Before rushing to the conclusion that Kolmogorov-Smirnov tests 'materialize' statistical similarity, one must assess the risk of methodological deception.
What is at stake here is the validity of a proof, not merely the execution of a procedure.
If the underlying benchmark is flawed, such as using a normal distribution for venture capital returns that are intrinsically asymmetric, then applying the KS test, even if performed with rigor, will not guarantee a valid conclusion.
The downside of such an approach is relying on an illusory similarity, ignoring the actual tail risks that may occur.
For example, a KS test that validates the similarity between a venture capital portfolio's returns and a Gaussian benchmark model is false security because it masks the impact of extreme events that dominate real venture capital returns.