Resolve the matched-volume safety difference between risk-targeted and random validation

Determine whether the safety difference between the risk-targeted validation policy {4} and the Random-Matched validation arm is statistically distinguishable when both operate at matched validation volume, thereby establishing whether validation targeting improves safety beyond random validation.

Background

The validation ablation compares policy {4}, which uses risk-targeted validation, with Random-Matched, which inserts validators at a matched random rate. The comparison is based on 60 instances and 120 requested deliverables per arm, with differences evaluated using scenario-bootstrap 95% intervals.

The reported interval for the difference in invalid releases per released deliverable does not resolve the safety comparison at the available sample size. Consequently, the evidence does not establish whether risk-targeted validation provides a safety advantage over random validation at matched validation volume, although policy {4} is reported to be cheaper.