The reproducibility crisis in science is conventionally understood as a methodological problem amenable to methodological solutions---pre-registration, transparency, replication. This paper argues it is an engineering problem. The institutional conflation of hypothesis generation and empirical verification creates structural conditions that maximise confirmation bias, suppress replication, sustain authority gradients, and exclude most of the world's researchers from meaningful participation in knowledge production. Software engineering confronted the identical problem---that developers who build a system cannot reliably verify it---and solved it through institutional separation, codified in standards (IEEE 1012 and its descendants) developed, tested, and revised over half a century. Science has the same problem, worse structural conditions for self-correction, and no comparable institution. The paper identifies a second, epistemological root: the reform movement's exclusive reliance on the hypothetico-deductive model marginalises abduction---the generation of explanatory hypotheses from surprising observations---which is the mode of reasoning through which Newton, Darwin, Watson and Crick, and Akira made the discoveries that define their fields. By treating hypothesis generation as a pre-scientific preliminary, the current institutional architecture suppresses the upstream process whose weakness may explain why so many published findings cannot survive independent testing. Open Experiment Design (OED) is proposed as a structural model that separates the roles of theorist and verifier, thereby protecting abductive reasoning while ensuring independent verification. The model requires no funding, no institutional permission, and no central coordination---features that make it available to any researcher with a testable hypothesis.
Ryuhei Ishibashi (Thu,) studied this question.