One shared label change gives a zero lower bound for the two-run error cost in the ds000102 model.

Using all 26 participants and 1,248 trials in the public ds000102 flanker event files, this analysis bounded the two-run error-cost mean under a balanced trial-label model. Error cost was the incongruent minus congruent error rate in percentage points. With no changes to labels on rows where the source condition fields agree, the bounds were 25/78 to 75/26 pp (0.320513 to 2.884615 pp). Allowing one such change made the lower bound zero; allowing two made it -25/78 pp (-0.320513 pp). The model could assign either condition to conflicting rows at no budget cost and retained 12 trials per condition in each run. These conditional results do not identify the correct labels or show that any label was wrong. The exact participant-bootstrap interval for the zero-budget two-run lower bound included zero. Sparse errors, fixed recorded correctness, condition balance, the assumed label-change budget, and participant sampling limit interpretation. The analysis does not establish a biological result or a positive population lower bound.

flanker task · error rate · error cost · condition labels · sensitivity analysis · participant bootstrap · dataset quality