Fracturing controls rates of mass, chemical and energy cycling within the crust. We use observed locations and magnitudes of microearthquakes (MEQs) to illuminate the evolving architecture of fractures reactivated and created in the otherwise opaque subsurface. We quantitatively link seismic moments of laboratory MEQs to the creation of porosity and permeability at field scale. MEQ magnitudes scale to the slipping patch size of remanent fractures reactivated in shear - with scale-invariant roughnesses defining permeability evolution across nine decades of spatial volumes – from centimeter to decameter scale. This physics-inspired seismicity-permeability linkage enables hybrid machine learning (ML) to constrain in-situ permeability evolution at verifiable field-scales (~10 m). The ML model is trained on early injection and MEQ data to predict the dynamic evolution of permeability from MEQ magnitudes and locations, alone. The resulting permeability maps define and quantify flow paths verified against ground truths of permeability.