The carbon isotopic composition of ecosystem CO₂ exchange provides information on the physiological and environmental controls of land–atmosphere carbon fluxes. Here, we use continuous CO₂ and δ¹³C–CO₂ measurements from five inlet heights at the Amazon Tall Tower Observatory (ATTO) in central Amazonia to derive apparent δ¹³Cs source signatures from vertical-profile Keeling-plot relationships without prescribing an external background. The resulting continuous 2.5-year record from 2022 to 2024 substantially extends previous campaign-based and irregular measurements and provides a unique observational basis for resolving source signature variability in the Amazonia rainforest over time. The observations show pronounced variability on diurnal and seasonal time scales. The diurnal cycle is characterized by more enriched source signatures during the day (-28.5 to -27.5 ‰) and more depleted values during the night (-29.5 to -28.5 ‰). In contrast, the seasonal cycle showed less variability, with an amplitude of approximately 1.5‰. We compare the observed temporal variability of the inferred source signatures with source signatures simulated by the process-based model QUINCY. The observed diurnal cycle is broadly consistent with the sign and timing of the QUINCY response. Although QUINCY is systematically more depleted than the observations, the comparable diurnal amplitude suggests that the model captures the short-term functional response. On seasonal time scales, however, QUINCY simulates substantially stronger variability and much tighter relationships with environmental drivers than observed. These results indicate that the current model representation of isotope discrimination and environmental responses cannot fully explain the observed seasonal behaviour. The weaker observed seasonal response to environmental drivers suggests that Amazonian carbon–water–isotope coupling may be more buffered than represented in QUINCY, for example through canopy microclimate or deep-water access. Overall, the continuous multi-year ATTO record provides a valuable data set for evaluating Amazonian vegetation dynamics and identifies key areas in which process-model representations of isotope discrimination and water-stress responses require improvement.