Accurate evaluation of AC losses under coupled DC transport currents and time-varying magnetic fields is critical for the thermal reliability of large-scale HTS coils. However, traditional homogenization techniques frequently suffer from severe staircase artifacts and numerical discontinuities due to their reliance on rigid, resulting in significant loss deviations. To address these limitations, this paper proposes an advanced Self-Adaptive Homogenized (SAH) model based on dynamic critical boundaries of electromagnetic evolution. Utilizing a field-decoupling strategy to extract external field extrema, an effective thickness correction is incorporated into the parabolic analytical equations to precisely capture the asymmetric spatial profiles of the flux penetration boundary. This boundary evolution is then dynamically mapped onto the coil submodels via a parabolic gradient auxiliary function, effectively smoothing current density transitions and eliminating discrete artifacts. The baseline fidelity of the reference model is rigorously validated against a customized experimental setup. Quantitative benchmarks demonstrate that while the traditional LH model exhibits computational errors exceeding 20%, the proposed SAH model robustly restrains the numerical deviation within 7% across all operating envelopes. Furthermore, the SAH model exhibits superior robustness in extreme scenarios involving rotating fields coupled with high DC bias currents by self-consistently tracking the topological shifts of subcritical regions. This work provides a highly scalable and reliable numerical framework for the design and optimization of complex, large-scale HTS devices such as superconducting turbines, magnets, and cables.