact are 7, 3, 1, 1, 13, 6620 and 30 seconds. All six printed regressions of Table A reproduce at the printed precision. The slope of a regression on a binary indicator is the difference of the two group means. Without the seven workers, I find that the three slopes are $1535/459$, $-7/351$ and $2987/1404$.\n\n\\begin{audittable}{lrrrr}{Table A, restricted columns: coefficient (standard error clustered by worker) and $N$.}\nQuestion & As printed & $N$ & Without the seven workers & $N$ \\\\\nPositive-Negative & 3.26 (0.23) & 112 & 3.34 (0.24) & 105 \\\\\nRelevance & 0.03 (0.25) & 113 & $-0.02$ (0.23) & 106 \\\\\nFavorability & 2.08 (0.24) & 113 & 2.13 (0.25) & 106\n\\end{audittable}\n\nThe times are those of accepted and submitted assignments and do not show what a worker viewed. One retained worker accepted four substantive facts in the same second, so the order of viewing cannot be established for every worker.", "problem": "S12 File counts the 49 quality-control items among the facts that a worker could be shown: facts were sampled ``from the set of 551 facts included in the five studies, plus an additional 49 facts that were seeded as quality control'', up to 50 per worker. The program drops the quality-control items in line 49 and ranks the submissions of each worker afterwards, in lines 51--53. The variable \\texttt{view\\_first} therefore marks each worker's first submitted substantive fact. Seven of the 113 workers had completed a quality-control item before that fact. For these seven the selected fact is not the first that the worker saw, and the restriction does not exclude earlier exposure to an item, which is the purpose stated for it.", "what_survives": "The six printed regressions are correct for the samples that the program selects. Without the seven workers the pattern is the same: large positive differences for the positive-negative question and for favorability, and a difference near zero for relevance. The columns for all responses and the Wikipedia analyses do not use this restriction." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E013", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "page_check", "exhibits": [ { "alt": "A box plot with two groups on the horizontal axis and seven tick labels, set at an angle, on the vertical axis. Red boxes mark the two group labels, which read 'Nega2ve' and 'Posi2ve', and the top tick label, in which a percent sign is printed over the gap between 'Very' and 'Pos'. From top to bottom, the other tick labels are Pos, Somewhat Pos, Neither, Somewhat Neg, Neg and Very Neg, and each label of two words has a percent sign in place of its space.", "caption": "Figure A of the S12 File, in an earlier rendering of the embedded image. The marked group labels read “Nega2ve” and “Posi2ve”, and in the marked tick label a percent sign stands between “Very” and “Pos”. How the figure appears in Word with its original fonts was not checked.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E014.png" } ], "formal_release": null, "formal_statements": [], "headline": "The labels stored in the three figures of the S12 File are malformed: “Negative” and “Positive” are stored with a digit in place of the letters “ti”, and the tick labels have a percent sign in place of a space.", "id": "P041-E014", "location": "S12 File, Figures A–C (images embedded in a Word document without page numbers); the S12 File is listed on p. 6 (PDF p. 6) of the article", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "\\begin{audittable}{lll}{Text stored in the three figures of the S12 File.}\nFigure & Group labels & Tick labels of two words \\\\\nA & Nega\\hl{2}ve, Posi\\hl{2}ve & Very\\hl{\\%}Neg, Somewhat\\hl{\\%}Neg, Somewhat\\hl{\\%}Pos, Very\\hl{\\%}Pos \\\\\nB & Nega\\hl{3}ve, Posi\\hl{3}ve & Very\\hl{\\%}Irr, Somewhat\\hl{\\%}Irr, Somewhat\\hl{\\%}Rel, Very\\hl{\\%}Rel \\\\\nC & Nega\\hl{5}ve, Posi\\hl{5}ve & Very\\hl{\\%}Unfav, Somewhat\\hl{\\%}Unfav, Somewhat\\hl{\\%}Fav, Very\\hl{\\%}Fav \\\\\n\\end{audittable}", "corrected": "Figures A--C are drawn again from the same data, with the group labels ``\\hl{Negative}'' and ``\\hl{Positive}'' and with spaces in the tick labels: ``Very Neg'', ``Somewhat Neg'', ``Somewhat Pos'' and ``Very Pos'' in Figure A, ``Very Irr'', ``Somewhat Irr'', ``Somewhat Rel'' and ``Very Rel'' in Figure B, and ``Very Unfav'', ``Somewhat Unfav'', ``Somewhat Fav'' and ``Very Fav'' in Figure C.", "evidence": "The evidence is a check of the source file. I read the text that is stored in the three images embedded in the file, which consists of ordinary characters in the font Calibri. I did not check how the labels appear in Word with the original fonts.", "problem": "Each figure of the S12 File is an image embedded in the Word document, and the image stores its labels as text. In all three figures this text is malformed. The commands in the replication files that draw the figures contain the intended labels: ``Negative'' and ``Positive'' for the two groups, and tick labels such as ``Very Neg'' and ``Somewhat Pos''.", "what_survives": "No number depends on the labels, and the two groups and the response scale keep their intended meaning." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E014", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "The three-line caption of Figure 1. A red rectangle marks the word fare in the sentence, Results show that among active senators, negative facts are fare more likely to be removed more quickly than positive facts.", "caption": "Fig 1 caption, p. 4 (PDF p. 4), with the word “fare” marked in the last sentence.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E015.png" }, { "alt": "Four lines of text. A red rectangle marks the word pages in the sentence, We created 5 Wikipedia pages and made around 20 political, non-Senate edits using these accounts before beginning the experiment.", "caption": "S3 File, “Description of implementation”, last paragraph, on the single page of the supplement as rendered from the published Word file. The word “pages” is marked; the next line continues “using these accounts”.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E015-S3.png" }, { "alt": "Three lines of bold text beginning, S8 File. Investigates whether Senate class is a moderator of the effect if positive and cited edits. A red rectangle marks the word if at the end of the first line.", "caption": "S8 File, opening sentence, on the single page of the supplement as rendered from the published Word file, with the word “if” marked.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E015-S8.png" }, { "alt": "Eight lines of the codebook in typewriter type, from committeechair to treatment_pre_order. Red rectangles mark the words page count in the entry, pagechar: Senator's Wikipedia page count before experiment began, and the words Study 5 in the entry, treatment_pre_order: Order in which edits were made in Study 5, pre-election.", "caption": "Codebook of the replication files (PDF p. 1), entries pagechar and treatment_pre_order, with “page count” and “Study 5” marked.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E015-codebook.png" } ], "formal_release": null, "formal_statements": [], "headline": "Five slips of a word or label: read “far” for “fare”, “accounts” for “pages” and “effect of” for “effect if”, and in the codebook read Study 4 for Study 5 in the entry for the pre-election order and “character count” for “count” in the entry for page length.", "id": "P041-E015", "location": "Fig 1 caption, p. 4 (PDF p. 4), last sentence; S3 File, “Description of implementation”, last paragraph; S8 File, opening sentence; replication codebook, entries treatment_pre_order and pagechar", "pdf_page": 4, "printed_page": "4", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "\\begin{itemize}\n\\item Fig~1 caption: ``Results show that among active senators, negative facts are \\hl{fare} more likely to be removed more quickly than positive facts.''\n\\item S3 File: ``We created 5 Wikipedia \\hl{pages} and made around 20 political, non-Senate edits using these accounts before beginning the experiment.''\n\\item S8 File: ``Investigates whether Senate class is a moderator of the effect \\hl{if} positive and cited edits.''\n\\item Codebook: ``treatment\\_pre\\_order: Order in which edits were made in \\hl{Study 5}, pre-election.''\n\\item Codebook: ``pagechar: Senator's Wikipedia \\hl{page count} before experiment began.''\n\\end{itemize}", "corrected": "\\begin{itemize}\n\\item Fig~1 caption: ``Results show that among active senators, negative facts are \\hl{far} more likely to be removed more quickly than positive facts.''\n\\item S3 File: ``We created 5 Wikipedia \\hl{accounts} and made around 20 political, non-Senate edits using these accounts before beginning the experiment.''\n\\item S8 File: ``Investigates whether Senate class is a moderator of the effect \\hl{of} positive and cited edits.''\n\\item Codebook: ``treatment\\_pre\\_order: Order in which edits were made in \\hl{Study 4}, pre-election \\hl{(also carried into the corresponding Study 5 records)}.''\n\\item Codebook: ``pagechar: Senator's Wikipedia page \\hl{character} count before \\hl{the} experiment began.''\n\\end{itemize}", "evidence": "The evidence is a check of the page, the supplements and the codebook, and a recalculation from the replication data. In the data, \\texttt{treatment\\_pre\\_order} is missing in Studies 1 to 3, is filled for all 100 Study~4 records, and is copied in the same order into the matching 100 Study~5 records. The values of \\texttt{pagechar} range from 26686 to 631299.", "problem": "The caption and S8 File contain misprints. The sentence of S3 File concerns the accounts created to reach autoconfirmed status and itself continues ``using these accounts''. S2 File describes the same step as creating five Wikipedia accounts. The pre-election study is Study~4 (30--31 October 2014), and Study~5 followed the election (6--7 November 2014). The variable \\texttt{pagechar} is the length of the page in characters, as p.~4 and the notes to the supplementary tables say.", "what_survives": "These five wording changes do not alter any number or fitted result." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E015", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "The three Kaplan-Meier panels of Figure 1, each with a horizontal axis from 0 to 250 hours and a vertical axis from 0 to 1, and the legend below them. A red rectangle encloses the lower right corner of each panel, where survival zero meets the 250-hour boundary. No curve or segment lies at survival zero inside the rectangles.", "caption": "Fig 1, p. 4 (PDF p. 4), with the lower right corner of each panel marked. The published panels stop at 250 hours and carry no segment at survival zero there.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E017.png" }, { "alt": "Two renders of Figure 1, one above the other, each with three panels of survival curves against hours from 0 to 250. In the upper render the horizontal axis runs a little past 250, and a short red segment at survival zero sits to the right of the 250 tick in each panel, circled in blue. In the lower render the axis stops at 250 and there is no segment at survival zero.", "caption": "The plotting script run from the replication files. (a) As supplied, with R’s default axes: a segment at survival zero lies to the right of 250 hours in each panel (circled). (b) With the extension drawn only when a curve ends before 250 hours and with xaxs=\"i\": no such segment.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E017-default-axes.png" } ], "formal_release": null, "formal_statements": [], "headline": "In the plotting script, each survival curve should be extended to the 250-hour boundary only when the curve ends before 250 hours, and at its last survival level; as written, the loops draw a segment at survival zero beyond 250 hours, which the published Fig 1 does not show.", "id": "P041-E017", "location": "Fig 1, p. 4 (PDF p. 4), right-hand edge of the three panels; R plotting script of the replication files, lines 28 and 34–37, with the same loop at lines 43–46 and 51–54 and in the loops for Figure A of S6 File", "pdf_page": 4, "printed_page": "4", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "\\texttt{maxtime <- 250}\n\n\\texttt{for (i in 1:length(fit1\\$strata)) \\{ \\#number of curves}\n\n\\texttt{temp <- fit1[i]}\n\n\\texttt{\\hl{lines}(c(max(temp\\$time), maxtime), rep(min(temp\\$surv),2),}\n\n\\texttt{col=plot\\_colors[i],lty=plot\\_lty[i])}\n\n\\texttt{\\}}", "corrected": "\\texttt{maxtime <- 250}\n\n\\texttt{for (i in 1:length(fit1\\$strata)) \\{ \\#number of curves}\n\n\\texttt{temp <- fit1[i]}\n\n\\texttt{\\hl{if (max(temp\\$time) < maxtime)}}\n\n\\texttt{lines(c(max(temp\\$time), maxtime), rep(\\hl{tail(temp\\$surv,1)},2),}\n\n\\texttt{col=plot\\_colors[i],lty=plot\\_lty[i])}\n\n\\texttt{\\}}\n\nThe same change applies to the loops for \\texttt{fit2} and \\texttt{fit3} and to the loops that draw Figure~A of S6 File. To match the boundary of the published figure, each \\texttt{plot} call also takes \\texttt{\\hl{xaxs=\"i\"}}.", "evidence": "The evidence is a recalculation from the replication files and a check of the page. I ran the supplied plotting code with R's default axes and recorded the coordinates of each extension. For all eight curves the extension runs from 3463.58037 hours to 250 hours at survival zero. The render shows a zero-level segment to the right of 250 hours in each panel. A second render, with the extension drawn only when \\texttt{max(temp\\$time) < maxtime} and with \\texttt{xaxs=\"i\"}, shows none. On the published page, viewed at 220 dpi, there is no mark at survival zero at the right-hand edge of any panel.", "problem": "After each panel is plotted, a loop extends every curve to the display limit of 250 hours. The extension is meant for a curve whose fit stops before 250 hours, but the loop has no such condition. It uses the final fitted survival level, which need not be the level at 250 hours. For the eight curves of Fig~1 the fit ends at 3463.58037 hours with survival zero. This is the time that the script assigns to facts that were never removed (\\fid{P041-N002}). The loop draws a segment at survival zero from 3463.58037 hours back to 250 hours. With R's default axis style, \\texttt{xaxs=\"r\"}, the horizontal range 0 to 250 is widened to $-10$ to 260, and the part of the segment between 250 and 260 hours appears inside each panel. The published Fig~1 stops at 250 hours and shows no such segment.", "what_survives": "The published Fig~1 and the survival curves it shows over 0 to 250 hours are unaffected. Only a new render of the supplied script with default axes shows the extra segments. The same loop, without a condition, draws Figure~A of S6 File." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E017", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "clarity_note", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "recalculation_from_source", "exhibits": [ { "alt": "Crop of the eight-line paragraph that introduces Table A. Red boxes mark 'more positive,' at the end of one line and 'equally relevant, and more favorable at statistically significant levels' at the start of the next.", "caption": "S12 File, paragraph introducing Table A, from a rendering of the published Word file (the supplement has no fixed page numbers). The marked words join “equally relevant” to “at statistically significant levels”.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-E018.png" } ], "formal_release": null, "formal_statements": [], "headline": "S12 File detects no difference in mean relevance between positive and negative facts ($-0.03$, standard error $0.07$); the sentence can be read as if equal relevance had been established at a significance level.", "id": "P041-E018", "location": "S12 File, Results, paragraph introducing Table A (Word supplement, no fixed pagination)", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": null, "subject": "paper", "target_ids": [], "text": { "as_printed": "Consistent with the box-plots, we see that positive facts were perceived to be more positive, \\hl{equally relevant}, and more favorable \\hl{at statistically significant levels}.", "corrected": "Consistent with the box-plots, positive facts were perceived as more positive and more favorable at statistically significant levels; \\hl{no statistically detectable difference in mean relevance was found}.", "intended_meaning": "The phrase ``at statistically significant levels'' applies to the two differences that Table A shows to be significant, $2.96$ (standard error $0.08$) for the positive-negative question and $2.30$ $(0.09)$ for favorability. For relevance the table shows $-0.03$ $(0.07)$, and the two mean ratings are $5.00$ and $5.03$, so ``equally relevant'' means that no difference in mean relevance was detected. Read in this way the sentence is correct. A test of a zero difference that does not reject is not a test of equivalence, and S12 File reports no equivalence margin or test." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-E018", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "exact_counterexample", "exhibits": [ { "alt": "The lower edge of the three panels of Figure 1, with horizontal axes labelled Hours and ticks from 0 to 250, the legend, and the caption. Red rectangles mark the tick label 250 of each panel and the caption title, Fig 1. Kaplan-Meier survival curves across all studies and all conditions.", "caption": "Fig 1, p. 4 (PDF p. 4): the title of the caption and the last tick of each horizontal axis are marked. The panels stop at 250 hours; the removals that the script invents lie at 3463.58 hours, outside the figure.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-N002.png" } ], "formal_release": null, "formal_statements": [], "headline": "The R script that draws the Kaplan–Meier figures builds its survival object without an event indicator, and the 257 facts that were never removed are entered as removals at an invented time of 3463.58 hours; line 14 should pass event=!wiki$dead.", "id": "P041-N002", "location": "Fig 1, p. 4 (PDF p. 4), and Figure A of S6 File; R plotting script of the replication files, lines 8–14", "pdf_page": 4, "printed_page": "4", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "\\texttt{wiki\\$dead <- is.na(wiki\\$time\\_alive)}\n\n\\texttt{wiki\\$time\\_alive2 <- wiki\\$time\\_alive}\n\n\\texttt{wiki\\$time\\_alive2[wiki\\$dead] <- max(wiki\\$time\\_alive*2,na.rm=TRUE)}\n\n\\texttt{wiki\\$time\\_alive2[wiki\\$time\\_alive==0] <- 0.1}\n\n\\texttt{wiki\\$hours\\_alive <- wiki\\$time\\_alive2*2.77778e-7}\n\n\\texttt{wiki\\$survdata <- \\hl{Surv(time=wiki\\$hours\\_alive)}\\#,event=dead)}", "corrected": "\\texttt{wiki\\$survdata <- Surv(time=wiki\\$hours\\_alive\\hl{, event=!wiki\\$dead})}\n\nLines 8--13 are unchanged. The indicator is the negation of \\texttt{dead}, since in this script \\texttt{dead} is \\texttt{TRUE} for a fact that was never removed.", "evidence": "The evidence is an exact counterexample and a recalculation from the replication data. I take two records, one removed at raw time 3600000 and one never removed. The script assigns the second record the raw time 7200000, and with the conversion constant $2.77778\\times10^{-7}=138889/500000000000$ the two times are $t_1=1250001/1250000$ and $t_2=2t_1$ hours. The Kaplan--Meier estimator is\n\\[\n\\widehat S(t)=\\prod_{u\\le t,\\ d(u)>0}\\frac{n(u)-d(u)}{n(u)},\n\\]\nwhere $n(u)$ is the number of records with time at least $u$ and $d(u)$ is the number of events at $u$. At $t_1$ I have $n=2$ and $d=1$ whatever the status of the second record, which gives $\\widehat S(t_1)=1/2$. With the statuses $(1,1)$ of the script, $n(t_2)=d(t_2)=1$ and $\\widehat S(t_2)=0$. With the statuses $(1,0)$, no factor is added at $t_2$ and $\\widehat S(t_2)=1/2$.\n\nIn the data, the largest recorded duration is 6234439680 in raw units. The invented time is then $169120135296/48828125=3463.58037086208$ hours, and all 551 statuses are 1. A fact that was never removed has time $3463.58>250$ in the survival object of the script and in the repaired one, and every recorded removal is the same in both. For every $u\\le250$ the two objects then have the same $n(u)$ and $d(u)$. This implies that the Kaplan--Meier products agree on $[0,250]$. In exact arithmetic on the data, the two products agree for all 26 fitted curves, 8 in Fig~1 and 18 in Figure~A of S6 File. The table lists the curves of Fig~1.\n\n\\begin{audittable}{llrrrrr}{Curves of Fig~1, as scripted and with the event indicator repaired. The value at 250 hours is the same in both cases. The last two columns give the value at the invented time of 3463.58 hours.}\nStudies & Condition & $N$ & Never removed & $\\widehat S(250)$ & As scripted & Repaired \\\\\n1 and 4 & Positive, cited & 51 & 16 & $16/51$ & 0 & $16/51$ \\\\\n1 and 4 & Positive, uncited & 49 & 5 & $5/49$ & 0 & $5/49$ \\\\\n1 and 4 & Negative, cited & 49 & 10 & $10/49$ & 0 & $10/49$ \\\\\n1 and 4 & Negative, uncited & 51 & 3 & $1/17$ & 0 & $1/17$ \\\\\n2 & Positive, cited & 50 & 37 & $4/5$ & 0 & $37/50$ \\\\\n2 & Negative, cited & 50 & 30 & $16/25$ & 0 & $3/5$ \\\\\n3 & Positive, cited & 75 & 67 & $68/75$ & 0 & $67/75$ \\\\\n3 & Negative, cited & 76 & 64 & $73/76$ & 0 & $16/19$ \\\\\n\\end{audittable}", "problem": "In R's survival package, \\texttt{Surv} called with a time and no event indicator gives every row status 1, an observed event. The outcome of the paper is the time until a fact is removed. In the data, 257 of the 551 facts were never removed: their \\texttt{time\\_alive} is missing and \\texttt{removed} is coded ``NO''. Line 10 assigns these facts twice the largest recorded duration, that is, 3463.58 hours, and line 14 enters them as removals at that time. The fitted curve of any group that contains such facts then falls to zero at 3463.58 hours, where a Kaplan--Meier curve with these facts censored keeps its last level. The column \\texttt{dead} is the missing-time indicator, not the removal indicator. It is \\texttt{TRUE} for the facts that were never removed, and passing it as \\texttt{event} would invert every status.", "what_survives": "Every Kaplan--Meier value shown in Fig~1 and in Figure~A of S6 File, which stop at 250 hours, is unchanged when only the event indicator is repaired. The Cox regressions, the regression tables and the counts at fixed times come from the Stata do-file and do not use the R object. The do-file sets \\texttt{dead} to 1 for a recorded removal and declares it with \\texttt{stset hours\\_alive, failure(dead)}. The fitted curves beyond 250 hours and any summary of the full fit do change. The repair keeps the time that the script assigns to facts that were never removed. Whether that time matches the actual end of monitoring is a separate question, which this correction does not settle." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-N002", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "page_check", "exhibits": [ { "alt": "Four lines of text beginning, We find no systematic evidence that this effect is moderated by covariates. Two red rectangles on the third line mark the words date of entry, and time of entry, and the citation see Table A in S9 File and S10 File.", "caption": "Results, p. 4 (PDF p. 4): the last sentence of the first paragraph, with “date of entry, and time of entry” and the citation “see Table A in S9 File and S10 File” marked.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-N005.png" } ], "formal_release": null, "formal_statements": [], "headline": "The sentence cites Table A of S9 File and S10 File for eight moderators of the valence effect, but the test for Senate class is in S8 File, S10 File is not a moderation analysis, and no test for date or time of entry is reported: cite S8 File and Table A of S9 File, and remove date and time of entry from the list.", "id": "P041-N005", "location": "Results, “Survival of Positive and Negative Facts on Wikipedia”, last sentence of the first paragraph, p. 4 (PDF p. 4)", "pdf_page": 4, "printed_page": "4", "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [], "text": { "as_printed": "We find no systematic evidence that this effect is moderated by covariates (state population, Senate class, character count of page, geographic region, length of incumbency, political party, \\hl{date of entry, and time of entry}; \\hl{see Table A in S9 File and S10 File}), as specified in our registered pre-analysis plan.", "corrected": "We find no systematic evidence that this effect is moderated by covariates (state population, Senate class \\hl{dichotomized as facing reelection in 2014}, character count of page, geographic region, length of incumbency, \\hl{and} political party; \\hl{see S8 File and Table A in S9 File}), as specified in our registered pre-analysis plan.\n\nDate and time of entry may return to the list once their interactions with valence are reported. The covariate called length of incumbency is the year of first election (\\fid{P041-E004}).", "evidence": "The evidence is a check of the page against the twelve supplements and the replication do-files. I write the Cox model with interactions as $h(t\\mid V,Z)=h_0(t)\\exp\\{\\beta V+\\gamma^\\top Z+\\delta^\\top(VZ)\\}$, where $V=1$ for a positive fact and $Z$ is the vector of covariates. Dividing the hazard of a positive fact by the hazard of a negative fact at $Z=z$, I find the hazard ratio $\\exp(\\beta+\\delta^\\top z)$. Moderation is a statement about $\\delta$, and S10 File estimates only $\\gamma$. The joint Wald tests of Table~A in S9 File have $p=0.321$ and $p=0.221$. From the rounded entries of S8 File, the interaction with Senate class has $p\\approx0.075$ without covariates and $p\\approx0.155$ with them.", "problem": "Table~A of S9 File reports interactions of the indicator for positive facts with party, influence, three regions, length of incumbency, log page length and log state population. It has no interaction with Senate class, date of entry or time of entry. S10 File regresses survival on covariates alone, with no valence indicator and no interaction, and therefore cannot test moderation. The one reported test of Senate class as a moderator of the valence effect is in S8 File, which the sentence does not cite. There Senate class is dichotomized as facing reelection in 2014, and the interaction is 1.750* (0.550) without covariates and 1.557 (0.485) with them. No supplement and neither copy of the do-file reports an interaction of valence with date or time of entry. These two variables enter only as additive controls.", "what_survives": "The analyses that are reported do not contradict the statement of no systematic evidence. The joint tests do not reject, and the interaction with Senate class is marginal only without covariates. The cited files do not document three of the eight listed moderators, and the corrected sentence covers six. The pooled valence estimates, the survival curves and the citation results do not depend on this sentence. I could not compare the sentence with the pre-analysis plan, which was not available." }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-N005", "work": { "arxiv_id": null, "doi": "10.1371/journal.pone.0136327", "title": "Editorial Bias in Crowd-Sourced Political Information", "url": "https://pmaronow.github.io/llm-audit/works/P041/" }, "work_id": "P041" }, { "before_review": null, "class": "error", "date_modified": "2026-10-25", "date_published": "2026-10-25", "evidence_type": "exact_counterexample", "exhibits": [ { "alt": "Crop of the four-line last paragraph of S1 File. Red boxes mark the sentence 'Because the order in which edits were made was randomly assigned, this does not bias our results.', which runs over the last two lines.", "caption": "S1 File, “Description of implementation”, last paragraph, from a rendering of the published Word file (the supplement has no fixed page numbers). The marked sentence is the assurance.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-N011.png" }, { "alt": "Crop of the last five lines of S5 File. Red boxes mark the sentence 'Because the order in which edits were made was randomly assigned, this does not bias our results.', which runs over the last two lines.", "caption": "S5 File, “Description of implementation”, end of the last paragraph, from a rendering of the published Word file. The marked sentence is the same assurance.", "url": "https://pmaronow.github.io/llm-audit/works/P041/exhibits/P041-N011-b.png" } ], "formal_release": null, "formal_statements": [ { "id": "P041-T5", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T5" }, { "id": "P041-T6", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T6" }, { "id": "P041-T13", "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T13" } ], "headline": "Random assignment of the order of the edits does not by itself ensure that speeding up Study 1, and converting the remaining uncited edits of Study 5 to cited ones, leave the comparisons unbiased.", "id": "P041-N011", "location": "S1 File and S5 File, “Description of implementation”, last sentence of each (Word supplements, no fixed pagination); the main text refers to S5 File on p. 3 (PDF p. 3)", "pdf_page": null, "printed_page": null, "reaches": [], "related_entries": [], "revisions": [], "same_defect_as": [], "severity": "Minor", "subject": "paper", "target_ids": [ "P041-T13", "P041-T5", "P041-T6" ], "text": { "as_printed": "S1 File: Because other Wikipedia editors became suspicious of this account's uncited edits and threatened to block the account, we sped up the implementation and finished implementing the experiment sooner than anticipated, on 15 July 2014. \\hl{Because the order in which edits were made was randomly assigned, this does not bias our results.}\n\nS5 File: After implementing the first 50 edits on 6 November 2014, other Wikipedia editors became suspicious. We decided to convert all of the uncited edits in the remaining 50 edits to cited edits. Thus the first 50 edits had all four conditions while the last 50 edits only had positive or negative conditions, with all facts cited. \\hl{Because the order in which edits were made was randomly assigned, this does not bias our results.}", "corrected": "S1 File and S5 File, in place of the marked sentence: \\hl{Random assignment of the original order prevents deliberate ordering by treatment. It does not by itself rule out treatment-dependent account responses or effects of the adaptive changes.}\n\nIf the comparisons are read as effects of an edit's own treatment, the text should also state the protocol and the conditions of exposure to which they refer.", "evidence": "The evidence is an exact counterexample to the sufficiency of random order, with a recalculation of two sensitivity checks. I take four pages, two assigned a cited fact (C) and two an uncited fact (U), as in one block of Study 1, and let the 24 orders of insertion be equally likely. Suppose that an uncited edit starts a period in which every later insertion from the account is removed, whatever its own citation status, and that no other removals occur. This response is hypothetical and is not a description of the actual editors. On the planned, slow schedule the period ends before the next insertion, no fact is removed, and the difference $D$ between the removal rates of cited and uncited facts is $0$. On the accelerated schedule, all remaining insertions fall inside the period that the first uncited edit starts. The edit at rank $r$ is then removed if and only if an earlier edit is uncited,\n\\[\nY_r=\\mathbf{1}\\{z_s=\\mathrm{U}\\ \\text{for some}\\ s1 or negative t is treated only as an algebraic function identity, not a survival-probability assertion.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.exponential_survival_power_laws", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Hazards.lean#L80", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:37f6d187cc2d04e2182dd3e3c5d21a5fc859604d5177199f542199d203462284", "conditional_on": [], "credit_reason": "The extra q≥1 cases are algebraic extensions; on 0p. 5/8 (PDF p. 5); the claim recurs in the Abstract, p. 1/8 (PDF p. 1). The stipulated exponential countermodel has common non-null hazard ratios, exact unequal 12-hour contrasts, and positive binomial expressions at all four observed counts.

", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’, last two sentences (‘This suggests that the editorial bias cannot be explained by … Instead, our findings suggest the editorial bias is limited to active politicians’), and Discussion, first paragraph (‘suggesting that this bias is limited to the pages of active politicians and is not systemic to Wikipedia’): the inference from the Study 3 comparison to confinement of the bias; the witness is the validator's construction and is not in the paper", "location": "p. 5/8 (PDF p. 5); the claim recurs in the Abstract, p. 1/8 (PDF p. 1)" }, "id": "P041-T1", "text": { "body": "p. 5/8 (PDF p. 5); the claim recurs in the Abstract, p. 1/8 (PDF p. 1). The stipulated exponential countermodel has common non-null hazard ratios, exact unequal 12-hour contrasts, and positive binomial expressions at all four observed counts.", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’, last two sentences (‘This suggests that the editorial bias cannot be explained by … Instead, our findings suggest the editorial bias is limited to active politicians’), and Discussion, first paragraph (‘suggesting that this bias is limited to the pages of active politicians and is not systemic to Wikipedia’): the inference from the Study 3 comparison to confinement of the bias; the witness is the validator's construction and is not in the paper", "location": "p. 5/8 (PDF p. 5); the claim recurs in the Abstract, p. 1/8 (PDF p. 1)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T1" }, { "declarations": [ { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The row covers the statistic bound, with the printed-pair p-value covered by R240-06.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_former_nondetection", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L375", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The target explicitly stipulates 1.96, so using 49/25 instead of the exact Gaussian quantile is faithful; the redundant active-SE restriction excludes no allowed four-tuple.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_former_imprecision", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L410", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "Independence is the target’s interpretation of the numerical standard-error formula, not an unproved independence conclusion in this scalar theorem.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_between_relative_effects", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L448", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The row covers the quantitative endpoint enclosures, with the interval-membership claims covered separately by R240-08.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.former_wald_endpoint_enclosures", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L757", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The count inputs are stipulated by the target; no archive extraction, timestamp authentication, or cited-only comparison is asserted.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.observed_early_contrasts", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Hazards.lean#L149", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The exported existential is sufficient for the non-implication component; the exact numerical witness is additionally established by R240-02.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.smaller_contrast_does_not_exclude_common_bias", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Hazards.lean#L167", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "conditional_on": [], "credit_reason": "The claimed p-values are approximations at the printed central pairs; the row is not assigned uniform normal-tail bounds over all rounding choices.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.corrected_normal_p_approximations_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GP041P042P045B1.lean#L195", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:12bc4b2711a3c8d9f7360df9bc5e7efae9ce2d73f20794368ad3627ab82c9fa1", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T2", "decision": "restricted", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T2" }, "html": { "body": "

p. 5/8 (PDF p. 5); the tables used are S6 Table A (rendered p. 1) and S6 Table C (rendered p. 2). The supplied rounded table entries give imprecise former-senator relative effects, smaller pooled early contrasts, and a countermodel to absence of relative bias.

", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’, last two sentences, and Discussion, first paragraph, last sentence, as repaired by Sol: nondetection in Study 3 with an imprecise relative estimate, smaller early absolute contrasts, and no exclusion of a relative bias on former senators' pages", "location": "p. 5/8 (PDF p. 5); the tables used are S6 Table A (rendered p. 1) and S6 Table C (rendered p. 2)" }, "id": "P041-T2", "text": { "body": "p. 5/8 (PDF p. 5); the tables used are S6 Table A (rendered p. 1) and S6 Table C (rendered p. 2). The supplied rounded table entries give imprecise former-senator relative effects, smaller pooled early contrasts, and a countermodel to absence of relative bias.", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’, last two sentences, and Discussion, first paragraph, last sentence, as repaired by Sol: nondetection in Study 3 with an imprecise relative estimate, smaller early absolute contrasts, and no exclusion of a relative bias on former senators' pages", "location": "p. 5/8 (PDF p. 5); the tables used are S6 Table A (rendered p. 1) and S6 Table C (rendered p. 2)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T2" }, { "declarations": [ { "claim_revision_id": "catalog:92536e78c0ea8983ded5cd9e25486abdb80cfff72bde9abd91a068358305e0c7", "conditional_on": [], "credit_reason": "The reference group and direction are exactly the printed wording’s natural reading fixed by the target; this does not refute the counts, qualitative direction, or Cox ratios.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.risk_percentages_counterexample", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Risks.lean#L30", "status": "counterexample_proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:92536e78c0ea8983ded5cd9e25486abdb80cfff72bde9abd91a068358305e0c7", "conditional_on": [], "credit_reason": "The scalar domain a>b>0 is exactly that in T3(iv), with no artificial probability upper bound.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.symmetric_understates_relative", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Risks.lean#L54", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:92536e78c0ea8983ded5cd9e25486abdb80cfff72bde9abd91a068358305e0c7", "conditional_on": [], "credit_reason": "Using counts rather than proportions leaves both measures unchanged when the two arms have equal size, as stipulated here.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.risk_measure_gaps", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Risks.lean#L71", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:92536e78c0ea8983ded5cd9e25486abdb80cfff72bde9abd91a068358305e0c7", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T3", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T3" }, "html": { "body": "

Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5). The three printed directional percentages fail individually, while the symmetric percentages round as printed and strictly understate directional increases.

", "headline": "Abstract, ‘Negative facts are 36% more likely to be removed by Wikipedia editors than positive facts within 12 hours and 29% more likely within 3 days’, and ‘Effects of Citations on Survival’, ‘uncited (but true) facts are 64% more likely to be removed than cited facts within four hours’, as printed; the counts are the validators' recounts of the deposited data and are not printed in the paper", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5)" }, "id": "P041-T3", "text": { "body": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5). The three printed directional percentages fail individually, while the symmetric percentages round as printed and strictly understate directional increases.", "headline": "Abstract, ‘Negative facts are 36% more likely to be removed by Wikipedia editors than positive facts within 12 hours and 29% more likely within 3 days’, and ‘Effects of Citations on Survival’, ‘uncited (but true) facts are 64% more likely to be removed than cited facts within four hours’, as printed; the counts are the validators' recounts of the deposited data and are not printed in the paper", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T3" }, { "declarations": [ { "claim_revision_id": "catalog:c8c636f52d67a96714db16d95f1013015af70fd9838ab45b8b8101f9194fddb6", "conditional_on": [], "credit_reason": "The conditional 76-count case is represented by evaluating the function at that distinct input; this neither asserts that 76 is the stored-duration count nor that a timestamp repair has occurred.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_risk_percentages", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Risks.lean#L81", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:c8c636f52d67a96714db16d95f1013015af70fd9838ab45b8b8101f9194fddb6", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T4", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T4" }, "html": { "body": "

Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, pp. 4/8–5/8 (PDF pp. 4–5). The supplied counts give the corrected directional percentages, the conditional 76-removal variant, and the symmetric alternative repair.

", "headline": "Abstract, the sentence with ‘36%’ and ‘29%’, and ‘Effects of Citations on Survival’, the sentence with ‘64%’, as repaired by Sol: 44%, 33% and 94%, computed as relative increases in the proportion removed", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, pp. 4/8–5/8 (PDF pp. 4–5)" }, "id": "P041-T4", "text": { "body": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, pp. 4/8–5/8 (PDF pp. 4–5). The supplied counts give the corrected directional percentages, the conditional 76-removal variant, and the symmetric alternative repair.", "headline": "Abstract, the sentence with ‘36%’ and ‘29%’, and ‘Effects of Citations on Survival’, the sentence with ‘64%’, as repaired by Sol: 44%, 33% and 94%, computed as relative increases in the proportion removed", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, pp. 4/8–5/8 (PDF pp. 4–5)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T4" }, { "declarations": [ { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "The broader finite-length quantifier loses no target case. The record’s generic mention of unbalanced inputs as stronger is unnecessary for T5(i), whose own text already includes unbalanced sequences.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.scrutiny_no_direct_effect", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L57", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "{'source': 'adjudication', 'decision': 'rejected', 'what': 'P041/P041_Order.lean:68 states that citationContrast m z (fun _ => 0) is zero for every m and every cited-rank subset. This matches T5(ii), which explicitly stipulates zero slow-schedule outcomes and concludes that the slow contrast is zero for every order; the clock mechanism is not an additional conclusion of this component. The m=0 extension uses totalized division, but every intended m≥1 case is retained and holds without division by zero, so the row is not vacuous or materially restricted.'}", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.slow_schedule_zero", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L68", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "The witness refutes sufficiency of random order under the target’s explicit reading. It does not refute actual unbiasedness of the study’s Cox estimates or model the distinct Study 5 implementation change.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.random_order_block_counterexample", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L94", "status": "counterexample_proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "This row supplies T5(iv), not a standalone assertion about actual Wikipedia bias. The target explicitly stipulates the balanced type-sequence law, so no extra assignment restriction is imposed.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.balanced_scrutiny_general_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GP041P042P045B1.lean#L359", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "Redundant compressed-weight certificate; the credited general and size-50 proofs establish the actual balanced-subset law and its moments.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.compressed_scrutiny50_certificate", "manifest_use": "not_used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L168", "status": "proved", "support_scope": "historical declaration", "used_in_solution": false }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "The m=50 specialization is exactly the component assigned to this row; it is not presented as an estimate of the actual study’s bias.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.study_one_size_scrutiny_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GP041P042P045B1.lean#L406", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.fixed_calendar_block_unbiased", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L110", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.independent_citation_boundary", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L131", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:bfbfa96c7ff29af85106046eef97c121db8ae4a01e8508def9a357b4cfac00a6", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T5", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T5" }, "html": { "body": "

S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2); the main article points to S5 on p. 3/8 (PDF p. 3). A uniformly balanced order with scrutiny gives zero direct effects and slow contrast but a positive accelerated expectation, with the full six-order table, general formula, size-50 instance, zero expected contrast for every fixed treatment-independent outcome pattern at every m >= 1, its four-rank example, and the independent-assignment boundary.

", "headline": "S1 File and S5 File, ‘Description of implementation’, last sentence of each (‘Because the order in which edits were made was randomly assigned, this does not bias our results’), as printed; the witness is the validator's construction and is not in the paper", "location": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2); the main article points to S5 on p. 3/8 (PDF p. 3)" }, "id": "P041-T5", "text": { "body": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2); the main article points to S5 on p. 3/8 (PDF p. 3). A uniformly balanced order with scrutiny gives zero direct effects and slow contrast but a positive accelerated expectation, with the full six-order table, general formula, size-50 instance, zero expected contrast for every fixed treatment-independent outcome pattern at every m >= 1, its four-rank example, and the independent-assignment boundary.", "headline": "S1 File and S5 File, ‘Description of implementation’, last sentence of each (‘Because the order in which edits were made was randomly assigned, this does not bias our results’), as printed; the witness is the validator's construction and is not in the paper", "location": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2); the main article points to S5 on p. 3/8 (PDF p. 3)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T5" }, { "declarations": [ { "claim_revision_id": "catalog:74a813acf3f8d7c2bef10ed732590baf12e2b7c82094de94a4c6174f63ac6f8c", "conditional_on": [], "credit_reason": "The no-interference condition is encoded by the argument type of g rather than a separate proposition; all real-valued arrays allowed by T6(a) are represented.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.random_order_without_interference", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L234", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:74a813acf3f8d7c2bef10ed732590baf12e2b7c82094de94a4c6174f63ac6f8c", "conditional_on": [], "credit_reason": "The scope as a four-rank auxiliary identity is explicit; it is not credited as proving the all-m clause alone.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.block_without_interference", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L265", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:74a813acf3f8d7c2bef10ed732590baf12e2b7c82094de94a4c6174f63ac6f8c", "conditional_on": [], "credit_reason": "This is the m=2 supporting example specifically described by the row; the general result is R240-17.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.adaptive_protocol_estimands_differ", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Order.lean#L276", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:74a813acf3f8d7c2bef10ed732590baf12e2b7c82094de94a4c6174f63ac6f8c", "conditional_on": [], "credit_reason": "The balanced law is stipulated by the target; the scrutiny rule is an example rather than an empirical assertion about editors, and the direct-effect clause also covers the off-balance interventions.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.adaptive_protocol_estimands_general_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GP041P042P045B1.lean#L422", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:74a813acf3f8d7c2bef10ed732590baf12e2b7c82094de94a4c6174f63ac6f8c", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T6", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T6" }, "html": { "body": "

S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2). Rank-and-own-treatment outcomes give the average direct effect, whereas the scrutiny protocol separates the direct, planned, and implemented effects.

", "headline": "S1 File and S5 File, last sentence of ‘Description of implementation’, as repaired by Sol: random order prevents ordering by treatment and does not by itself rule out effects of earlier treatments on later edits or of the adaptive changes", "location": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2)" }, "id": "P041-T6", "text": { "body": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2). Rank-and-own-treatment outcomes give the average direct effect, whereas the scrutiny protocol separates the direct, planned, and implemented effects.", "headline": "S1 File and S5 File, last sentence of ‘Description of implementation’, as repaired by Sol: random order prevents ordering by treatment and does not by itself rule out effects of earlier treatments on later edits or of the adaptive changes", "location": "S1 File, last paragraph (rendered p. 2); S5 File, last paragraph (rendered p. 2)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T6" }, { "declarations": [ { "claim_revision_id": "catalog:5cc1b2741de12ff08aa9f0d39472dfa56e9f11744937f86036994fb079d859e5", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.citation_ols_exact_fit", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L145", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:5cc1b2741de12ff08aa9f0d39472dfa56e9f11744937f86036994fb079d859e5", "conditional_on": [], "credit_reason": "The matrix and counts are stipulated target inputs; this row does not claim to certify data extraction.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.citation_gram_inverse_is_inverse", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L171", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:5cc1b2741de12ff08aa9f0d39472dfa56e9f11744937f86036994fb079d859e5", "conditional_on": [], "credit_reason": "The assigned component asks for squared influence rows, sufficient for this variance. Signed ±1/100 influence values and an expanded generic hc1Covariance theorem are not separately exported here.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.citation_hc1_exact_variance", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L177", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:5cc1b2741de12ff08aa9f0d39472dfa56e9f11744937f86036994fb079d859e5", "conditional_on": [], "credit_reason": "The arithmetic component is exact. The target’s statistical interpretation using normal or Student-t critical values is not concluded by these rows; this is reflected in target coverage, not charged as a failure of the expressly arithmetic row.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_citation_standard_error", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L205", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:5cc1b2741de12ff08aa9f0d39472dfa56e9f11744937f86036994fb079d859e5", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T7", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T7" }, "html": { "body": "

S11 File, Table A (rendered p. 1); main text p. 4/8 (PDF p. 4). The four binary-outcome cells give the stated OLS fit and HC1 variance. The standard error is 0.06446... and rounds to 0.064. The erroneous printed statistic has a two-sided p value rounding to 0.62 and is not significant at 10%, under either the normal or Student t(197) reference. The corrected statistic exceeds both genuine 1% critical points, retaining three stars. The usual measure-theoretic conditions are assumed. These are reference calculations, not an assertion of an exact sampling law for robust OLS.

", "headline": "S11 File, Table A (‘OLS Estimates on Survival at 6 Hours and 12 Hours’), second column (6 hrs; Studies 1,4; N 200): the standard error of the Cited coefficient, printed ‘(0.645)’, corrected to (0.064)", "location": "S11 File, Table A (rendered p. 1; the file has one page); the main article cites the table on p. 4/8 (PDF p. 4)" }, "id": "P041-T7", "text": { "body": "S11 File, Table A (rendered p. 1); main text p. 4/8 (PDF p. 4). The four binary-outcome cells give the stated OLS fit and HC1 variance. The standard error is 0.06446... and rounds to 0.064. The erroneous printed statistic has a two-sided p value rounding to 0.62 and is not significant at 10%, under either the normal or Student t(197) reference. The corrected statistic exceeds both genuine 1% critical points, retaining three stars. The usual measure-theoretic conditions are assumed. These are reference calculations, not an assertion of an exact sampling law for robust OLS.", "headline": "S11 File, Table A (‘OLS Estimates on Survival at 6 Hours and 12 Hours’), second column (6 hrs; Studies 1,4; N 200): the standard error of the Cited coefficient, printed ‘(0.645)’, corrected to (0.064)", "location": "S11 File, Table A (rendered p. 1; the file has one page); the main article cites the table on p. 4/8 (PDF p. 4)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T7" }, { "declarations": [ { "claim_revision_id": "catalog:a8ab5076a533e10cc4493b31784ab1b098158548de6a877cef933c96c581ba53", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.corrected_star_wald_bounds", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L790", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:a8ab5076a533e10cc4493b31784ab1b098158548de6a877cef933c96c581ba53", "conditional_on": [], "credit_reason": "The target’s own mathematical content explicitly specifies the strongest-attained-tier convention and the normal Wald reference. This is a proof of that correction, not proof that the first three printed weaker nested cutoffs are false.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.corrected_normal_star_tiers_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GP041P042P045B1.lean#L458", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:a8ab5076a533e10cc4493b31784ab1b098158548de6a877cef933c96c581ba53", "conditional_on": [], "credit_reason": "The supplied p ranges are legitimate premises of the expressly auxiliary logical implication; R241-13 separately establishes them for the four relevant normal p values. conditional_on is empty: no published theorem is being imported as an assumption.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.nested_significance_reading", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L808", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:a8ab5076a533e10cc4493b31784ab1b098158548de6a877cef933c96c581ba53", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T8", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T8" }, "html": { "body": "

S6 File, Tables B and E (rendered pp. 2–3), and S9 File, Table A (rendered p. 1); main text p. 4/8 (PDF p. 4). All four rounding rectangles have the stated Wald and Gaussian p enclosures, central approximations, and strongest tiers 3, 2, 2, 2. The genuine normal critical points have the stated approximations and give exactly the same tier rule. Under nested cutoffs the first three printed weaker levels hold; the fourth printed 1% assertion fails. The usual measure-theoretic conditions are assumed. Decimal ellipses are approximations with explicit errors.

", "headline": "The significance markers of four cells of the Cox tables: S6 Table B, Positive (‘0.670**’); S6 Table E, Cited, Study 5 (1st Half) (‘0.480*’) and Positive, Study 5 (2nd Half) (‘0.274*’); S9 Table A, Positive X NE Region, first column (‘0.316***’); corrected to ***, **, ** and **", "location": "S6 File, Table B (rendered p. 2) and Table E (rendered p. 3); S9 File, Table A (rendered p. 1); the main article cites S6 and S9 on p. 4/8 (PDF p. 4)" }, "id": "P041-T8", "text": { "body": "S6 File, Tables B and E (rendered pp. 2–3), and S9 File, Table A (rendered p. 1); main text p. 4/8 (PDF p. 4). All four rounding rectangles have the stated Wald and Gaussian p enclosures, central approximations, and strongest tiers 3, 2, 2, 2. The genuine normal critical points have the stated approximations and give exactly the same tier rule. Under nested cutoffs the first three printed weaker levels hold; the fourth printed 1% assertion fails. The usual measure-theoretic conditions are assumed. Decimal ellipses are approximations with explicit errors.", "headline": "The significance markers of four cells of the Cox tables: S6 Table B, Positive (‘0.670**’); S6 Table E, Cited, Study 5 (1st Half) (‘0.480*’) and Positive, Study 5 (2nd Half) (‘0.274*’); S9 Table A, Positive X NE Region, first column (‘0.316***’); corrected to ***, **, ** and **", "location": "S6 File, Table B (rendered p. 2) and Table E (rendered p. 3); S9 File, Table A (rendered p. 1); the main article cites S6 and S9 on p. 4/8 (PDF p. 4)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T8" }, { "declarations": [ { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "The generalization loses no target case: choosing the first basis vector recovers the intercept change and reversal of every other coefficient. Empty finite types are allowed but do not make the theorem vacuous; no covariance occurs in this declaration.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.ols_complement_coefficients", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L40", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "The target specialization has an ordinary first-column intercept and n>k. The theorem explicitly exports all three conclusions its row claims; equality of standard errors is an immediate consequence of equality of covariance diagonals.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.ols_complement_residuals_and_covariance", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L52", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "A genuine auxiliary identity in the target argument. Its hypotheses do not assume a treatment-slope result; full rank is explicitly part of T9(i) and holds for the fixed models in T9(ii).", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.ols_normal_equations", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L74", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "Auxiliary calculation for precisely the specified model P, not a claim about arbitrary sample sizes or every balanced design. The normal-equation premise is the advertised bridge rather than the slope conclusion.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.positive_balanced_slope", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L89", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "Auxiliary fixed normal-equation calculation for model C, not an independent certification of data extraction; its matrix and design sizes are specified by the target.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.cited_balanced_slope", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L120", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "Auxiliary two-group normal-equation calculation for the stipulated 75/76 design; the target supplies the empirical interpretation of sT and sB.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.former_two_group_slope", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L130", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "conditional_on": [], "credit_reason": "{'source': 'adjudication', 'decision': 'rejected', 'what': \"P041/P041_OLS.lean:249 gives the target's exact arm-survival differences and verifies nonzero values, agreement with printed precision, and failure of that enclosure for their negatives. T9(ii) explicitly gives the same former-senator counts (72,75) and coefficient −51/1900 at 6, 12 and 24 hours, so storing that pair once loses no numerical conclusion among the twelve rows. The arm denominators and rounding tolerance match the supplied claim and standing assumptions; this component is arithmetic on stipulated counts, not a claim to verify their extraction from the data.\"}", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.s11_survival_slopes_and_not_removal", "manifest_use": "not_used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_OLS.lean#L238", "status": "proved", "support_scope": "historical declaration", "used_in_solution": false } ], "formalization": { "claim_revision_id": "catalog:cde04615b3994006805b1509be8398ea12a54007c2610572180f397398fcab01", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T9", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T9" }, "html": { "body": "

S11 File, first paragraph and Tables A and B (rendered p. 1); main text p. 4/8 (PDF p. 4). Complementing an outcome reverses all slopes and residuals and preserves the robust meat. HC1 covariances and standard errors agree whenever their divisor exists. The stated balanced designs imply full rank and the difference-of-arm survival slopes, reproducing all twelve entries and excluding their removal interpretation. The six-hour full vectors and the initialization/replacement rules give the stated coding correction. The usual measure-theoretic conditions are assumed; the sample-size guard applies only to HC1.

", "headline": "S11 File, opening paragraph, second sentence (‘These results, presented in Tables A and B estimate the likelihood that an edit was removed by that point in time’), corrected: the outcome of Tables A and B is survival beyond the horizon, and the printed coefficients are the negatives of the removal coefficients", "location": "S11 File, first paragraph and Tables A and B (rendered p. 1); the main article cites the tables on p. 4/8 (PDF p. 4)" }, "id": "P041-T9", "text": { "body": "S11 File, first paragraph and Tables A and B (rendered p. 1); main text p. 4/8 (PDF p. 4). Complementing an outcome reverses all slopes and residuals and preserves the robust meat. HC1 covariances and standard errors agree whenever their divisor exists. The stated balanced designs imply full rank and the difference-of-arm survival slopes, reproducing all twelve entries and excluding their removal interpretation. The six-hour full vectors and the initialization/replacement rules give the stated coding correction. The usual measure-theoretic conditions are assumed; the sample-size guard applies only to HC1.", "headline": "S11 File, opening paragraph, second sentence (‘These results, presented in Tables A and B estimate the likelihood that an edit was removed by that point in time’), corrected: the outcome of Tables A and B is survival beyond the horizon, and the printed coefficients are the negatives of the removal coefficients", "location": "S11 File, first paragraph and Tables A and B (rendered p. 1); the main article cites the tables on p. 4/8 (PDF p. 4)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T9" }, { "declarations": [ { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "conditional_on": [], "credit_reason": "The arbitrary real treatment parameter includes the target’s binary treatment case, and the target explicitly asks for the predictor identity for all real parameters. The optional transformation of the full covariance matrix is not among this target’s numbered mathematical components.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.year_tenure_reparameterization", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Tenure.lean#L27", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "conditional_on": [], "credit_reason": "The same universally quantified exponential identity applies with the S10 main coefficient b substituted for g.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.tenure_multiplier_is_reciprocal", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Tenure.lean#L41", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.printed_tenure_reciprocal_enclosures", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Tenure.lean#L60", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "conditional_on": [], "credit_reason": "{'source': 'adjudication', 'decision': 'rejected', 'what': 'P041/P041_Tenure.lean:83 assumes V≥0 and s=exp(g)√V, and concludes the reciprocal standard-error identity and equality of the two squared Wald expressions. These are precisely the algebraic identities of T10(iv), with r=exp(g) and V representing Var(g); nonnegativity is inherent in a variance. Allowing V=0 adds a conventional algebraic boundary case without excluding any ordinary V>0 Wald test, so the theorem is faithful with a boundary note.'}", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.reciprocal_standard_error_and_wald", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Tenure.lean#L83", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "conditional_on": [], "credit_reason": "Fidelity sweep: ok", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.year_tenure_exact_instance", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Tenure.lean#L100", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:0457de1aeb8104975ebf61c0c1f53e326ab8e0c137799054544d37838a589e04", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T10", "decision": "whole", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T10" }, "html": { "body": "

S9 File, Table A (rendered p. 1) and Table B (rendered p. 2); S10 File (rendered p. 1); the same covariate label occurs in the notes to S6, S9 and S11 and on p. 4/8 (PDF p. 4). The fitted covariate is first-election year. Replacing it by tenure preserves the predictor and fitted hazards after transforming the year, treatment and interaction coefficients and baseline. The complete coefficient covariance transforms by congruence, with every other coefficient retained. The five printed ratios reverse direction through reciprocals, with the stated uniform enclosures; the two 1.056 reciprocals round to 0.947. Reciprocal delta-method standard errors obey s/r^2 even at zero variance. Where a Wald test is defined, its squared statistic, two-sided normal p value and cutoff decisions are preserved; both S9 Table B rounding rectangles retain three stars under this Wald reading. The exact 132/125 illustration has the stated hazards and tenure values. The usual measure-theoretic conditions are assumed; finite second moments are explicit for covariance, and positive variance is required only for the Wald test. Hazard ratios formed by division require a nonzero baseline.

", "headline": "S9 File, Table A, row ‘Positive X Length of Incumbency’ (0.989, 0.981) and Table B, row ‘Cited X Length of Incumbency’ (1.056***, 1.056***); S10 File, row ‘Incumbency Length’ (1.004); corrected: the fitted covariate is the year of first election, so each printed ratio is per year of first election and its reciprocal is the ratio per year of tenure", "location": "S9 File, Table A (rendered p. 1) and Table B (rendered p. 2); S10 File (rendered p. 1); ‘length of incumbency’ also appears in the notes of the tables of S6, S9 and S11 and on p. 4/8 (PDF p. 4)" }, "id": "P041-T10", "text": { "body": "S9 File, Table A (rendered p. 1) and Table B (rendered p. 2); S10 File (rendered p. 1); the same covariate label occurs in the notes to S6, S9 and S11 and on p. 4/8 (PDF p. 4). The fitted covariate is first-election year. Replacing it by tenure preserves the predictor and fitted hazards after transforming the year, treatment and interaction coefficients and baseline. The complete coefficient covariance transforms by congruence, with every other coefficient retained. The five printed ratios reverse direction through reciprocals, with the stated uniform enclosures; the two 1.056 reciprocals round to 0.947. Reciprocal delta-method standard errors obey s/r^2 even at zero variance. Where a Wald test is defined, its squared statistic, two-sided normal p value and cutoff decisions are preserved; both S9 Table B rounding rectangles retain three stars under this Wald reading. The exact 132/125 illustration has the stated hazards and tenure values. The usual measure-theoretic conditions are assumed; finite second moments are explicit for covariance, and positive variance is required only for the Wald test. Hazard ratios formed by division require a nonzero baseline.", "headline": "S9 File, Table A, row ‘Positive X Length of Incumbency’ (0.989, 0.981) and Table B, row ‘Cited X Length of Incumbency’ (1.056***, 1.056***); S10 File, row ‘Incumbency Length’ (1.004); corrected: the fitted covariate is the year of first election, so each printed ratio is per year of first election and its reciprocal is the ratio per year of tenure", "location": "S9 File, Table A (rendered p. 1) and Table B (rendered p. 2); S10 File (rendered p. 1); ‘length of incumbency’ also appears in the notes of the tables of S6, S9 and S11 and on p. 4/8 (PDF p. 4)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T10" }, { "declarations": [ { "claim_revision_id": "catalog:427a0b8111942279f28a729806268b32945ee5a5d2193c1ac44faf1c940bafea", "conditional_on": [], "credit_reason": "This is the auxiliary |z|<1 part of P1, not a proof that the former-senator effect is zero or that bias is confined to active politicians.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.former_subgroup_nondetections", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Wald.lean#L820", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:427a0b8111942279f28a729806268b32945ee5a5d2193c1ac44faf1c940bafea", "conditional_on": [], "credit_reason": "P1 numerical reconstruction only. The living centre gives approximately .69452; the interval of possible values permits a p value rounding to the printed .70, but the theorem correctly does not assert that the centre itself rounds to .70. Target decimal z and p expressions are understood as numerical approximations, not literal real equalities.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.Wave1.printed_former_normal_p_enclosures_proof", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/Wave1/Gap/GLP041P042P045B1.lean#L231", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:427a0b8111942279f28a729806268b32945ee5a5d2193c1ac44faf1c940bafea", "conditional_on": [], "credit_reason": "Historical excluded association.", "credit_status": null, "fidelity_status_baseline": null, "lean_name": null, "manifest_use": "not_applicable", "named_axioms": [], "says": null, "source": null, "status": "not_applicable", "support_scope": "historical declaration", "used_in_solution": false } ], "formalization": { "claim_revision_id": "catalog:427a0b8111942279f28a729806268b32945ee5a5d2193c1ac44faf1c940bafea", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T11", "decision": "restricted", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T11" }, "html": { "body": "

p. 5/8 (PDF p. 5); Abstract, p. 1/8 (PDF p. 1). The three former-senator table pairs have the displayed Gaussian p-value enclosures and central approximations, with absolute Wald statistics below one.

", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’ (the whole paragraph), with the last sentence of the first paragraph of the Discussion and the last two sentences of the Abstract; as printed", "location": "p. 5/8 (PDF p. 5); Abstract, p. 1/8 (PDF p. 1)" }, "id": "P041-T11", "text": { "body": "p. 5/8 (PDF p. 5); Abstract, p. 1/8 (PDF p. 1). The three former-senator table pairs have the displayed Gaussian p-value enclosures and central approximations, with absolute Wald statistics below one.", "headline": "‘Replication on Wikipedia Pages of Dead and Retired Senators’ (the whole paragraph), with the last sentence of the first paragraph of the Discussion and the last two sentences of the Abstract; as printed", "location": "p. 5/8 (PDF p. 5); Abstract, p. 1/8 (PDF p. 1)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T11" }, { "declarations": [ { "claim_revision_id": "catalog:a002997702fd0cec9b10e3ff0001fb2f990bf281f323e4688aaee38134f3b5be", "conditional_on": [], "credit_reason": "Precisely T12(P1) and (P3), with no claim that the symmetric percentages are directional relative increases.", "credit_status": "ok", "fidelity_status_baseline": "ok", "lean_name": "AronowAudit.P041.printed_percentages_arithmetic_and_direction", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_Risks.lean#L104", "status": "proved", "support_scope": "historical declaration", "used_in_solution": true }, { "claim_revision_id": "catalog:a002997702fd0cec9b10e3ff0001fb2f990bf281f323e4688aaee38134f3b5be", "conditional_on": [], "credit_reason": "{'source': 'adjudication+second review', 'adjudicator': \"P041/P041_PrintedRefutations.lean:7 concludes only ¬(A ∧ B ∧ C), where A, B and C are the directional rounding assertions for 36%, 29% and 64%. T12(P2) says ‘All three are false’, which requires ¬A ∧ ¬B ∧ ¬C; the exported statement establishes only that at least one assertion fails and identifies none individually. The reviewer's logical observation is correct, but treating the omitted individual refutations as a minor note understates the loss of the target's conclusions. Calling the row an exact refutation of the recorded conjunction does not supply those conclusions, and the stronger imported theorem in P041_Risks.lean does not strengthen this declaration's statement.\", 'suggested_fix': 'Replace the outer negation of the conjunction with a conjunction of the three individual negations, retaining the same relativeIncrease arguments and printed integers. The required individual refutations are already conjuncts of risk_percentages_counterexample in P041/P041_Risks.lean.', 'second_review': 'P041/P041_PrintedRefutations.lean:7 concludes only ¬(A ∧ B ∧ C), whereas target T12(P2) explicitly says “All three are false,” requiring the three individual negations for this component. The definitions of relativeIncrease at P041/P041_Risks.lean:17 and roundsWhole at P041/P041_Risks.lean:25 match the target’s directional reading and rounding convention, so this is an omitted conclusion. The omission is minor because each missing negation is a trivial closed rational calculation: 575/13 > 73/2, 100/3 > 59/2, and 375/4 > 129/2. Strengthen the declaration to a conjunction of individual negations, already stated at P041/P041_Risks.lean:37, P041/P041_Risks.lean:38 and P041/P041_Risks.lean:39; importing that theorem does not strengthen this declaration’s own statement.', 'severity': 'minor'}", "credit_status": "narrower", "fidelity_status_baseline": "narrower", "lean_name": "AronowAudit.P041.printed_directional_percentages_refuted", "manifest_use": "used", "named_axioms": [], "says": null, "source": "https://github.com/pmaronow/llm-audit-lean/blob/b9e9df44831afed4f2b8f63be5c0cc9169a5fda4/AronowAudit/P041/P041_PrintedRefutations.lean#L7", "status": "counterexample_proved", "support_scope": "limited", "used_in_solution": true } ], "formalization": { "claim_revision_id": "catalog:a002997702fd0cec9b10e3ff0001fb2f990bf281f323e4688aaee38134f3b5be", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T12", "decision": "refuted", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T12" }, "html": { "body": "

Abstract p. 1/8 (PDF p. 1); Effects of Citations on Survival, pp. 4/8–5/8 (PDF pp. 4–5). The printed claim is false: under the stated directional reading of “more likely than”, each of 36%, 29% and 64% fails separately. The exact directional increases are 575/13, 100/3 and 375/4, rounding to 44%, 33% and 94%. The symmetric percentages do round to the printed numbers and all three removal comparisons have the stated direction. The usual measure-theoretic conditions are assumed. This refutation concerns the three directional percentages.

", "headline": "Abstract, the sentence giving ‘36%’ and ‘29%’, and ‘Effects of Citations on Survival’, the sentence giving ‘64%’; as printed (headline claims P041-H2 and P041-H3)", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5)" }, "id": "P041-T12", "text": { "body": "Abstract p. 1/8 (PDF p. 1); Effects of Citations on Survival, pp. 4/8–5/8 (PDF pp. 4–5). The printed claim is false: under the stated directional reading of “more likely than”, each of 36%, 29% and 64% fails separately. The exact directional increases are 575/13, 100/3 and 375/4, rounding to 44%, 33% and 94%. The symmetric percentages do round to the printed numbers and all three removal comparisons have the stated direction. The usual measure-theoretic conditions are assumed. This refutation concerns the three directional percentages.", "headline": "Abstract, the sentence giving ‘36%’ and ‘29%’, and ‘Effects of Citations on Survival’, the sentence giving ‘64%’; as printed (headline claims P041-H2 and P041-H3)", "location": "Abstract, p. 1/8 (PDF p. 1); ‘Effects of Citations on Survival’, sentence running from p. 4/8 to p. 5/8 (PDF pp. 4–5)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T12" }, { "declarations": [ { "claim_revision_id": "catalog:48f6cb936f71650b3cddc74ff742d0e4e151b96a0ac95f68d2ddcdefd90775a7", "conditional_on": [], "credit_reason": "Historical excluded association.", "credit_status": null, "fidelity_status_baseline": null, "lean_name": null, "manifest_use": "not_applicable", "named_axioms": [], "says": null, "source": null, "status": "not_applicable", "support_scope": "historical declaration", "used_in_solution": null } ], "formalization": { "claim_revision_id": "catalog:48f6cb936f71650b3cddc74ff742d0e4e151b96a0ac95f68d2ddcdefd90775a7", "crosswalk_url": "../../data/lean-crosswalk.json#/targets/P041-T13", "decision": "omitted", "release_id": "sap-2026-11-08-assured-release", "target_id": "P041-T13" }, "html": { "body": "

No current theorem is supplied for this target.

", "headline": "S1 File, ‘Description of implementation’, last paragraph, and S5 File, ‘Description of implementation’, last four sentences; as printed", "location": "S1 File (rendered p. 2); S5 File (rendered p. 2); the main article refers to these changes on p. 3/8 (PDF p. 3)" }, "id": "P041-T13", "text": { "body": "No current theorem is supplied for this target.", "headline": "S1 File, ‘Description of implementation’, last paragraph, and S5 File, ‘Description of implementation’, last four sentences; as printed", "location": "S1 File (rendered p. 2); S5 File (rendered p. 2); the main article refers to these changes on p. 3/8 (PDF p. 3)" }, "url": "https://pmaronow.github.io/llm-audit/works/P041/#P041-T13" } ] }, "missing_sources": [ "EGAP preregistration plan and addenda referenced by the article and S1-S5 (not an official supplement; all twelve official supplements are present)" ], "reproduction": { "agreement": "agrees_in_part", "status": "full", "targets": [ { "agreement": "agrees_in_part", "location": "p. 1 (PDF p. 1), Abstract", "method": "I translated the Stata calculations into R. The code’s symmetric percentage differences reproduce 36% and 29% to rounding, but the relative-risk increases implied by “more likely” are 44.23% and 33.33% (P041-E001). The 12-hour count also depends on the time conversion corrected in P041-E002.", "status": "reproduced", "target": "Abstract: removal risk at 12 hours and 3 days" }, { "agreement": "to_rounding", "location": "p. 4 (PDF p. 4)", "method": "I ran the released R plotting code with file-path and graphics changes, including clipping the display at 250 hours. The curves, step locations and labels agree visually; all eight endpoint heights agree within an absolute read-off tolerance of 0.02. Here “To rounding” refers to graphical reading precision, not exact numerical annotations.", "status": "reproduced", "target": "Figure 1" }, { "agreement": "to_rounding", "location": "p. 4 (PDF p. 4), Survival of Positive and Negative Facts on Wikipedia", "method": "I translated the Stata Cox model into R, preserving Breslow ties, the supplied time and censoring conventions, and the robust-variance correction. The negative-versus-positive hazard ratio 1.58997 and p-value 0.00134 round to 1.59 and 0.001.", "status": "reproduced", "target": "Active senators: pooled Cox effect" }, { "agreement": "agrees_in_part", "location": "pp. 4-5 (PDF pp. 4-5), Effects of Citations on Survival", "method": "I translated the Stata calculations into R. The full-follow-up hazard ratio 2.32977 and the code’s symmetric percentage difference reproduce 2.33 and 64% to rounding. The relative-risk increase at four hours is 93.75%, and the Cox model is not restricted to four hours (P041-E001).", "status": "reproduced", "target": "Citations: removal risk and Cox effect" }, { "agreement": "agrees_in_part", "location": "p. 5 (PDF p. 5), Effects of Citations on Survival", "method": "I translated the adjusted Stata Cox model into R. The commands reproduce the printed interaction hazard ratio $0.32$ and $p<0.001$ to rounding, but the model controls for the seconds within the minute rather than the hour of entry. With the hour of entry, the interaction is $0.28$ (P041-E003). The model also omits the influence covariate listed in the table note (P041-E009).", "status": "reproduced", "target": "Citation effect moderated by 2014 reelection" }, { "agreement": "to_rounding", "location": "p. 5 (PDF p. 5), Replication on Wikipedia Pages of Dead and Retired Senators", "method": "I translated the Study 3 Cox models into R with the Stata-style robust-variance correction. The living, deceased and pooled p-values round to 0.70, 0.47 and 0.42. This reproduces the numerical nondetections, not the inference that bias is absent or limited to active politicians (P041-E006).", "status": "reproduced", "target": "Retired and deceased senators: null effects" }, { "agreement": "to_rounding", "location": "p. 3 (PDF p. 3), Details on the Facts", "method": "I translated the Stata data preparation into R and recalculated the relevance means and ordinary rating-level standard errors. They round to 5.0 and 0.04 in each group. Worker-clustered subgroup standard errors are about 0.07; all five study waves enter these calculations (P041-E018).", "status": "reproduced", "target": "Fact relevance validation" } ] }, "summary": "The paper reports four field experiments, and a partially aborted fifth, that added positive or negative facts described as true, cited or uncited, to the Wikipedia pages of U.S. senators and recorded the time to removal. It finds that negative facts are removed faster than positive ones on the pages of sitting senators (hazard ratio 1.59). 20 entries are classified as errors (1 Major, 2 Moderate, 17 Minor), and 2 are clarity notes. The most consequential is \\fid{P041-E006}: Study 3 (hazard ratio 0.69, 95\\% interval 0.28 to 1.70) does not show that the bias is limited to active politicians.", "what_was_checked": "I checked the following points against the replication data and programs. This is not a statement that the rest of the paper is correct.\n\\begin{itemize}\n\\item The pooled hazard ratios of Table A of S6 File, $0.629$ $(0.091)$ and $0.429$ $(0.069)$, reproduce, and the values 1.59 ($p=0.001$) and 2.33 ($p<0.001$) of the main text are their reciprocals after rounding. The pooled difference between positive and negative facts remains with standard errors clustered by senator. It also remains without the Study 1 edits of the last day and with Study 4 censored at the start of Study 5.\n\\item Table C of S6 File (Study 3) and the $p$-values $0.42$, $0.70$ and $0.47$ of the main text reproduce; for the conclusion drawn from them, see \\fid{P041-E006}.\n\\item The Cox estimates checked in S6--S10 Files reproduce to their printed precision from the commands as run. The exceptions and qualifications are in \\fid{P041-E003}, \\fid{P041-E008}, \\fid{P041-E009} and \\fid{P041-E012}. One entry of Table B of S7 File (printed $0.386$) is $0.389$ in the re-estimate, a difference in the third decimal that I could not settle without Stata.\n\\item With Study 5 included, the pooled hazard ratio for positive facts is $0.563$, with separate study effects for its two halves. Adding only its first half to the citation analysis gives a hazard ratio of $0.454$ for cited facts. These estimates preserve the direction and significance of the pooled results.\n\\item The coefficients and robust standard errors of Tables A and B of S11 File reproduce, one standard error aside (\\fid{P041-E011}; for the description of the outcome, see \\fid{P041-E010}).\n\\item The treatment counts and assignment patterns in the data match those described. Study 1 has 25 blocks of four, with two positive and two cited facts in each. The valence assignments in Studies 2 and 5 reverse those in Studies 1 and 4, respectively. The first 50 edits of Study 5 have all four conditions and the last 50 only cited facts, and the edit dates are those stated in S1--S5 Files.\n\\item The Kaplan--Meier values recomputed from the data match the three panels of Fig 1 within the 250 hours shown.\n\\item In S12 File, the following reproduce: the 237 of 245 quality-control responses, the 551 facts rated five times each, the entries of Table A, the six Fleiss kappas, and the relevance means of the main text (5.0 and 5.0, with unclustered standard error 0.04). See \\fid{P041-E013}, \\fid{P041-N015} and \\fid{P041-E018}.\n\\item The two example facts quoted on p.~3 are in the archived list of facts. The Harvard and Yale archives hold identical data, and only the main program differs between them.\n\\end{itemize}\nThe following limits apply.\n\\begin{itemize}\n\\item The pre-analysis plan registered with EGAP and its addenda were not available. Statements that an analysis was ``as specified in our registered pre-analysis plan'' are unchecked.\n\\item The sources do not say when observation of the facts that were never removed ended, so I could not check the common censoring time that the Cox models give them.\n\\item The revision histories of the Wikipedia pages were not consulted. Who removed each fact, from which account each edit was made, and why, are not in the data.\n\\item The truth of each inserted fact was not checked against the news reports, and the summaries of the literature were not checked.\n\\item No Stata run was possible. The Stata programs were re-estimated in R and Python, and the embedded figures of S6 and S12 Files were seen only in approximate renderings.\n\\item Two XML codebooks named in the Harvard metadata are not in either archive.\n\\item I did not reproduce every cell or the likelihood-ratio tests of the balance tables in S1--S5 Files.\n\\item I read the Discussion’s suggestion that such biases may build over time as a possibility. I found no estimate of accumulation or downstream public perceptions in the paper.\n\\end{itemize}", "work": { "arxiv_id": null, "audit_complete": true, "authors": [ "Joshua L. Kalla", "P. M. Aronow" ], "category": "Articles - Empirics", "citation": "Joshua L. Kalla and P. M. Aronow (2015). Editorial Bias in Crowd-Sourced Political Information. PLOS ONE 10(9):e0136327.", "doi": "10.1371/journal.pone.0136327", "edition": { "pages_in_audited_file": 8, "sha256": "9e916021e736e531a7833aac2bb408c78f91570d48fdbec9c4b2faf8726ab138", "source_url": "https://doi.org/10.1371/journal.pone.0136327", "venue": "PLOS ONE", "year": "2015" }, "id": "P041", "later_materials": null, "title": "Editorial Bias in Crowd-Sourced Political Information", "type": "Journal article", "url": "https://pmaronow.github.io/llm-audit/works/P041/", "venue": "PLOS ONE", "year": "2015" } }