The unevenness of digital development across Russian regions increases the demand for forecasting tools that are suitable for managerial decision-making while preserving an economically interpretable logic of influences. Composite digitalization indicators are convenient for monitoring; however, their dynamics are sensitive to structural shifts and regime changes. This reduces the reliability of simple trend-based approaches and complicates the use of opaque models in scenario analysis. The aim of this study is to develop a hybrid forecasting model for a regional composite digitalization index, that is robust under unstable dynamics and provides an interpretable representation of factor effects. The empirical basis is a region–year data mart constructed by the authors, combining observed values of the digitalization index with comparable socio-economic indicators from official statistics. The proposed methodology relies on a two-component architecture. Nonlinear dependencies of factors are specified by an interpretable layer of Kolmogorov–Arnold networks, where the influence of each feature is represented by a smooth univariate function parameterized by splines. The dynamic component is implemented as a recurrent module with chaotic modulation, designed to account for inertia and accelerated reorganization of the model's internal state during regime changes. To prevent over-complexification, the authors introduce a complexity selection procedure based on a trade-off between forecast accuracy on held-out partitions and the smoothness of the influence functions. The validity of the findings is confirmed by comparison with baseline econometric and neural network solutions and ablation experiments that separate the contribution of interpretable nonlinearity from that of the dynamic mechanism. The results indicate that selecting a minimally sufficient complexity level keeps predictive accuracy close to the best configurations by metrics, while preserving stable, economically readable form of the influence functions. Interpretation of the edge functions reveals nonlinear effects, including saturation of the influence of economic scale and a heterogeneous response of the index to labor market conditions. The practical value of the study lies in the possibility of using the model as an analytical module for monitoring and scenario-based decision support in regional digital policy, since forecasts are accompanied by a transparent structure of factor influences.