Computational molecular design can generate chemically novel structures and optimize predicted properties at a scale that exceeds conventional manual ideation. Yet the production of valid or high-scoring structures does not resolve the central pharmaceutical problem: a molecule must satisfy multiple interacting requirements before it can become an experimentally credible candidate. Potency may conflict with selectivity, exposure, safety, solubility, stability, synthetic tractability, route practicality, and material availability. Moreover, confidence in a predicted property does not establish that the underlying model is applicable to the proposed structure or that an acceptable synthesis and development path exists. This article develops an original constraint-aware account of molecular design in which pharmaceutical relevance is determined by non-substitutable feasibility conditions, conditional optimization objectives, explicit uncertainty, and evidence-dependent decision gates. The proposed Full-Weight Pharmaceutical Design Hierarchy distinguishes structural admissibility, pharmacological relevance, selectivity and safety boundaries, synthetic and material feasibility, developability requirements, preference-sensitive optimization, and experimental confirmation. It further separates molecule-level plausibility from route-level accessibility and prevents favorable performance on one objective from compensating for failure of a critical constraint. The contribution is conceptual and methodological rather than empirically validated: it organizes how computational outputs may be evaluated, compared, rejected, or advanced without treating a composite score as evidence of pharmaceutical readiness. Its applicability remains conditional on target biology, therapeutic modality, assay quality, model calibration, product concept, organizational capabilities, and access to experimental expertise. By repositioning generation as the production of testable pharmaceutical hypotheses rather than candidate drugs, the article provides a basis for more defensible model evaluation, interdisciplinary decision-making, and prospective validation of constraint-aware molecular design.