High eccentricity and fixed correlated noise reduce the 81-day coefficient mismatch in GJ 393

We tested whether 176 HARPS nights and 75 CARMENES nights of GJ 393 can share the coefficients of an 81-day Keplerian orbit. Earlier work tested fixed source curves. We allowed period, phase, amplitude, orientation and eccentricity to change. We used four fixed Gaussian error models, with and without time covariance from stellar activity. We saved 8118 orbit shapes, local fit refinements and a denser eccentricity check. At the source period, a circular orbit gives coefficient mismatch Q = 36.721 under independent errors with 1.4 m/s extra noise. A fixed activity covariance with 0.7 m/s extra noise gives Q = 16.280 for that circular shape. Its best common fit in the tested range has period 80.805471 days, eccentricity 0.783266, semi-amplitude 1.241 m/s and Q = 4.474. At one fixed true shape and known Gaussian covariance, the conditional 95% reference is Q = 5.991. Selected fits do not have global significance from that reference. Removing the 38.53-day nuisance terms gives best-common Q = 4.320. The known 7.02679-day control gives Q = 0.975 and amplitude 1.059 m/s. The high-eccentricity fits pass close to the known inner orbit; stability is untested. The result measures how orbit shape and stellar noise change a joint coefficient check. It does not confirm a planet or exclude all 81-day orbits. The package preserves raw input rows, hashes, code, all grid values, covariance matrices, figures and Gaussian calibration checks.

GJ 393 · Gl 393 · radial velocity · HARPS · CARMENES · stellar activity · eccentricity · orbit coherence · Gaussian covariance · reproducible astronomy