In 2015, the Open Science Collaboration tried to replicate 100 published psychology studies. Only 36 held up. The number sent shockwaves through the field — but anyone who had spent time in research already knew something was wrong.
Since then, similar patterns have surfaced in cancer biology, economics, machine learning, and drug discovery. Bayer reported in 2011 that its internal teams could reproduce only 20-25% of published preclinical findings. Amgen fared even worse: 6 out of 53 landmark studies replicated.
The default response is to blame scientists. Fraud. Sloppiness. Cutting corners. But this misdiagnoses the problem. Most researchers are honest people operating inside a machine designed to produce the wrong outputs.
Here is the machine: funding depends on publications. Publications depend on positive results. Positive results depend on hypotheses that happen to be correct. The system selects for researchers who can generate a steady stream of clean, positive, publishable findings — and punishes anyone who reports messy reality.
This is not a character flaw. It is an incentive flaw.
When survival depends on producing positive results, researchers unconsciously make methodological choices that inflate them. They stop experiments when results look good. They test multiple hypotheses and report the winners. They exclude inconvenient data points. Each step is rational within the incentive structure. The aggregate effect is a literature that systematically overstates what is true.
Fixing this requires changing what the system rewards. Not asking scientists to be braver. Not adding ethics training. Restructuring funding so that methodological rigor, transparent reporting, and honest null results receive the same support as flashy positive findings.
A DeSci DAO approach begins from this insight: build evaluation around what research actually contributes after publication, where replication attempts, negative results, and methodological soundness are visible and valued. Remove the survival pressure to produce clean narratives, and the literature starts reflecting reality.
The question: if you could redesign what gets rewarded in science, what single change would reduce the replication problem most?
Comments
0 approved