Methods for reconstructing the topology of complex networks from time-resolved observations of node dynamics are gaining relevance across scientific disciplines. Of biggest practical interest are methods that make no assumptions about the properties of the dynamics, and can cope with noisy, short and incomplete trajectories. Ideal reconstruction in such scenario requires an exhaustive approach of simulating the dynamics for all possible network configurations and matching the simulated against the actual trajectories, which of course is computationally too costly for any realistic application. Relying on insights from equation discovery and machine learning, we here introduce decoupling approximation of dynamical networks and propose a new reconstruction method based on it. Decoupling approximation consists of matching the simulated against the actual trajectories for each node individually rather than for the entire network at once. Despite drastic reduction of the computational cost that this approximation entails, we find our method's performance to be very close to that of the ideal method. In particular, we not only make no assumptions about the properties of the trajectories, but provide strong evidence that our methods' performance is largely independent of the dynamical regime at hand. Of crucial relevance for practical applications, we also find our method to be extremely robust to both length and resolution of the trajectories and relatively insensitive to noise.Actually, network reconstruction is becoming a field of its own within network science [15]. It brings together methodological disciplines such as computer science and statistics with domain sciences such as physics, sociology, biology and neuroscience. Within the context of physics, a myriad of methods have been proposed over the past decade. They are typically anchored in physical insights about network's collective/ emergent dynamics [16,17]. This primarily includes synchronization as the best researched paradigm of collective dynamics [9], in both its theoretical [18,19] and experimental aspect [20,21]. Methods applicable to sparse data have been developed [22], as well as methods that work in the presence of noise [23], or out of equilibrium [24]. Invasive methods assume that one is able to interfere with the system and extract the information from transients [25]. Other methods use compressive sensing [26,27] or elaborate statistics related to derivative-variable correlations [28,29]. Another set of methods attempts to grasp the situations relevant for inferring networks of neurons [30, 31], or even social networks based on infection statistics [32]. On the other hand, somewhat less effort has been invested in the development of methods that can reconstruct the interaction (coupling) functions and not necessarily the network topology [33].The problem of reconstructing the network from dynamical data is not to be confused with the problem of link prediction or network completion [34][35][36]. The latter refers to assessing the existence of ...