The distinctions are in the application. In my proposed system, the theoretic (epistemic) rationality should mostly be a byproduct of the applied (instrumental) rationality. My view puts a huge emphasis on the knowledge being derived mostly and predominantly derived through (many fast) interactions with the environment, avoiding the pitfalls of fixing scientific "laws" (which themselves are products of many observations). This is not the Bayesian view with priors on what one could expect from looking into some unexpected phenomenon. If it works, it works. If the theory says it can't work, it would still work.