> But it's a particular research programme, embedded in a social context, that generates knowledge, not a single reproducible experiment.

Somewhat disagree. That particular research programme must still have reproducible experiment as a unit. A million irreproducible experiments will generate exactly zero knowledge.