An older winter gave uncertain added RONI forecast gain in 1981-2024, while ONI error fell

We tested whether an older winter ENSO state adds next-winter forecast information after the current spring state and its change are known. We used fixed CPC ERSSTv5 RONI and ONI tables. Each spring issue used only earlier target years in its fit. There were 44 target winters in 1981-2024. Adding the older winter reduced RONI mean squared error by 7.52%. The conditional 95% five-year block interval was -1.47% to 15.88%, so the primary test gives no clear gain. The ONI check gave a 12.97% reduction, with limits of 1.44% to 23.36%. All planned RONI checks had intervals that include zero. ONI intervals also included zero under a 30-year fit and when three extreme targets were omitted. The result tests delayed index information, not a physical cause or real-time skill. Shared SST data, revised reference periods, 44 winters, model form and conditional intervals limit inference. Fixed public domain inputs, exact source hashes, all forecasts, code, a figure and numerical audits are included.

ENSO · RONI · ONI · delayed memory · seasonal forecast · chronological evaluation