Past flight dates gave small and uncertain gains in British butterfly index forecasts in 2005-2024

We tested whether past flight timing adds information about next-year butterfly abundance beyond two years of population history. We used public UK Butterfly Monitoring Scheme 2024 records. Models fitted through 2004 forecast 2005-2024 at selected established British sites. Adding prior-year first, mean and last flight dates reduced equal-species mean squared log-index error by 0.83% across 29 species and 25,510 test triples. A three-year test-block range was 0.23% to 1.42%, conditional on fixed fits. The 2015-2024 gain for 27 species after fitting 1995-2014 was 0.47%, with range -0.07% to 1.06%. Both full-model ranges crossed zero when training-block weights also varied. At sites with complete test coverage, point gains were -0.35% and -0.54%, with wide ranges. That check selects persistent recorded populations and does not prove a site-turnover cause. Half of the timing-model addition retained small positive training-weight sensitivity ranges. Thus measured timing may carry a small signal, but these data do not establish a robust general gain. These tests overlap and reuse one monitoring system. Whole-season dates mix broods. Selected sites, shared count errors, final-release inputs and short test periods limit inference. Code, exact source hashes, aggregate errors, numerical checks and figures are included. External records and joined databases remain outside the baseline bundle.