A fixed 37.90-day term improves Gl 725 B velocity prediction in three omitted years under three noise models with published correction vectors

We tested whether two reported Gl 725 B velocity terms predict an omitted year. We used 205 public SPIRou measurements and the four published WAPITI correction vectors. We fixed the periods at 4.765 and 37.90 days and fitted the offset, trend, correction coefficients, signal phase, and signal amplitude together. We integrated the coefficients and tested three noise models. The 37.90-day term improved the joint log score in each of 2019, 2020, and 2021 under all three models. The descriptive score sums were 13.55, 10.54, and 4.97 nats for white noise, rotation covariance, and short memory covariance. The 4.765-day term reduced the score for 2019 and increased it for 2020 and 2021 under all three models. The long term also scored above two fixed annual aliases and the nominal third rotation harmonic in each omitted year. Local period integration and tested nuisance changes kept the main score signs. The data processing, correction vectors, and period choices used the full record. These conditional tests do not give an independent planet discovery or a false alarm probability. We provide code, input hashes, score tables, figures, and numerical checks for reuse.

Gl 725 B · SPIRou · radial velocity · exoplanets · WAPITI · predictive density · Gaussian process · blocked validation · annual alias