One-bit face–house readers lost 0.064 AUC in one of six ECoG patients and tied a bit-count reader in one of six fMRI participants

We tested whether one bit at each recording site retains face–house ranking in two separate human studies. A training median changed each site's response to a high or low bit. One ECoG patient supplied development. A fixed test used six other ECoG patients and six fMRI participants. Mean later-test one-bit AUC was 0.7671 in ECoG and 1.0000 in fMRI. Mean changes from continuous readers were -0.0029 and +0.0000. Their pointwise 95% person-bootstrap intervals were -0.0291 to 0.0160 and 0.0000 to 0.0000. The shared rule required ranking above chance, a loss below 0.05 AUC, and ranking above a total-bit-count reader in both modalities. One ECoG patient lost 0.064 AUC. One fMRI count reader also reached AUC 1.000. The exact person sign-test conjunction p was 0.109375. The fixed rule failed. The fMRI ceiling cannot establish equal population information. A conservative 95% mean AUC bound at that ceiling was 0.5407 to 1. All individual results and controls remain in the package. These are new-endpoint tests in reused cohorts. Different tasks, selected clinical coverage, selected fMRI regions, small cohorts, repeated images, and offline processing limit scope. The result does not establish an optimal bit depth, a binary neural code, or a common cause. The package gives code, source versions and terms, frozen plans, figures, numerical checks, and repeat instructions.

human neuroscience · ECoG · fMRI · visual categories · quantization · binary features · independent cohorts