On August 31, 2026, three professors who blog under the name Data Colada — Uri Simonsohn, Leif Nelson, and Joseph Simmons — posted an analysis of a paper behavioral science had stopped questioning. Dan Ariely and Klaus Wertenbroch's "Procrastination, Deadlines, and Performance: Self-Control by Precommitment," published in Psychological Science in 2002, is assigned reading in economics and psychology courses and carries more than two thousand citations on Google Scholar: give people binding deadlines instead of leaving them to police themselves, and their output improves. The evidence that its numbers had been tampered with came from no whistleblower and no confession. It came from a spreadsheet Ariely had emailed to a graduate student, Kyle Hyndman, in 2006, and from arithmetic.
The paper's second experiment gave sixty students three ten-page documents to proofread, each seeded with a hundred errors. Participants working to evenly spaced, externally imposed deadlines made an average of 136.1 corrections; those facing a single deadline on the last day made 71.1. The gap is a Cohen's d of 2.5, a bigger difference than the one between men's and women's heights. Stranger still, eighteen of the twenty participants in the last-day group had what Data Colada called a "Corrections Twin": another participant, exactly ten identification numbers away, who had logged the identical number of corrections on all three proofreading tasks. No twins appeared in either of the other two conditions. People do not pair off like that by accident. Copied and lightly altered spreadsheet rows do.
The original files survived. Hyndman kept them, and passed them to Data Colada seventeen years later, when he and Alberto Bisin began a replication of their own. What the spreadsheet gave up was a texture problem. Scores on the three near-identical proofreading tasks correlate between +.74 and +.90 in the replication, and between +.03 and +.27 in the original. An invented dataset can mimic an average. It cannot easily mimic the thousands of small, irregular correlations that come free with any real measurement of real people.
Wertenbroch, a professor of marketing at INSEAD who says he never had access to any version of the data, did not defend the paper. On July 23, three days after Data Colada sent the authors its findings, he asked the journal's editor-in-chief to retract — a month before any of it was public. The evidence of tampering, he told them, was convincing. It shows, he wrote, that much or all of the data — and therefore the results — are false. Ariely's statement was narrower: he had been made aware of serious anomalies, and neither his records nor his memory could answer the questions raised. Psychological Science retracted the paper on September 2, citing Data Colada's posts and the replication alike.
Coauthorship runs on a trust nobody examines until it breaks: you put your name on data you did not collect, trusting that your collaborator's numbers came from where he says. That assumption held for twenty-four years, through careers built partly on a result nobody had reason to doubt. What ended it was the discovery that reality has a shape invented numbers rarely match. You can fabricate a result. You cannot fabricate the arithmetic underneath it, and eventually somebody checks.