Naive before/after measurement of SXG is swamped by seasonal bias, low SXG share of traffic, and the fact that SXG users are biased toward top SERP positions (window.isSXG tags them but they aren't comparable to normal visitors). Instead, compare SXG users with successful prefetching to SXG users with broken prefetching, using a canary sub-resource detected from JavaScript, and report via analytics. For LCP, compare normal traffic to SXG traffic with loaded sub-resources; for conversion, compare SXG with and without intentionally broken prefetching.