I found something in paper "2017 Validation of Molecular Simulation-An Overview of Issues" may be useful for the questions:

" Before addressing how to validate molecular simulation methodology, software and applications, one may ask whether validation of an answer to a question is always possible. This is definitively not the case. Some questions are based on unprovable premises or involve the choice of an arbitrary quantity as a reference. In other cases, a correlation between the values of two observable quantities measured under very different conditions may be due to hidden variables and not be directly related to the properties considered. It is easy to confuse correlation with causality. In general, molecular simulation is best used to address questions that are clear and concrete. In such cases the answers from simulations can be normally well validated. Often however, results of molecular simulation can only be partly or incompletely validated "

“ Validation of simulation results by comparison to experimental data is not straightforward. The amount of experimental data available for a particular system is limited, meaning the problem is often underdetermined. Experimental data, as well as the functions that connect a configuration to an observable quantity, are also subject to uncertainty and error. Finally, some quantities, the values of which are commonly considered to be experimental data, are in fact not directly observable. For example, the structure of a molecule in solution or a radial distribution function of a liquid cannot be observed directly. Such non-observable quantities should not be used as the sole means of validating a simulation. This is because they may represent the quality of the function or procedure, by which the value of the non-observable quantity was derived from values of observable quantities, rather than the quality of the simulation. Since experimental values are generally averages over space and time, a comparison with values calculated from a simulation should be based on comparable averaging. This implies that averaged values of a quantity should not be compared to instantaneous values of that quantity. Validation of molecular simulation studies would be enhanced if publishers, editors, reviewers and authors of manuscripts ensured that all information needed to reproduce the data was reported in the manuscript. Equally, funding agencies should encourage the repetition of research essential to maintain the integrity of the scientific literature. ”-cited from Validation of Molecular Simulation-An Overview of Issues

If interested, please also referred to "1998 Validation of molecular dynamics simulation".