Hepatoblastoma, the most common malignant liver tumor in children, has benefited from improved survival rates in recent decades through cisplatin-based chemotherapy, surgical refinement, and risk stratification models such as CHIC-HS. However, a subset of patients still experiences relapse or poor outcomes, often due to primary or acquired chemoresistance. This work aimed to identify novel biomarkers predictive of both treatment resistance and prognosis, and to better understand the developmental and molecular basis of tumor heterogeneity and evolution. Using a large collection of post-chemotherapy hepatoblastoma surgical samples, we integrated genomic, transcriptomic, histopathological, and clinical data to refine prognostic stratification. While canonical Wnt/β-catenin alterations are nearly ubiquitous but non-informative for outcome, we identified rare driver mutations (including NFE2L2 and NF1) in 37% of cases, significantly associated with relapse, mortality, and chemoresistance. We also demonstrated that the SBS35 mutational signature, linked to the tumor’s ability to repair cisplatin-induced DNA damage, and the 'Liver Progenitor' and 'Immune Cold' transcriptomic subtypes, were also strongly associated with poor response to neoadjuvant chemotherapy and predict poor outcomes independently of clinical parameters at diagnosis. Spatial transcriptomic analysis further revealed that embryonal tumor architecture correlates with the 'Liver Progenitor' subtype, suggesting that histopathology could serve as a rapid surrogate for molecular subtyping and guide postoperative treatment. To further dissect hepatoblastoma heterogeneity, we performed single-nucleus RNA sequencing and spatial transcriptomics, revealing that the three major tumor cell populations (Hepatocytic, Liver Progenitor, and Mesenchymal) organized along developmental trajectories and present more diversity of phenotypes. We reconstructed an oncogenic tree reflecting hepatoblastoma cells heterogeneity. These data suggest that plasticity and heritability jointly shape tumor architecture, clonal dynamics, and response to therapy. Finally, early developmental alterations were explored through the identification of clonal 11p15.5 mosaicism in non-tumor liver tissue of a subset of patients, suggesting a potential preneoplastic condition that may precede β-catenin activation and contribute to early tumor development. In summary, this work provides new insights into the molecular and developmental mechanisms underlying hepatoblastoma progression and resistance. It identifies clinically relevant biomarkers and highlights the value of integrating genomic, transcriptomic, and histopathological data to improve risk assessment and therapeutic strategies in pediatric liver cancer