A free scale on each released correction curve keeps at least 67% of annual circular signal information in 105 CARMENES series under the stated diagonal weights

We test how the released CARMENES correction curves change annual velocity fits and their information. The sample has 11,659 paired RVC and AVC measurements for 105 targets. Each target has at least 50 finite pairs over at least 730.5 days. We fit an offset, a linear trend, and a circular signal at 365.25 days. All paired fits use the same fixed diagonal weights, with the AVC error and a 2 m/s error floor. The correction changes the annual coefficient vector by more than 1 m/s in 68 series. The median vector change is 1.210 m/s. When one scale on the full applied correction curve is free, the smallest information ratio over signal phase ranges from 0.670551 to 0.998117. Its median is 0.943529. The common published nightly table gives a minimum ratio of 0.642541. In a separate model, a free Earth velocity term leaves less than 0.125% at some annual phase in every series. Thus a large measured coefficient change can occur with a small information cost when one scale on a known correction shape is free. The bound depends on sampling: a 30-exposure threshold gives a minimum of 0.4611. We preserve inputs, source hashes, per-target coefficients and information matrices, 13 controls, code, and figures. A fresh run gives identical results. These are conditional model results. They hold the correction shape fixed and omit correlated noise. They do not establish a planet or an error in the official calibration.

CARMENES · radial velocity · nightly zero point · annual signal · calibration · Fisher information · public data · reproducible research