Low-slow-small (LSS) radar targets often produce compact echoes whose internal components cannot be reliably separated by a discrete Fourier transform. Although super-resolution estimators can refine continuous parameters when the model order is known, an incorrect front-end order decision may merge distinct targets or introduce spurious components. This paper formulates the front-end task as a binary decision between one resolvable component and two resolvable components. We propose the binary split consistency operator (BiSCO), which discovers one- and two-component candidates from noncoherent residual evidence across short frames and evaluates their predictive consistency on held-out validation frames with independently fitted complex amplitudes. A split is accepted only when the validation residual reduction exceeds a fixed dimension-dependent penalty motivated by the Bayesian information criterion. The default decision threshold is determined by the effective validation dimension and does not require empirical calibration from target trials. Under an independently calibrated 1% false-alarm protocol for cross-method comparison, BiSCO achieves a mean detection probability of 0.913 at a stress spacing of 0.20 discrete Fourier transform cells and 0.982 at 0.35 cells. The corresponding worst-case values are 0.838 and 0.957; from 0.50 cells onward, the mean detection probability is at least 0.999. The default BIC-motivated rule yields no false split in 6000 independent single-target trials. BiSCO estimates a binary resolvable model order rather than the physical scatterer count and is intended as a lightweight front end for known-order super-resolution estimators.
