The Curve Must Be Earned: A Preregistered Closure-Activity Test for Black-Hole Waveforms (v2, amended baselines)

Abstract. We present a fully pre-registered pipeline testing whether the post-Newtonian waveform angle θPN predicts the NRSur7dq4 / SEOBNRv5PHM mismatch across 53 GWTC-1–4.1 binary black hole events. The primary GLS regression under a 1/f³ fractional Gaussian noise model (Hurst H= 0.989, MLE) finds a statistically significant coefficient (p=3.4×10−5) but fails the held-out RMSE criterion against three polynomial mass baselines. A chirp-mass tercile split reveals that the 1/f³ long-memory signal is concentrated in the low-mass half (H=0.963), not the high-mass closure regime (H=0.510, near white noise). A theoretical E=mc³ correction battery (four operationalizations, Bonferroni-corrected) fails uniformly (0/4 pass). We interpret the low-mass spectral signature as a coherent, chirp-mass-ordered PN phase accumulation effect rather than a closure-grammar phenomenon.