Weather regression restored ozone forecast skill with an old temperature slope at 30 rural US sites in 2022-2025

We tested whether weather calibration can restore forecast skill without an ozone response update. We crossed a fixed 2003-2007 ozone slope or a recent five-year slope with bias correction or regression calibration of archived GFS temperature. All fits used earlier summers. On 16,594 matched site-days at 30 eastern US sites in 2022-2025, regression changed old-slope MSE reductions from -0.9% to 13.6% with 24-hour weather inputs, and from -9.8% to 10.0% with 48-hour inputs. The reference used previous ozone and weather. Recent slopes with regression gave 18.3% and 15.0% reductions and beat the old slope in every summer. Input calibration recovered useful skill but did not remove the value of a response update. Four summers, final observed history, one selected network and uncertain archive delivery times limit operational inference. Code, source hashes, aggregate errors and numerical checks are supplied; source records remain external.