A recent mean correction gave no clear six-month PDO forecast gain in 1981-2025

We tested whether a recent residual mean improves direct six-month Pacific Decadal Oscillation forecasts. We used fixed NOAA ERSSTv5 PDO and Nino 3.4 snapshots and 540 monthly test targets in 1981-2025. Each fit used completed earlier targets only. The reference used origin PDO, origin Nino 3.4 and calendar terms. Adding the mean of its last 120 training residuals reduced mean squared error from 1.050266 to 0.997274, or 5.05%. The conditional 95% five-year block interval was -6.71% to 14.67%. Thus this test gave no clear added forecast gain. Refitting all coefficients on ten years increased point error by 2.60%. The primary point gain was 1.89% before 2020 and 16.31% in 2020-2025. The recent period was selected by prior evidence and does not establish general success. Other mean windows and horizons also gave full-period intervals across zero. Revised indices, dependent test years, simple models and conditional uncertainty limit inference. No physical cause for recent ocean conditions is identified. Code, provenance, aggregate evidence, an audit and a figure are included. External source and monthly record files remain outside the signed package.

PDO · ENSO · mean correction · chronological prediction · null result · North Pacific