Do stimulus-related trajectories repeat across runs?
A group-average trajectory can hide inconsistent repetitions. This page compares each run’s event-aligned shape with the other three runs from the same participant, then shows how that agreement varies across people and networks. Agreement concerns the timing and shape of expression, not rediscovery of the anatomical network.
Explore the figures
Start with the cohort agreement figure. Select a participant/network cell to inspect its four run trajectories; focus on one run when you want its other-three-run reference. Similar rises and falls give positive agreement, opposing trajectories negative agreement. A low value is a reason to inspect the traces, not automatically exclude a recording.
Do task-linked network trajectories repeat within participants?
Every participant has four runs expressed in the same cohort basis. For each run and network, we correlate its block-average curve with the equal-run mean of the other three runs from that same participant. The cohort view summarizes those four agreement scores per participant. It does not compare a person’s chronological recording with another person’s scan.
Main findings
How much does agreement vary across participants?
Each point represents one participant’s mean of four within-participant agreement scores. Bars average participants equally; error bars are participant SD, not confidence intervals. A high cohort mean can conceal people or runs with weak agreement. These descriptive arithmetic averages of correlations are not Fisher-z inference.
What do differences imply?
Repeated task-linked trajectories support temporal repeatability within this fitted representation. Weak agreement can reflect variable responses, attention, motion, timing dependence or estimation noise; this analysis does not identify the cause. Review the Preparation & QC tab alongside unusual curves, including retained motion flags.
Onset and offset windows overlap and contain temporally autocorrelated samples. There are 15 independent participants, not 60 independent people or 720 independent task repetitions. Participant-level uncertainty and held-out refitting are future validation steps; no significance is claimed here.
More about the measures and interpretation
The reference excludes the run’s curve, but the model was trained on all 60 runs. This is in-sample repeatability of expression in a common learned representation, not held-out model validation or independent rediscovery of spatial networks. No participants are excluded or flagged by a new agreement cutoff.
Positive Pearson correlations describe similar rises and falls, negative correlations opposing trajectories. Correlation removes mean offsets and ignores positive amplitude scaling. Click a cell to inspect its participant and network. Network 1 is hidden in figures and retained in exports.
For a full-cohort curve, blocks are averaged within run, runs within participant, then participants equally. Curves are standardized within each full run before event alignment. With one participant selected, the default shows only its four run curves; focusing on a run adds the other-three-run reference. No all-run mean is superimposed in that view.