"DeepSentinel Reproducibility Dataset: Campaign Outputs, Metadata, and Representative Traces for D1\u2013D4"

"Predictive control can improve the timing of ledger hardening in quantum-aware edge ledgers, but in security-critical consensus the forecasting layer must remain policy-bounded rather than unconditionally authoritative. In this paper, we present DeepSentinel, an uncertainty-aware predictive meta-consensus architecture that combines deep sequence forecasting, predictive uncertainty estimation, an uncertainty-aware resilience index, and a bounded fallback mechanism that reverts control to a classical autoregressive path when forecast reliability degrades. Our evaluation across four campaign families shows a clear regime-dependent pattern. Under nominal conditions, DeepSentinel remains non-disruptive, with zero fallback activity and zero envelope exceedance. Under abrupt regime switching, the autoregressive baseline remains the stronger one-step forecaster, showing that deep prediction does not provide a uniform advantage across all disturbance classes. Under oscillatory disturbance, however, DeepSentinel delivers its strongest result by eliminating the phase lag observed in the autoregressive baseline at intensities 1.0 and 1.2, while also reducing RMSE from 0.022979 to 0.018519 and from 0.025569 to 0.022393, respectively. Our uncertainty-tuning study further shows that uncertainty-aware fallback remains operationally stable and can improve forecast accuracy within the deep forecasting family, although these gains are accompanied by a measurable fallback burden and do not establish a broad dominance frontier. Taken together, our results show that deep prediction is most valuable when disturbance carries exploitable temporal structure, and that its practical value in secure ledger control depends on uncertainty-aware governance of forecasting authority."