The dataset is unbalanced in terms of repetitions, mainly due to scanning errors, and for some participants we were able to rescan to make it more consistent. We decided to ship as much usable data for each participant and dataset users can decide what to exclude, based on their own criterion (motion level for instance, or excluding the first run for learning/exposure effects).
The design of the task used the HCP WM eprime script, which contained 4 variants of the task (due to the original HCP design: 2 repeats x AP/PA), and we cycled these 4 variants across the repeats, but due to excluded or failed runs this ordering/balancing is sometimes disrupted.
Yes, the duration programmed on the scanner was a bit longer than the task, which itself included a delay to record the return to baseline after the last event. The countdown numbers are cueing/preparation stimuli displayed by the task, and we exported these in the event files for completeness (as HRF for these events will overlap task trials HRF). All the trials in the events.tsv are covered by the fMRI file (as onset times are all above 0), and we did not discard any TRs in the BIDS leaving the choice to the dataset user. Also, with the multiband acquisition, some single-band and multi-band reference scans are acquired prior to the exported scan (and the first TTL sent to start the task), helping to reach T1 steady-state and making the removal of "dummy volumes" less necessary depending on the analysis.
From @bpinsard:
The dataset is unbalanced in terms of repetitions, mainly due to scanning errors, and for some participants we were able to rescan to make it more consistent. We decided to ship as much usable data for each participant and dataset users can decide what to exclude, based on their own criterion (motion level for instance, or excluding the first run for learning/exposure effects).
The design of the task used the HCP WM eprime script, which contained 4 variants of the task (due to the original HCP design: 2 repeats x AP/PA), and we cycled these 4 variants across the repeats, but due to excluded or failed runs this ordering/balancing is sometimes disrupted.
Yes, the duration programmed on the scanner was a bit longer than the task, which itself included a delay to record the return to baseline after the last event. The countdown numbers are cueing/preparation stimuli displayed by the task, and we exported these in the event files for completeness (as HRF for these events will overlap task trials HRF). All the trials in the events.tsv are covered by the fMRI file (as onset times are all above 0), and we did not discard any TRs in the BIDS leaving the choice to the dataset user. Also, with the multiband acquisition, some single-band and multi-band reference scans are acquired prior to the exported scan (and the first TTL sent to start the task), helping to reach T1 steady-state and making the removal of "dummy volumes" less necessary depending on the analysis.