A phase-lag line-width fit reduces GJ 3543 held-out shape error by 7 percent but lacks support under block noise

We test a fixed phase-lag activity model in 90 public HARPS-RVBank version 02 observations of GJ 3543. Five observing groups each have at least ten visits. An activity sine and its quarter-cycle component predict velocity shape in a group left out of the velocity fit. At 9.161 days, line width reduces total held-out squared error by 7.12 percent. Conditional tail fractions are 0.0063 with one-day signs, 0.0684 with ten-day signs, and 0.0668 with thirty-day signs. Whole-group signs give 0.0625. H-alpha, chromatic index, and all three short-alias fits raise error. Group removal, visit removal, and indicator-error perturbations can reverse the line-width gain. Thus the tested correction lacks stable support. This does not exclude stellar activity or a planet. No new signal or independent instrument confirmation is reported. Original code, aggregate results, a figure, source hashes and external retrieval steps are included. Source measurements remain outside the baseline bundle.

astronomy · GJ 3543 · phase lag · stellar activity · HARPS · cross validation · null result