A shared nightly error model gives variance ratios consistent with one in 32 short Ross 128 HARPS pairs

We test quoted velocity errors in 32 disjoint Ross 128 HARPS exposure pairs less than one hour apart. Independent corrected errors give squared-change to variance ratios of 0.477 for SERVAL and 0.500 for DRS. A model with one fully shared nightly correction error gives 1.197 and 0.817. Night-resample 95 percent ranges are 0.626 to 1.951 and 0.341 to 1.903. The error columns show that the shared model cancels almost all added correction variance. This is a tested covariance assumption, not a measurement of true covariance. Pairing changes and the small sample limit inference. Thirty-one pairs come from one late season. The known planet changes little within these pairs. No new planet, physical signal or instrument fault is found. The bundle contains code, aggregate results, source hashes, a figure and external retrieval steps. Source database records are excluded.

astronomy · Ross 128 · GJ447 · HARPS · nightly calibration · error covariance · short exposure pairs · null result