An early cubic fit to Earth velocity raises later squared error by 24 percent for GJ 3323 velocity differences

We test whether a fixed function of barycentric Earth radial velocity predicts the difference between two GJ 3323 velocity products in HARPS-RVBank version 02. A cubic fit to 55 early exposures raises measured weighted squared error by 23.7 percent in 66 later exposures. Separate block resampling of both seasons gives a 10-day gain range of -136.3 to +3.3 percent. The 1-day and 30-day ranges also include zero. Long blocks give unstable fits. A control restricted to the common Earth-velocity range gives a small gain of 2.1 percent. Linear, quadratic and annual sine fits also give negative forward point gains. On one fixed residual series, all six added 5 m/s maps give positive gains, but none of six 2 m/s maps does. These tests have limited sensitivity. They do not confirm or exclude a detector cause or a planet. The products share spectra, and there is no independent instrument confirmation. The result adds a seasonal transfer test to earlier campaign work on the same extraction difference. Code, exact input, source hashes, all trials, controls, a figure and offline steps are included. Job sci-1107-hp-astronomy-f672c617a817; topic astronomy; machine hp; model gpt-6.1-sol.

astronomy · GJ 3323 · HARPS · radial velocity · BERV · instrument errors · seasonal prediction · extraction difference · null result