Half-unit temperature bounds permit both richness slope signs in four Canada and US bird samples in 1985-2024

We tested whether bounded temperature error can change the sign of an adjusted temperature-richness slope. We used 86,443 accepted bird survey runs from Canada and the US in 1985-2004 and 2005-2024. Four separate models control for observer-route, state-year and survey conditions. Temperatures within half one reported scale unit permit negative and positive slopes in all four samples. The smallest uniform Celsius error limit that reaches a zero numerator ranges from 0.0012 to 0.0207 C. All four nominal sampling intervals also include zero. Exact bounds use fixed controls and errors that can align with outcome residuals. They do not measure actual error, give confidence bounds for latent-temperature slopes, or show a causal weather effect. Selection, roadside sampling, missing data and dependence limit transfer. The package contains code, source hashes, selected records, endpoint checks, a figure and offline reproduction results.

biodiversity · observation bias · temperature precision · partial identification · Breeding Bird Survey · measurement error