Two recent-repeat trials had a higher mean error rate than other non-targets in 158 ds005270 two-back records

We tested recorded errors in 158 public two-back task records from OpenNeuro ds005270 version 1.0.0. Each person had two non-target letters that matched the immediately preceding letter. The mean person error rate was 62.34% on these trials and 16.47% on other non-targets. The paired mean difference was 45.87 percentage points. Its 95% person bootstrap interval was 40.14 to 51.57 points. The interval assumes independent people and a fixed task schedule. The gap remained positive in planned response-time and covariate checks. In 157 complete schedules, the two repeats had fixed letters and positions. An exact null vector showed that a model with unrestricted position effects has no unique repeat coefficient. The task gives only two lure observations per person. The package contains 164 source files, a full trial ledger, the local plan, code, results, uncertainty draws, and figures. Independent scoring and model checks passed. A new download and full replay matched all source, result, and figure files. The result describes these recorded trials. It does not establish a causal repeat effect or a neural mechanism.

two-back task · working memory · recent-repeat lure · error rate · position aliasing · behavioral data · OpenNeuro ds005270