Half of social-science studies fail replication test in years-long project (nature.com)

it's not just social science, even AI papers that would be trivial to replicate dont