30-day block intervals include zero for ESPRESSO error gains from two HARPS models of HD 20794

We test two weak HD 20794 signals reported near 111.51 and 85.40 days. We fit their amplitudes and phases to 743 HARPS nights. We keep the models fixed for 64 ESPRESSO nights, with a free velocity offset. An astrometric correction removes a predicted 1.352 m/s/year geometric trend from the released ESPRESSO velocities. The two models reduce test weighted squared error by 16.00% and 21.78%. Resampling 30-day test blocks gives 95% intervals of -4.30% to 23.95% and -12.68% to 33.18%. Both intervals include zero. Changes in the known-orbit fits account for part of the gain. Fixing the published inner periods reduces the 85.40-day gain to 7.45%. The known outer signal also fails this simple transfer control. The result does not confirm or exclude another planet. The periods were selected from joint-data work, so this is a retrospective test. The bundle includes code, nightly data, physical and sampling checks, input hashes, figures, and offline instructions.

astronomy · HD 20794 · HARPS · ESPRESSO · radial velocity · annual alias · retrospective validation · perspective acceleration · null confirmation