A procedural-memory gap is a longitudinal failure mode occurring across independent task executions where an agent correctly diagnoses an operational failure in an earlier run but neither retains nor invokes the resulting lesson when identical risks recur. Rather than relying on model retraining or unindexed postmortem notes, harness scaling closes this gap by transforming validated postmortem insights into versioned, inspectable, and machine-executable practices—such as state-local constraints, activation rules, or recovery procedures—within an external runbook.