Magma degassing models typically invoke volatile depletion of a single parental melt, with permeable loss of exsolved gas having served for many years as the paradigm for the transition from volatile-rich, explosive eruptions to volatile-depleted lava flows. These degassing models are guided by measurements of H2O, CO2, and hydrogen isotope variations retained in melt that quenched to glass, but the existing models are not uniquely constrained by the data. There also remains uncertainty surrounding the origin and significance of volcanic glass fragments. We show that individual obsidian pyroclasts from Mono Craters, California (USA), are heterogeneous in dissolved H2O and CO2, suggesting that clasts are assembled from juvenile melt and rewelded ash during magma ascent. This is in contrast to the conventional view that clasts are chemically homogeneous and sample the chilled, glassy margins of conduit walls. The new measurements of dissolved H2O and CO2 help reconcile existing open-system degassing models used to explain elevated CO2/H2O ratios, provide time scales based on diffusion modeling for pyroclast formation, and show that magma does not necessarily lose volatiles monotonically during ascent-driven decompression.