A radial decoder reversed a continuous odor value model; one pooled spike panel and three fMRI adults did not confirm fixed linear transfer

We tested whether failed prediction across odor pleasantness ranges identifies separate neural codes. A noiseless model had exactly one continuous value axis. A radial decoder gave mean transfer correlation -0.824; a linear decoder gave +1.000. With the same continuous code and four actual rating lists, radial correlations ranged from -0.894 to -0.806. A broader radial width restored positive order. This is a decoder diagnostic, not measured neural evidence. We then applied fixed readers to public amygdala data from separate human studies: one pooled panel of 622 implanted units and three fMRI adults. Odor identities were excluded across low and high median ranges. Mean linear correlations were -0.141 and -0.076. Conditional 95% odor ranges were [-0.706,+0.719] and [-0.181,+0.063]. The proposed shared positive transfer rule failed. The MRI person interval was [-0.305,+0.154]. Source-zero and mean-only checks were discordant. All negative scores remain. An exact-constant numerical guard changed some resampled fits but no primary forecast. A second solver reproduced every primary forecast within 1.8e-15. The spike release lacks donor labels; MRI has only three adults. Cohort, target, coverage and source processing differences limit inference. The model disproves an unconditional decoder-to-code inference. It does not establish a continuous or separate human odor value code, a positive neural result across measurements, or causality. Code, permitted public inputs, source pins, plans, uncertainty draws, figures and repeat steps are supplied.

human amygdala · odor pleasantness · neural decoding · extrapolation · radial kernel · single neurons · fMRI · negative result