arXiv:2611.13699v1 Announce Type: new Abstract: Grid mismatch in wideband channel estimation leaves structured cancellation residuals after strong-path subtraction. These residuals generate spurious correlation peaks or mask weak arrivals, turning continuous parameter errors into path-count errors. We propose RC-SAGE-Brent, a residual-closed-loop estimator that jointly refines continuous delay-scale parameters and corrects the unknown-order path set. Within each feedback cycle, ambiguity-constrained alternating Brent searches suppress strong-path cancellation residuals. Candidate validation, resolution-constrained pruning, and full-domain residual redetection then revise the active path set. For linear frequency modulation (LFM) probes, an observable signal-to-residual ratio adaptively scales the model-order penalty to mitigate overestimation caused by delay-scale coupling. Theoretical analysis links off-grid parameter errors to local cancellation-residual curvature and weak-path detectability. Factorial ablations, paired comparisons against standard sparse and off-grid estimators, and measured channel replay demonstrate improved path-set recovery and waveform reconstruction while delineating the operational boundaries of the method.