Dear Editor-in-Chief, My co-authors and I are pleased to submit our manuscript entitled “Neurosymbolic Theory Revision through Predicate Invention” for consideration as a research article in Neurosymbolic Artificial Intelligence. The topic of this manuscript is the revision of imperfect symbolic theories within neurosymbolic systems. Existing approaches typically assume complete and correct background knowledge, which limits their applicability in realistic settings. We introduce NeTheR (Neurosymbolic Theory Revision), a framework that iteratively refines an initial logical theory through targeted structural modifications, including the insertion and removal of symbolic literals and the invention of neural concepts learned from subsymbolic data. By integrating these neural concepts into compiled logical circuits and selecting revisions using a risk-adjusted heuristic, NeTheR enables data-driven improvement of symbolic models while preserving their structure. Empirical results demonstrate consistent and statistically significant performance gains over the initial theory and competitive baselines. This manuscript has not been previously published and is not under consideration elsewhere in any form. All authors have approved the manuscript and agree with its submission to the journal. The authors declare that there are no known conflicts of interest related to this work. Sincerely, Sieben Bocklandt, PhD student Declarative Languages and AI KU Leuven, Department of Computer Science Celestijnenlaan 200A box 2402, 3001 Leuven, Belgium sieben.bocklandt@kuleuven.be On behalf of all co-authors: Vincent Derkinderen Wannes Meert Luc De Raedt