Exact image trees did not ensure tree distances between category centers; shared EEG and fMRI rejection tests failed

We tested if category response centers give repeatable evidence against one tree in two human measurements. Separate public studies supplied scalp EEG and fMRI responses to the same 48 source photographs. Two EEG people and five fMRI people supplied development. The fixed test used eight other EEG people and ten other fMRI people. Neither of two fixed joint tests passed Holm correction; both adjusted p values were 1. Some EEG gaps changed sign. Image resampling also left fMRI uncertainty. An exact model then showed a separate limit of the method. All 48 image patterns had unit length and zero channel mean. Their ordinary image distances fit one star tree exactly. Four equal groups had 12 images each. Distances between group centers did not fit any tree. The least possible maximum pair error was 6.12% of the mean center distance. This model was not fitted to the human data. It applies to either measurement when its input is the stated correlation distance. It is not evidence that human responses follow the model. The contribution is an inspection tool, a derived benchmark, an exact counterexample and two fixed negative human tests. No positive cross-modal biological effect or cause is established. The package gives source versions and terms, code, all outcomes, figures and repeat instructions.

representational geometry · EEG · fMRI · tree metric · category centers · data reuse · negative results