While significant research efforts have been directed toward developing more capable neural decoding architectures, comparatively little attention has been paid to the quality of training data. In this study, we address the challenge of constructing effective training datasets to maximize the potential of existing syndrome-based neural decoder architectures. We emphasize the advantages of using fixed datasets over generating training data dynamically and explore the problem of selecting appropriate training targets within this framework. Furthermore,we propose several heuristics for selecting training samples and present experimental evidence demonstrating that, with carefully curated datasets, it is possible to train neural decoders to achieve superior performance while requiring fewer training examples. Code to reproduce all results is available at https://github.com/lebidan/sbnd.