Neutrino oscillation established that neutrinos are massive and mixed, motivating precision measurements of the mixing parameters. Long-baseline (LBL) experiments address key open questions, including the CP-violating phase δ_CP and the neutrino mass ordering, but the coming precision era demands unprecedented control of systematic uncertainties. This thesis presents three complementary projects towards this goal, mainly within the T2K experiment. First, the characterization, hardware validation and performance of the SuperFGD front-end electronics for the upgraded ND280 near detector, together with a novel time-calibration method. Second, the first measurement of the visible superscaling variable ψ′_vis in neutrino–nucleus interactions using MINERvA data, compared to theoretical predictions through forward folding, probing interaction physics that dominates oscillation uncertainties. Third, a novel machine-learning approach using Normalizing Flows to learn the multidimensional T2K Near Detector likelihood, providing a more accurate analytical description than existing methods.