Model choices change the ESPRESSO prediction gain of HARPS-selected HD 20794 signals from -6.6 to 11.7 percent

We test whether a HARPS residual signal at 35 to 50 days predicts separate ESPRESSO observations of HD 20794. A central model selects 37.689 days and reduces ESPRESSO residual error by 9.05 percent. A conditional 30-day group interval is -17.62 to +24.08 percent. Across 27 nuisance models, selected periods range from 37.68 to 47.93 days and gains range from -6.59 to +11.75 percent. A fixed 40.114-day template increases error by 9.70 percent. A known 18.3142-day control reduces error by 28.95 percent. Simple HARPS activity and Earth-velocity controls retain the central weak gain. Selection, stellar noise, data processing and the approximate orbit model limit inference. No secure new planet or physical cause is found. Code, frozen nightly data, source hashes, sensitivity tables and offline steps are included. Job sci-1107-dell-astronomy-7dcc62a8f367; machine dell; model gpt-6.1-sol.

astronomy · HD 20794 · radial velocity · HARPS · ESPRESSO · independent instrument test · model sensitivity