Infrared radiofluorescence (IR-RF) dating of K-feldspar is commonly performed on multi-grain aliquots, implicitly treating the IR-RF dose response curve (DRC) and its apparent saturation behavior as properties of a sample. Here, we test the alternative hypothesis that bulk DRC curvature and plateau dose can emerge from mixtures of grains with contrasting mineralogy and luminescence behavior. We combine (i) single-grain, multi-spectral RF measurements (710, 850, 880 nm) on feldspar reference materials spanning the ternary diagram, (ii) controlled mixed-grain aliquot experiments using a detection window centered at 850 nm, (iii) spatially resolved IR-RF (SR IR-RF) imaging of natural samples (X7363, X7368) using an RF<sub>70</sub> style protocol and grain-scale DRC classification, and (iv) synchrotron-based µ-XRF mapping at 10 keV to assess grain-scale chemistry and evaluate surface versus interior controls.</p> <p>Reference measurements confirm strong dependence of IR-RF behavior on feldspar type and polymorph. Albite and microcline commonly show high initial intensities and decaying DRCs at 850–880 nm, whereas Ca-rich feldspars are weak and near-flat, and sanidine exhibits variable, sometimes increasing DRCs. Mixing experiments demonstrate that adding a small number of grains with contrasting behavior measurably shifts microcline DRC curvature and reduces apparent saturation. SR IR-RF reveals that both natural samples tested contain grains with decreasing, increasing, and near-flat regenerative DRCs, and that polishing produces minimal change in DRC shape, arguing against a purely surface-controlled origin for anomalous behavior. Qualitative 10 keV µ-XRF maps show that grain-scale elemental detections and spatial patterns do not uniquely predict the DRC category, implying that IR-RF behavior is not controlled by composition alone. For dating purposes, these results imply that saturation limits and equivalent dose estimates should not be inferred from averaged signals in heterogeneous samples, particularly those of volcanic origin. We propose a practical two-step screening workflow that combines DRC-shape classification with the dose corresponding to 95 % of the fitted asymptotic level (D95) to prioritize high-capacity grains. This approach now requires validation on independently well-dated and/or stratigraphically constrained sequences.