The evergrowing demand for state-of-the-art machine learning (ML) models has introduced systems with extremely large, often combinatorial, state spaces, which severely impact existing validation and verification techniques needed for ensuring the ML model’s behaviour is guaranteed, safe and dependable. Tsetlin Machines (TMs) are a promising ML model, as their discrete event-based behaviour is naturally explainable and aligns with formal modelling techniques, where prior work shows how they can be captured using ordinary 1-safe Petri nets. However, those TM-Petri nets lack automated construction support, are only validated on a toy-sized dataset, do not cover the assembling process, and use Cartesian product-based operations that scale poorly to the TM’s hyperparameters and larger datasets. In this paper, we present ANIMATE as a holistic data-driven framework that supports automatic construction of our improved TM-Petri nets, which replace the Cartesian product-based operations with parametric mechanisms that scale to any binarised dataset and TM hyperparameter configuration. ANIMATE integrates existing Petri net tools like Workcraft and TINA, and introduces a tool called MaestroPN that supports data-driven simulation. Our experimental results demonstrate the scalability of our new TM-Petri net design, and report the complete net’s assembling and simulation times.