To address prominent problems prevalent at the regulatory end of the dual-prevention mechanism, including data island, risk identification relying on fixed thresholds, and difficulties in penetrating enterprise underlying data, and to explore an effective path for digital-intelligent transformation of the regulatory end, a four-level regulatory framework of "data penetration-early warning grading-closed-loop response-law-enforcement implementation" was proposed. By realizing real-time interconnection between enterprise underlying sensing data and the regulatory end, this framework conducts cross-verification on the authenticity of hazard investigation, rectification management and monitoring data using methods such as spatio-temporal behavior verification, spatial static verification, and data consistency verification. On this basis, an early warning comprehensive index model was established to realize dynamic risk identification and hierarchical response. At the same time, a differentiated and accurate law enforcement path based on a digital evidence chain and "dynamic safety risk profile" was explored to improve the timeliness and accuracy of supervision. The empirical results show that this model improves the efficiency of mine supervision by more than 60% and effectively reduces the frequency of accidents. It provides a scientific scheme for promoting the transformation of the dual prevention mechanism from formal operation to substantive operation, and provides a feasible path for building a more efficient and intelligent safety governance system for non-coal mines.