Science must be able to repeat and reproduce.
The social sciences and broader domain of human sciences certainly needs more replication in terms of experimental results reported by original studies. But justification for an approved _X_ in overall science isn't confined to literally repeating "on the table" its developmental stages if such are subsumed in an explanation of it.
Some disciplinary areas (like biological evolution) can't directly and immediately reproduce the asserted, lengthy origin process of a target of research; and thereby depend upon verifying the mechanisms and regularities that yielded it. Along with preserved evidence of the past affairs being intellectively coupled with the former.
Astronomical and cosmological endeavors depend heavily upon gathered data managed by tested and reliable principles -- and inferential and mathematical explorations -- because the objects/curiosities of interest are distant and not subject to "hands-on" examination. The successful predictive ability of outputted claims and abstract frameworks contributes to their revisable acceptance, and computer simulations may even offer a tentative semblance of replaying the past or modeling the future.
Supersensible intelligence, angels, etc are pre-conditionally ruled out of the explanatory arena by methodological naturalism; i.e., the latter is part of the constraints of how the "game" is played. It's the philosophical guideline orientation of science practice to proceed cataloging, describing, and explaining the Earth/universe with the latter's own interdependent content rather than via any traditional beliefs whose agencies are prior in rank to detectable entities. Science may project products of reason into its depiction of nature which do not correspond to concrete or observable residents. But those generalizations and technical description dependent concepts, while seemingly non-corporeal or noological furniture of a science field, validate themselves with their persistent usefulness and arguable indispensability.