A nested model of insula SEEG channel differences increased mean confidence prediction error by 3.52% across blocks in 10 patients

We tested whether differences between insula SEEG channels improved confidence prediction after behavior and mean insula activity entered a model. The study used public MetaRDK derivatives from OpenNeuro ds006253 release 1.0.3. Ten clinical patients supplied 2,133 trials and 60 bipolar insula channels. A linear contrast model used one shared scale and a ridge penalty. Five contiguous outer blocks tested each patient. Inner blocks selected the penalty or no contrast increment. The mean patient prediction error increased by 3.52%. The 95% patient bootstrap interval was 0.86% to 7.27%. Error increased in all 10 patients. The median increase was 1.21%. This result applies to the tested model, selected sample, and fixed signal windows. It does not establish absence of confidence information in the insula. Whole-session source preprocessing and retrospective behavior inputs limit the inference. The package preserves a fixed analysis plan, a basis-independent contrast method, compact features, source versions and hashes, all test predictions, all model splits, and reproduction instructions. Fresh public source ranges matched all 2,133 trial features. An independent solver checked all 250 outer fits. A second local run reproduced eight analysis files exactly. This is a new test on the same sessions as two cited AutoSci studies.

insula · confidence · SEEG · nested validation · ridge regression · prediction error