Model and readout dependence of radical pair yield anisotropy

Directional reaction-yield contrast in radical-pair models depends on the Hamiltonian, reaction operator, noise channel, absolute rates and readout. We correct an earlier preprint by separating two historical one-nucleus conventions and independently checking a newly frozen computational benchmark. For an isotropic-hyperfine model, rotationally covariant noise gives no directional yield contrast; a fixed dephasing axis supplies the directional reference. At one specified noise strength, a finite observation window has an interior rate-ratio maximum that is absent from the corresponding complete-yield scan. The historical left-rectangle readout can exceed unit yield, exposing a separate quadrature error. In an axial-hyperfine model, two rate-dependent response peaks are visible at reference rates, but their locations and amplitudes change with absolute reaction rate and with genuine Jones-Hore rather than Haberkorn kinetics. Molecular-axis dephasing can enhance the declared absolute contrast while two explicitly matched isotropic channels suppress it. Full angular profiles expose antipodal argmax ties and do not support the previously claimed switch near 1.8 mT. Archived two-nucleus calculations retain only five of seventeen qualified branches within the earlier location window, all in the generic perturbation arm; fourteen of seventeen have more than ten percent discrepancy from a legacy meridian scan. A separate implementation reproduces selected tensor profiles and agrees with the new production outputs within frozen tolerances. The resulting contribution is a reproducible corrective benchmark of specified spin models. It establishes no universal optimal ratio, generic noise protection, source-article replication or biological optimization. Corrective release with the corrected manuscript, source code, data and figures. Related Research Square record: https://www.researchsquare.com/article/rs-8798749 . Earlier universal shelf, scaling and inversion claims are withdrawn as detailed in the manuscript. This preprint is not peer reviewed.