Accurate precipitation data for the Tibetan Plateau (TP) is critical for understanding regional water resources and global climate dynamics. However, existing datasets struggle with observational bottlenecks due to complex terrain and high precipitation variability. Here, we introduce the CRISP (Conformal Regression and Integrated Stacking for Precipitation) dataset, providing a continuous, 27-year (1998-2024) daily precipitation record at a 0.1° spatial resolution for the TP. Unlike conventional statistical datasets, CRISP integrates physical atmospheric conditions and terrain features into a machine-learning framework. The CRISP dataset effectively identifies false drizzle signals commonly seen in reanalysis data without missing most real rainfall events. Furthermore, CRISP can provide reliable 90% uncertainty intervals that adaptively adjust to the intensity of rainfall compared to other datasets, and it explicitly classifies precipitation into rain, snow, and mixed phases to directly support cryospheric research. The independent validations indicate that CRISP shows potential in reducing inconsistencies associated with the “Third Pole precipitation paradox” (Miao et al., 2024). By providing consistent precipitation estimates together with uncertainty information and phase classification, CRISP offers a valuable basis for hydroclimatic research over the Third Pole and its downstream regions. The CRISP dataset is openly available at (<a href="https://doi.org/10.11888/Atmos.tpdc.303469">https://doi.org/10.11888/Atmos.tpdc.303469</a>; Yong & Lyu, 2026).