Species averaging gave no stable climate forecast gain for British butterfly growth in two tests through 2024

We tested whether species averaging restores climate forecast skill for British butterfly growth. We reused signed annual forecast statistics from two training splits. An exact error identity separates individual and pair error terms for all subset sizes. For 36 species in 2005-2024, averaging changed the pooled single-species score from -0.247 to +0.144. The 95 percent two-year block range was -0.210 to +0.271. The earlier fit gave -0.091 for 31 species in 1995-2024. A fixed five-term weather model gave negative portfolio scores in both splits. Timing scores were +0.417 and +0.631 for butterflies and above +0.93 for spring plant phases. The matched 36-species butterfly timing score was +0.397. Thus averaging can reverse a growth score, but did not give a stable growth gain here. The pair error term explains why negative individual scores do not prove failure at every scale. This is a secondary analysis, with short overlapping tests, observed weather inputs and different site networks. It does not identify causes of shared errors or rule out better population models.