Vote Tracing: Make RF Ensembles Auditable, Explainable, and Deployable TL;DR — Per-model “vote traces” turn your RF modulation ensembles into fully auditable systems: exact Shapley attribution in microseconds, gorgeous timelines, and an open-set detector that… 

We quantify when a parentHierarchicalMLClassifier beats a flat ensemble andvice versa. We report per-class win profiles, confusiondeltas, and latency trade-offs, with code paths mapped tosuper().classify_signal() vs the ensemble voting block.We find a hierarchical classifier is… 

On RadioML 2016 (5 classes, 20k examples), hierarchical strictly dominates flat (0 unique flat wins). On TorchSig Sig53 (53 classes, 5M examples), hierarchical gains 15% on PSK/QAM families but flat wins 12% on impaired FSK/OFDM,… 

Your modulation paper showed hierarchical ≥ flat with almost no cases where flat uniquely wins — that’s the exception, not the rule. In radar, the opposite is true: flat almost never wins on realistic taxonomies… 

*November 15, 2025 | Ben Gilbert* Academic reproducibility has long been the holy grail of scientific research, yet most papers remain frustratingly opaque about their implementation details. After spending years wrestling with broken experiment scripts,… 

We investigate post-softmax calibration for weightedensemble voting in RF signal classification. Neural network confidence scores are often miscalibrated, leading to overconfidentpredictions that degrade ensemble performance. Using per-modeltemperature scaling, we reduce Expected Calibration Error(ECE) from 15.4%… 

Ensemble methods for RF signal classification combinepredictions from multiple neural networks to achieve superioraccuracy over individual models. However, modern neural networks often exhibit poor calibration—their confidence scoresdo not reflect actual prediction accuracy [1]. This miscalibration…