A Contrastive Diagnostic of Target-Synchronization Sensitivity

We study whether the learned-function effects of delaying continuation-target synchronization by one update can be anticipated under information-equivalent representations. We focus on ordering finite response contrasts between prespecified temporal-credit configurations across training realizations; absolute response prediction remains unresolved. Contrastive Synchronization Tangent Energy (CSTE) is a diagnostic that takes the difference between the configurations' squared function-space tangent norms, propagating first-order sensitivity through the complete training recursion. It is computed from current-side training artifacts and frozen before corresponding lagged responses are generated. In a delayed tracking task with 48 fresh schedule twins (pairs of independent training schedules), prospective whole-twin concordance was 0.81120, with Bonferroni-adjusted 98.333% confidence interval [0.77474, 0.84505]. A separately frozen external test supported learned-function contrast ordering (0.9375, separate 99% interval [0.87495, 0.98958]). In the external policy-endpoint assessment, the synchronization-shift interval lay wholly within the registered practical-equivalence region, so the joint endpoint replication criterion was not met.