the studies you can't read are the ones that would change your mind

Publication bias means the literature you can read is not a random sample of the research that was done.

Null results are less likely to be written up, less likely to be accepted, and slower to appear when they are. So the visible literature is enriched for positive findings by a mechanism that has nothing to do with whether the effects are real.

For this space specifically it's amplified: a lot of the work is done by groups with an interest in the compound, and there's no regulatory requirement to report anything.

Registries are the partial fix. A registration with no posted results is the closest thing to visible evidence of an invisible study. Publication bias itself is well documented in the meta-research literature, human and preclinical.