Prior short-season rain gave no clear added forecast skill for March-May rain in an East Africa box in 2000-2025

We tested whether rain in the preceding October-December season adds forecast information for March-May rain in 5 S-5 N, 35 E-40 E. Each fit used earlier years from 1980. The baseline used February local rain, February Nino 3.4, February DMI and a time trend. Across 26 forecasts in 2000-2025, adding prior short-season rain changed mean squared error by -0.12%. Positive values mean lower error. The conditional 95% three-year block interval was -13.34% to 9.60%. All one-, three- and five-year block intervals included zero. Root mean squared error was 1.131 mm/day for the baseline and 1.132 for the added model. The earlier-year target mean had lower error, 1.051 mm/day. Thus the primary null result depends on a baseline with no demonstrated gain over that simple reference. Checks of trend, lead month, region and score period gave small or negative added gains. A short record, one small box, revised and interim data, model choice and conditional intervals limit inference. The result does not establish absent physical memory or independent drought risk. The package gives source hashes, monthly evidence, forecasts, code, saved draws, a figure and separate numerical checks.

East Africa · rainfall · seasonal prediction · seasonal persistence · null result · GPCP · ENSO · Indian Ocean Dipole