A recurrent Transformer model that reduces transformer parameters by using a single layer in a circular manner, enhanced by adaptive level signals from low-rank matrices. (Findings of the Association for Computational Linguistics: EMNLP 2025)