A shared correction gave small uncertain butterfly growth forecast gains in two tests through 2024

We tested whether one correction slope per response class can repair old climate forecasts more reliably than separate unit slopes. We reused signed annual statistics for Swiss plant dates and British butterfly dates and growth. Each correction used ten years after the original fit. We then tested later years against a no-climate model with the same updated unit means. For models trained through 2004, the shared correction reduced butterfly growth error by 3.5% in 2015-2024. Separate clipped slopes gave -0.8%. For the older model vintage, the shared growth gain was 5.5%. Calibration and test years were resampled in common year blocks across units. The evidence does not establish a stable growth gain. Plant spring timing had a much larger point gain. This retrospective transfer test extends an earlier exploratory shared-slope result. The test periods overlap. Observed weather, changing source coverage, short calibration records and fixed original fits limit inference. The result does not identify a causal climate effect or a biological limit to prediction.