Thin-interbedded low-permeability sandstone reservoirs are characterized by frequent sand–mudstone alternations and strong vertical heterogeneity, which make accurate sweet-spot identification and reliable development-performance prediction difficult. Conventional sweet-spot evaluation methods commonly rely on fixed weighting or linear parameter superposition, and thus cannot fully represent the interlayer flow imbalance and cross-layer stress barriers that govern thin-interbedded reservoirs. Supervised data-driven models can capture nonlinear relationships, but their application is limited when reliable sweet-spot labels are scarce. This study proposes a physics-constrained adaptive weight-optimization strategy based on an ANFIS–VAE dual-pathway network to establish an integrated sweet-spot evaluation system from the dimensions of production potential and fracturing potential. Two thin-interbedded-specific indicators are introduced. Permeability contrast quantifies interlayer petrophysical heterogeneity, and interlayer stress difference characterizes cross-layer fracture-propagation constraints. The optimized sweet-spot favorability index (SFI) is classified into three tiers using K-means++ clustering. Results show that oil saturation is the leading control on production potential, with a weight of 0.3052, while permeability contrast contributes approximately 20%. Interlayer stress difference dominates fracturing-potential evaluation, with a weight of 0.2534. Class I, Class II, and Class III sweet spots account for approximately 22%, 45%, and 33% of the study area, respectively. Compared with the conventional method, the proposed approach reduces the RMSE from 0.35 to 0.06 on a held-out well excluded from model training. Reservoir numerical simulation further confirms the geological rationality of the classification. Class I zones exhibit relatively uniform depletion with an effective recovery of approximately 65%, Class II zones show interlayer-differentiated depletion with an effective recovery of approximately 52%, and Class III zones display channelized flow under mudstone shielding with recovery below 40%. The proposed framework provides a physically consistent basis for sweet-spot grading, differentiated stimulation design, and zone-specific development deployment in thin-interbedded reservoirs.