The ds000102 lower bound is negative at 57 shared label errors when labels in two untrimmed runs can change.

This conditional sensitivity analysis asks how many shared label errors can make the equal-run mean response-time (RT) contrast negative in OpenNeuro ds000102. The source contains 26 participants, two runs per participant, and 24 trials per run. Two recorded condition fields agree on 624 run-1 trials and 320 run-2 trials, and conflict on 304 run-2 trials; the presented labels remain unknown. The primary model allows agreeing labels in either run to change at a cost of one error and allows conflicting labels to change at no cost. It preserves recorded RT and correctness, requires 12 labels per condition in each participant/run and at least eight usable RTs per condition, and gives each of 52 participant/run contrasts equal weight. With correct positive RTs and no upper RT cap, the lower bound is positive at 56 errors (2137/2860 ms) and negative at 57 (-2083/17160 ms). With a 2000 ms cap, the corresponding budgets are 59 (15391/102960 ms) and 60 (-36607/51480 ms). When run-1 labels stay fixed, the first negative bounds occur at 123 errors without a cap (-769/10296 ms) and 122 with the cap (-1397/4680 ms). Exact rational dynamic programming calculated bounds for every budget; saved assignments and separate checks reproduce the reported endpoints. These are bounds under a specified model, not confidence intervals, estimates of actual error counts, or evidence of biological causality. Labels may depend on RT, and the model does not assign probabilities to label configurations. The presented labels remain unresolved.

response time · condition labels · sensitivity analysis · error bounds · flanker task · OpenNeuro · dynamic programming · rational arithmetic