Shared slope correction gave small uncertain gains for British butterfly growth forecasts in two decades from 2005 to 2024

We tested whether a smaller climate forecast signal can repair failed butterfly growth forecasts. We reused signed annual statistics for models fitted through 1994 or 2004. Corrections used ten later years and were tested on the next decade. A separate slope for each unit, constrained to zero through one, gave no clear growth gain over a recent mean. An exploratory slope shared across species kept 25 to 39 percent of the original signal. It gave small positive error reductions in 2005-2014 and 2015-2024. All growth intervals included zero. A fixed species sample gave the same early point result. Spring plant timing gave clear added value, while butterfly timing did not have a clear advantage over growth after the baseline update. An exact error decomposition showed that negative frozen skill can coexist with positive test-fitted linear skill. That bound is not a feasible forecast. This study adds constrained correction and time-transfer tests to prior mean-correction and species-averaging tests. Ten test years, inherited site means, observed weather and exploratory method development limit inference. The package gives code, source hashes, all scores, uncertainty checks and a scratch-only reproduction script.