Modern distributed computing environments suffer from resource thrashing, cascading failures, and inefficient allocation due to reliance on static utilization thresholds. This paper presents a governance-aware dynamic resource allocation framework integrating a novel Centrality-Entropy Index (CEI) with adaptive weight recalibration, oscillation suppression via hysteresis windows, fault propagation probability modeling, and pre-modification k-hop simulation. Unlike conventional autoscaling that reacts to instantaneous CPU or memory metrics, the proposed system reconstructs sustained workload behavior through longitudinal telemetry analysis, incorporates mission-criticality and disaster recovery dependencies into allocation decisions, and validates modifications against a dependency graph before execution. Empirical validation in a production cloud environment demonstrated sustained cost savings of 27% ($6,000/month from $22,000 baseline), 35-40% reduction in scaling oscillation events, and 18-22% improvement in compute utilization efficiency. The framework is platform-agnostic, operating on standardized telemetry across AWS, Azure, Google Cloud, and Kubernetes environments. A provisional patent application (USPTO No. 63/999,378) has been filed for the system architecture.