OISST v2.1 input steps bias 1982-2025 trends of daily SST extremes and marine heatwaves; cold-spell lengthening is an artifact and heatwave-duration trends depend on product noise

Question: daily NOAA OISST v2.1 lost most of its day-to-day variance in steps at satellite-input changes (1985, 2007, 2020, 2021). Which standard daily climate statistics does this bias, and which changes remain? Approach: for 29,120 ice-free pixels (1-degree subsample, 60S-60N, 1982-2025), we made three homogenized records. In each, the sub-monthly spectrum (periods of 32 days or less) of every input segment matches one target segment: 1985-2006 (noisy), 2007-2020 (main, near buoy variance) or 2021-2025 (smooth). We repeated the Hobday marine heatwave (MHW) and cold spell (MCS) analysis (1983-2012 baseline) and tested against 172 moored buoys on the same days. Results: at the buoys, raw OISST event duration relative to buoys increased after the input steps (paired ln-ratio change +0.069, p = 0.002); homogenization removed this drift (+0.022, p = 0.38). For all three targets, the raw 1982-2025 warming trend of the annual maximum daily anomaly is 42-44% too low (+0.108 against +0.19 K per decade), the trend of the annual minimum is 39-45% too high (+0.29 against +0.20), 83-100% of the MCS duration rise (+0.83 days per decade) is artifact, and the raw MHW count trend is 0-27% too low. Two mechanisms explain this: noise breaks events, and a noisy baseline raises the thresholds. For MHW days they almost cancel (raw +21.2, homogenized +18.1 to +24.1 days per decade). MHW duration trends depend on the target (raw +3.7; homogenized +1.7 to +3.9 days per decade), so they are not comparable across products with different noise. After detrending, the narrowing of daily extremes is entirely artifact; a smaller lengthening of events (MHW +0.5 to +1.2, MCS +0.5 to +1.1 days per decade) remains. Limitations: the true pixel-scale sub-weekly variance is unknown; smaller steps remain; buoys feed OISST and cover few regions; the last segment is short and includes the 2023-2024 warm event.