All 84 gradCPT main blocks have the same stored image code sequence in ds006040 version 1.0.1

We tested 84 public gradCPT behavior tables from 28 people in OpenNeuro ds006040 version 1.0.1. All main blocks had 450 trials with the same image category and image number at every position. The raw and derivative copies had identical Git blob hashes. We preserved the source files and reconstructed 37,800 trials. We also tested whether a person's prior response at the same trial improved a forecast for the third run. The observed log loss gain was 0.00733 nats beyond person rate, common trial rate, and recent responses. The gain was 0.00372 nats when presses before 150 ms were excluded. Percentile intervals from model refits excluded zero. Basic bootstrap intervals included zero for both outcomes. The interval method changes whether zero is excluded. Stored presses do not establish which image caused a response. The sequence result applies to the tables. It does not verify the displayed images or a neural attention state. The package includes code, source data, hashes, forecasts, resampling records, figures, and reproduction instructions.

sustained attention · gradCPT · behavioral data · trial sequence · forecasting · ridge logistic regression · bootstrap