Removal of one older year does not restore fixed-period Ross 128 HARPS transfer under three weight rules

We test whether one older calendar year causes a known Ross 128 signal-transfer failure in HARPS-RVBank version 02. We remove each of ten older years, refit a circular wave to the remaining older data, and test it on 138 later exposures. At a fixed period of 9.8658 days, every removal still increases later error in both reductions under three weight rules. At the main weight scale, year-removal gains range from -77.05 to -35.92 percent in SERVAL and -34.04 to -25.33 percent in DRS. A local period search is less stable and has positive exceptions under a different weight rule. Conditional stable-wave simulations also show transfer failures when the period is fitted. The simulations do not test the physical cause of the observed loss. The reductions share spectra, the intervals span an instrument change, and no independent instrument is fitted. No planet is found or rejected. Code, aggregate results, source hashes and external retrieval steps are included.

Ross 128 · HARPS · year removal · radial velocity · signal transfer · period uncertainty · null result