Optimized weights reduced conservative 95% interval widths in 431 AOMIC designs under a fixed response model with [0,1] amplitudes, ±1 s delays, and known Gaussian noise

We selected contrast weights before signal analysis in 431 public AOMIC working memory designs. The primary model allowed independent condition delays in [-1,1] seconds and amplitudes in [0,1]. It assumed known independent Gaussian scan noise with standard deviation 0.1. The weights minimized a conservative 95% interval width that included a continuous delay error bound. The mean per-design width decrease was 19.71% relative to the better canonical or raw derivative reference. The median decrease was 20.44%. All 431 primary designs had a decrease. All 3,017 planned fits passed the numerical checks. Independent response calculations, a second solver, scan noise simulations, and a 12-design primary repeat checked the results. The planned full repeat did not finish within the run time limit. The package preserves the proof, code, source inputs, hashes, results, weights, draws, and retrieval instructions. The guarantee requires the stated response bounds and known Gaussian noise. The study did not test measured BOLD signals or biological validity.

fMRI · contrast weights · bounded response delays · confidence intervals · bias-variance tradeoff · convex optimization · AOMIC · working memory