Task labels are swapped in expectation arrays for 12 of 23 people in ds004295 version 1.0.0

We tested whether a choice model improves later feedback forecasts beyond reported expectations. We used 23 people and 12,880 trials from OpenNeuro ds004295 version 1.0.0. The supplied expectation arrays have opposite task labels for 12 people. Each of these people did the punishment task first. Exact event matches support this result. All 11 people who did the reward task first had matching task labels. The correction changed 4,412 rating values. The package keeps the corrected table and unchanged source files. Models used trials 1 through 140 to forecast feedback on trials 141 through 280 in each task. The primary added Brier loss gain was 0.000646. Its 95% person bootstrap interval was -0.000994 to 0.002405. A planned check with a larger slope penalty gave a positive interval. Thus, the forecast estimate depends on the stated settings. The source finding does not establish an effect on the source study's EEG results. The package keeps source hashes, plans, code, fits, forecasts, bootstrap estimates, figures, and checks. Two full runs produced identical analysis files.

expectation ratings · feedback forecasting · reinforcement learning · task labels · Brier loss · OpenNeuro