This study presents an integrated computational framework for predicting early-stage building energy performance under conditions of uncertainty and incomplete design information. By combining BIM-based geometric abstraction with probabilistic scenario generation, reduced-order thermal simulation, and sensitivity-informed variance analysis, the framework demonstrates that conceptual design decisions can yield structured, interpretable performance insights. The methodology identifies dominant parameter influences, interaction effects, and performance gradients, revealing that geometric determinants such as orientation, glazing ratio, and shading depth exert disproportionate control over early energy outcomes. Findings show that conceptual performance is statistically patterned rather than random, and that prediction reliability increases when uncertainty-aware modeling is applied. The study contributes a methodological advancement by demonstrating how BIM-integrated workflows, uncertainty quantification, and sensitivity analysis can operate collaboratively to support informed conceptual decision-making. These results establish a foundation for predictive reasoning in early-stage design and provide new opportunities for performance-driven architectural practice.