Scalar variance restoration raised held-out daily SST-change errors at 67 OISST moorings in 2022-2025

We tested five scalar corrections of OISST SST increments against matched buoy increments. Whole-year holdouts in 2022-2025 used 67 moorings and a 2016-2019 reference. Restoring earlier OISST daily increment variance raised median centred mean square error by 28 percent. Matching buoy variance raised it by 32 percent. Regression calibration lowered it by 6 percent, with a larger benefit at tropical moorings. A forward-year check retained higher variance-restoration error. These scalar tests differ from full spectral correction. They show that matching variance is not sufficient for measured-change accuracy. Buoy assimilation, point-to-grid differences, retrospective anomaly fits, short eras and conditional uncertainty limit the result. No global climate trend or latent sensor error is estimated. Code, fitted gains, held-out scores, source hashes and checks are included.