The six-score model gave flanker correlation gap bounds of 0.179–0.419 for 133 adults with fixed error variances and zero covariance between errors and latent scores

We tested whether error covariance could remove the higher average latent correlation of mean response time than of conflict cost. We used 133 adults in OpenNeuro ds004883 version 1.0.1. Six scores had fixed error variances for each person. Covariance between every error score and every latent score was zero. A full semidefinite model gave sharp bounds for the supplied moments. The numerical latent correlation gap ranged from 0.178540 to 0.419036. The lower bound exceeded the earlier outer bound by 0.000434. All 499 site resamples passed for each of three error rules. Primary lower-endpoint percentiles were 0.068078 to 0.303649. With doubled error variances, they were -0.089003 to 0.152678. These percentiles are sampling checks, not validated confidence limits. The result reuses the earlier cohort. It does not identify a neural mechanism or prove the error assumptions. The archive contains the proof, code, exact source bytes, hashes, feasible matrices, dual certificates, resample indices, and reproduction commands.

semidefinite programming · measurement error · latent correlations · flanker task · response time · conflict cost · covariance bounds