The email from Utrecht arrived at 6:40 on a Tuesday morning, while Nadia Okonjo was still deciding whether to reheat yesterday's coffee. Dr Anneke Vos wrote with the elaborate courtesy that scientists reserve for bad news. Her laboratory had attempted a direct replication of Okonjo (2019), using four hundred participants and the original stimuli, and had found nothing at all. Nadia read the message twice. The premise underpinning nine years of her work — that handwriting, even glimpsed for a second, strengthens later recall — sat in the attachment's tables like a body on a slab, pale and entirely unmoving.
By nine she had both datasets open across two monitors, her office door shut against the corridor's cheerful noise. The Dutch effect size hovered somewhere near zero, with confidence intervals wide enough to drive a lorry through. Nadia went looking for a culprit. Perhaps their participants were older, or their screens slower, or their recall tests scored by a different rubric. She isolated one subgroup whose scores looked genuinely anomalous, felt a brief surge of hope, then checked the sample size: eleven people. Eleven. She had built entire lectures on less, and the thought turned her stomach over.
Marco Terzi, her doctoral student, found her still at the desk at two in the afternoon, barricaded behind printouts. His thesis rested on an extension of her paradigm; his funding ran out in March. "Protocol drift," he said hopefully. "They ran it online, didn't they?" Nadia drafted a reply to Vos that evening and attributed the null result to differences in presentation timing, screen resolution and the online setting. Every sentence was defensible. Every sentence was also, she suspected, a small act of cowardice. She saved the draft, did not send it, and walked home the long way along the canal.
Vos replied before breakfast and politely dismantled every excuse. The Dutch team had pre-registered their analysis, measured the refresh rate of each machine, and run a laboratory arm alongside the online one. The rigour of their design was, Nadia admitted to her kitchen ceiling, considerably greater than her own had been in 2018, when she was thirty-four, unfunded and desperate for a result before her contract expired. She remembered the weeks of testing participants alone at night. She remembered, rather less comfortably, how often she had stopped collecting data the moment the p-value slipped below .05.
The original analysis lived in a folder labelled STUDY3_FINAL_v7, which ought to have been warning enough. Nadia opened the script she had written six years earlier and read it like a stranger's confession. Line 212 removed trials with response times above two seconds. Line 213, added later and never reconciled with the line above it, removed everything below three hundred milliseconds — and, thanks to a misplaced bracket, inadvertently deleted a quarter of the control condition as well. She sat very still. The effect did not survive without that bracket. She checked four times, hoping arithmetic might develop a merciful streak.
She told Marco first, in the stairwell, because she could not bear to do it across a desk. He listened with his arms folded and asked, twice, whether the bracket might be defended as a principled exclusion. Nadia could have argued it; reviewers had accepted stranger things. Instead she heard herself concede that the finding was an artefact, that six years of citations rested on a typing error, and that his central chapter would need rebuilding from the studs. Marco said nothing for a long moment. Then he asked, quietly, whether she intended to tell the funder or wait.
Professor Halvorsen, who ran the department and its accounts with equal caution, received the news standing up. A public correction, he observed, would undermine not merely Nadia's standing but the institute's case for renewing a four-million-euro consortium grant on which three other laboratories depended. He proposed a quieter route: a short commentary, a footnote in next year's review, no drama. Nadia watched a gull land on the windowsill behind him. She thought of the doctoral students in Lisbon and Seoul who had built designs on her paradigm, and of how long a wrong idea survives when nobody says so aloud.
That night she wrote the retraction notice eleven times. The early drafts were crowded with context: the funding pressure, the loose norms of the era, the reviewers who had waved the paper through. By the fourth coffee she deleted all of it. The salient fact was one line long — the published effect depended on a coding error, and without that error there was no effect. Everything else was mitigation, and mitigation belonged in a different document, or possibly in a bar. She attached the corrected script, the raw files and Vos's data, and sent the package to the journal at 4:12 a.m.
The journal's editor answered within a week with a provisional expression of concern, pending review by two statisticians in Melbourne. Nadia found the wording strangely consoling. Nothing in the message pretended to be final; it simply held the paper up to the light and invited others to look. Social media, predictably, was less careful. A thread describing her as a cautionary tale collected nine thousand reposts, and an old rival gave an interview about standards. Marco forwarded her none of it. He was busy, he said, redesigning his second study, this time with a pre-registration and a sample size that would make Vos nod.
Eighteen months later Nadia stood in a lecture theatre in Utrecht, invited by Vos herself, and explained the bracket on slide four to two hundred graduate students. The consortium grant had survived, narrowly, restructured around a replication arm that Halvorsen now described in interviews as his own initiative. Her retraction had given unexpected impetus to a network of eleven laboratories that audited each other's analyses before submission, and Marco chaired its scripting group. Someone asked whether she regretted the paper. Nadia said she regretted the error and not the correction, which was the truest sentence she had managed in years.