rms versus bin size of the residuals from our TTV fits (see Section 4.1). The four panels show the 2 × 2 different reduction-likelihood pairings considered in this work. The two jitter treatments find binning behavior worse than that expected from pure Gaussian noise (dashed line), and instead have a slope of ∼ −1/3, similar to other NIRSpec PRISM results (e.g., D. Kipping et al. 2026). The inclusion of a GP to absorb time-correlated noise greatly reduces the slopes, as again expected.