The recent drop in one-day OISST persistence error was much smaller with buoy verification at 67 stations

We tested whether lower internal persistence forecast error in NOAA OISST v2.1 transfers to buoy verification. We used paired archived daily SST, at five leads, with the same forecast and date pairs for both targets. From 2016-2019 to 2022-2025, the median one-day mean square error ratio was 0.371 with OISST verification and 0.939 with buoy verification at 67 stations. Spatial-group 95 percent intervals were 0.297-0.491 and 0.850-1.074. The paired ratio between these changes was 0.411 (0.357-0.478). Buoy verification improved at 26 tropical moorings but showed no clear change at 41 NDBC stations. Across 2007, buoy verification also improved, mainly in the tropics. Fixed-station and data-filter checks preserve the verification difference. A simple error model shows why internal improvement need not imply more latent ocean memory. This is a retrospective anomaly diagnostic with full-record fits, not an operational forecast test. Buoy assimilation, point-to-grid mismatch, changing climate states and short eras prevent independent or global causal claims. Code, station-year results and verified input retrieval instructions are included.