A five-shift fit reduced 2024-2025 buoy-change error where a variance match raised it at 30 OISST moorings

We tested whether local corrections of NOAA OISST v2.1 improve measured buoy temperature changes in later years. We fitted each station in 2022-2023 and scored 2024-2025 at 30 moorings. A variance match raised median mean squared error by 40.7 percent for one-day changes and 9.9 percent for seven-day changes. A five-shift ridge fit reduced error by 9.1 percent and 10.8 percent. The error ratios were 0.909 (conditional 95 percent spatial-group interval 0.820-0.933) and 0.892 (0.855-0.946). The fit also beat scalar regression. At 24 common stations, fits from 2016-2017 had 3.3 percent and 6.3 percent more error than recent fits on the same 2024-2025 targets. Both test years showed the main median directions. Removing the gross-pair filter kept the results. Stronger regularization made the seven-day benefit uncertain. This is a new held-out test on an existing daily archive, not independent observations or a new estimator. Buoys feed OISST, the sample has only four tropical stations, and two test years cannot establish global or long-term skill. The fit uses future OISST and a separate filter for each change interval. It does not produce a common corrected SST record or test heatwave trends. Code, aggregate errors, fit parameters, source hashes and reproduction steps are included. Daily source records stay outside this package.