M-TriX: A Hybrid-Causal Neural Architecture for Balanced Intelligence, Stability, and Interpretability

This work introduces M-TriX, a Dynamic Balance Architecture designed to enhance stability, interpretability, and efficiency in convolutional networks. The architecture is built around the Gated Residual Bridge, producing spatial p-maps that act as transparent decision-control fields. This version (v0.6.1) includes the complete Master Proof, covering noise-augmented training, the reinforced dual-collapse sweep test, the noise robustness test (M-TriX vs Dense), and the visual clean-filter evidence. A full reproducibility supplement and the official implementation notebook are provided to enable exact replication of all experimental results.