Drug–drug interaction knowledge is distributed across molecular assays, enzyme and transporter studies, pharmacokinetic models, clinical investigations, computational predictions, knowledge bases, and prescribing alerts. These resources frequently describe different portions of an interaction while using incompatible entities, relation types, temporal assumptions, evidence labels, and levels of causal interpretation. Consequently, a molecular perturbation may be represented as equivalent to an exposure change, a predicted association may appear indistinguishable from an observed interaction, and a pairwise alert may omit the dose, timing, patient state, or physiological process that determines its relevance. This article proposes a causal grammar for representing drug–drug interactions as evidence-bearing chains extending from a context-specific molecular initiation event through enzymes, transporters, disposition processes, systemic or tissue exposure changes, and bounded clinical consequences. The grammar distinguishes interacting-drug roles, molecular mechanisms, biological mediators, pharmacokinetic transitions, contextual qualifiers, evidence assertions, contradictions, uncertainty, and applicability boundaries. It is designed to preserve direction, magnitude, temporal sequence, dose and regimen conditions, anatomical location, genotype, current metabolic phenotype, organ function, inflammation, and other patient-specific modifiers without reducing them to a single interaction score. The proposed construct also separates molecular plausibility, computational prediction, mechanistic explanation, quantitative simulation, clinical observation, and decision relevance as different evidence states requiring different forms of validation. Its purpose is not to generate dosing recommendations or establish clinical causality autonomously, but to provide a coherent semantic architecture for evidence integration, mechanistic reasoning, model documentation, and context-aware decision support. The grammar remains conceptual and requires formal ontology testing, pharmacological adjudication, quantitative evaluation, interoperability assessment, and prospective human-centered validation before operational use.