Reproducibility is widely treated as evidence of reliable causal inference in environmental epidemiology. Yet in environmental time-series studies, exposures, behaviours, and mortality often share strong seasonal structure, making causal inference highly dependent on modelling assumptions. Temperature, air pollution, and mortality co-vary systematically across the calendar year, creating tightly coupled temporal patterns that limit empirical separability between competing explanations. In such settings, reproducible associations may partly reflect shared analytical conventions for representing seasonality rather than uniquely identifiable causal effects. Using daily mortality data from Stockholm, Sweden, we show that estimated associations for temperature and nitrogen dioxide (NO₂) vary substantially across alternative specifications of seasonal adjustment, even when all other model components are held constant. Estimated effects attenuate or shift as progressively stronger seasonal controls are introduced, indicating that models redistribute overlapping seasonal variation between correlated predictors. These findings suggest that the central challenge is not simply confounding, but limited identifiability within highly correlated seasonal systems. Reproducibility should therefore be treated as a starting point for inference rather than definitive evidence of causal isolation. Greater transparency regarding modelling choices and sensitivity to seasonal specification is essential for responsible interpretation of environmental epidemiological evidence.