Limited incremental prediction of cardiac synchrony judgments from heartbeat locked scalp activity

Heartbeat-locked scalp responses can differ across attention and expectation conditions, but this does not establish incremental prediction of a later subjective judgment. We tested a fixed 200–280 ms post-R representation in 1,731 internal-attention omission trials from an existing 34-person archive. Five folds held out participant labels. L2-regularized logistic models compared task/history, ECG/R–R and pre-R scalp covariates with and without sixteen principal components of post-R channel means. A separate outcome-blind median/MAD calibration used each participant's full omission-epoch waveforms, including held-out participants; the design is therefore offline and transductively calibrated. The mean participant-balanced log-loss gain was −0.005176 nats per trial, with 95% participant-bootstrap interval [−0.011304, 0.000593] and one-sided 1,000-shift correspondence p = 0.230769. The upper bound was below the locally specified 0.006931-nat benchmark, while the interval included zero. No inferential secondary survived Holm correction, and the external-attention endpoint was not estimable under the unchanged singleton-stratum null. October replay reproduced ten saved output tables from preserved parser outputs; independent code verified summary scoring, not the model fitting or waveform parser. The result limits this representation and prediction procedure, not heartbeat-related neural information generally, cortical source identity or causal interoception.

This is an author preprint and has not been peer reviewed. The supplement provides author code and aggregate results only. No individual/trial data are redistributed, and the package is not a self-contained raw-to-results reproduction. See the manuscript and methods history for the preserved nulls, unexecuted checks and provenance limitations.