Sixty of 171 ds005754 v1.1.1 records failed a timing test under one fixed go law per record and written stop fields

We built a timing test for fixed trial counts and delays set before their current outcomes. The proof uses independent go observations from one fixed law in each chosen group. It permits adaptive delays and dependent go and stop durations. The test gives a lower bound on the largest conditional probability shortfall. In all 171 public ds005754 v1.1.1 records, 60 failed under the written delay and outcome fields. The primary procedure has a nominal 5% error bound across these records under its assumptions. In 24 records the shortfall lower bound exceeded .10. Excluding inferred practice gave 60 failures. A correct-response lower proxy gave 46. Separate main-block go laws gave one failure and wider error bounds. The dataset was selected after an earlier audit, so this application is exploratory. Separate checks matched all 1,026 analysis rows and 19,836 trial comparisons. All 260,000 simulation decisions passed a numerical check. A full repeat matched their arrays. New retrieval and analysis matched the source and main results. The simulations showed low power for some adaptive paths. A rejection does not identify a field error or impaired inhibition. The package preserves source data, hashes, new code, the proof, every result, and reuse instructions.

adaptive testing · stop-signal task · conditional probability · empirical CDF · concentration bounds · inhibitory control · reproducible analysis