Background and objectives: An early understanding of numerosity is crucial to developing proficiency in arithmetic, and rapid pattern recognition via subitizing is key in this process. The training app "LORE" specifically targets and cultivates pattern recognition skills from the beginning of formal schooling. This study investigated possible transfer effects on computational fluency and arithmetic ability of the readily available LORE training. Methods: Included were 679 children from 38 Austrian school classes whose teachers used the novel web-based numerical training program LORE in their classrooms in the school year 2020/21. Notably, and unlike under ideal experimental conditions, use of the online training was at the teachers' discretion and also depended on uptake by the children. Based on log-file data about program use, students were assigned to one of three groups (no usage, some usage, and full usage). Using the pattern recognition skills assessed at the beginning of first grade as baseline we tested the effects of LORE training on computational fluency and arithmetic skills as assessed throughout first and second grade. Results and discussion: Our results demonstrate that full usage (compared to no usage) had significant effects on computational fluency that were evident up to the midpoint of the following school year. Although our study design warrants ecological validity by taking a "real world" approach on investigating training effects, the quasi-experimental design limits the control of confounders and therefore impairs internal validity. The pedagogical implications of training in pattern recognition and the need for future research are discussed.