When researchers set out systematically to repeat well-known psychology experiments, fewer than half produced the original result. Similar exercises in cancer biology and in economics returned comparable figures. Outright fraud accounts for very little of this. The larger causes are structural: journals prefer surprising positive findings, careers depend on publication, and a researcher working through a rich dataset can test dozens of hypotheses while reporting only the one that reached significance. None of these practices feels dishonest from the inside. The response — pre-registering hypotheses, publishing raw data, funding replications — has been substantial, though uneven across fields. Uncovering the problem was itself an achievement of the scientific method, and sceptics who cite the crisis as proof that science cannot be trusted draw exactly the wrong conclusion.