Robustness of evidence reported in preprints during peer review