With fixed models and Student tests, mean boundary label gains were below 0.5 percentage points in 12 iEEG patients and ten fMRI adults

We tested whether narrative boundary labels improve prediction of hippocampal activity in film sections excluded from fitting. Separate public studies supplied bipolar intracranial EEG power and BOLD fMRI. Four patients and five adults supplied development. A local plan preceded tests in 12 other patients and ten other adults. The model added a boundary term after removal of its training relation to cut and shot history inputs. Mean gains were -0.0020 and +0.0325 percentage points of residual signal energy. Their 95% person Student intervals were [-0.0133, +0.0092] and [-0.0404, +0.1055]. Mean full-model skills were +0.0621 and -0.0167 points. The useful shared prediction rule failed. A separate fixed test bounded each mean gain below 0.5 point, with Holm p=1.52157e-07. This bound assumes independent people and adequate Student inference. It concerns one reader and whole-film residual energy, not absent neural boundary responses. Seven non-hard intracranial narrative changes lie in the editing control span. Only five hard narrative transitions supply distinct label events. Separate cohorts, films, labels, clinical coverage and source processing limit scope. The package gives new power and regional BOLD derivatives, compact model inputs, code, source pins and terms, all outcomes, figures and repeat steps. Original voxel arrays and voltages stay external.

human neuroscience · intracranial EEG · fMRI · hippocampus · event boundaries · blocked validation · gain budget · stimulus confounds · data reuse