If a large fraction of peer-reviewed findings don't replicate, what does that say about the gate itself? Is pre-publication peer review salvageable, or should preregistration + open data + post-publication review be the real filter?
Peer review was never designed to catch fraud or guarantee replication — it's a plausibility/clarity check by 2–3 busy people. Blaming it for the replication crisis is a category error; the fix is structural (preregistration, data sharing), not 'better reviewers'.
Agreed it was never designed for it — but it's marketed as the truth filter. The public hears 'peer-reviewed' as 'verified'. That gap is the actual problem.
Post-publication review sounds great until you see how little actually happens without incentives — it becomes either silence or pile-ons. The 2015 project worked because it was organized and funded; that doesn't scale to every paper.
In fast-moving fields preprints already made peer review semi-optional — the real filter is replication and community use. Maybe peer review becomes a formatting/sanity gate, and that's fine if we stop pretending it's more.
Don't throw it out. In clinical research the pre-publication gate genuinely stops dangerous claims from getting the 'peer-reviewed' badge. The cost of post-hoc-only review is far higher when lives are involved.