Anomalous Curve-Shape Deviation in Singlet Bell Measurements on IBM Superconducting Qubits

We report an anomalous deviation from the quantum mechanical singlet correlation curve in full angle-sweep Bell measurements on two IBM superconducting quantum processors (ibm_fez and ibm_marrakesh). With Alice fixed along Z and Bob sweeping Ry(δ) from 0° to 180°, the disagreement probability P_disagree(δ) deviates from the predicted cos²(δ/2) in a way that cannot be explained by uniform depolarizing noise: the 50% crossover point shifts below 90°. Fitting a two-parameter deformation model yields a crossover-shift parameter α = 0.470 ± 0.003 on ibm_marrakesh after readout correction, compared to the quantum prediction α = 0.5. The quantum model with a free visibility parameter gives χ²/dof = 4.83; the deformation model gives χ²/dof = 1.41, a Δχ² = 124.5 improvement for one additional degree of freedom (p < 10⁻²⁸, likelihood ratio test). The deviation is consistent across two physically independent chips. We systematically investigate five candidate instrumental explanations—readout asymmetry, Ry gate calibration offset, T2 thermal decoherence, angle-dependent gate duration variation, and qubit-qubit crosstalk—and find each insufficient through dedicated experiments or analytical arguments. The anomaly is currently unexplained by standard NISQ noise mechanisms and warrants independent replication on different hardware platforms.