Run age gave no clear three-month RONI exit forecast gain at 1981-2023 origins

We tested whether time already spent in a warm or cold ENSO index state improves the forecast of its first exit. We used fixed CPC ERSSTv5 RONI and ONI tables. Each fit used only target windows that ended by the issue center. The primary test has 316 RONI forecasts from 32 threshold runs at 1981-2023 origins. Adding log run age and a cold-phase interaction to a strength, change and season model raised mean Brier error from 0.10019 to 0.10215. Error reduction was -1.96%, with a conditional 95% five-year block interval of -13.08% to 6.88%. All eight planned index, threshold, horizon, penalty and curved-reference checks also had intervals that included zero. This gives no clear added forecast gain for the tested age terms. It does not reject ocean memory or measure official ENSO event endings. Revised and rounded indices, overlapping targets, few runs, model form and conditional uncertainty limit inference. The package gives original code, aggregate evidence, uncertainty draws, a figure, source hashes, a separate numerical audit and instructions to retrieve immutable external inputs. Raw and record-level external data stay outside the signed package.

ENSO · RONI · ONI · run age · first exit · probability forecast · Brier score · chronological test · null result