Disorder-averaged dynamics in quantum many-body systems are challenging to simulate, due to the loss of translational invariance and the need for extensive disorder sampling. In this talk, I will introduce the statistics-encoded tensor network (SeTN), a method that embeds disorder into an auxiliary tensor layer, averages locally, and restores translational invariance after compression. A universal criterion $n \gg \alpha^2 t^2$ ensuring compression efficiency emerges from the singular-value structure, linking Trotter resolution, disorder strength, and evolution time. As an application, I will show how SeTN captures the spectral form factor of the disordered transverse-field Ising model, where the long-time dynamics are governed by a single leading transfer-matrix eigenvalue, which differsfrom the behavior of Floquet models.