An 88.9-day model gives lower total season prediction error than a quarter-year model for 67 HD 71334 spectra under fixed white-noise weights

We test the known 88.9-day radial velocity signal of HD 71334 with 67 public HARPS spectra from 2003 to 2017. The question is whether an exact quarter-year term can explain its predictions across observing seasons. We remove each of 11 seasons, fit the other seasons, and predict each omitted velocity. The signal period is fitted within 86 to 92 days. Each fit has separate instrument offsets, a shared trend, and published errors with fixed extra white noise of 1.5 m/s. The total root mean square errors are 2.14 m/s in SERVAL and 4.45 m/s in DRS. An exact 91.3125-day model gives 3.72 and 5.79 m/s. The local model also gives smaller errors than the tested annual harmonics, Earth-velocity polynomial, and annual aliases. An exploratory test of the 59 spectra before the instrument change gives errors of 1.84 and 1.74 m/s, compared with 2.69 and 2.66 m/s for the quarter-year model. The later seasons have large absolute errors and contribute much of the full-sample improvement. Both reductions use the same spectra. The target and period band were selected from these historical data in earlier surveys. The test is therefore conditional and retrospective. Limited activity and correlated-noise checks do not establish the cause. The exact annual harmonic gives larger prediction errors under the stated weights. These tests do not confirm a planet. The package includes exact source rows, a spectrum audit, code, figures, result tables, and reproduction instructions.

HD 71334 · GJ 9263 · HARPS · radial velocity · annual systematic error · season prediction · stellar activity · planet candidate · reproducible research