Digital-twin terminology increasingly frames the patient as a computational entity whose changing state can be reproduced, predicted, and tested under simulated interventions. In pharmacotherapy, however, no model can fully reproduce the biological, experiential, behavioral, social, and organizational conditions through which medication effects emerge. Treating a pharmacotherapy digital twin as a replica may consequently conceal model selection, incomplete data, disagreement, uncertainty, and the distribution of decision authority. This article proposes Negotiated Pharmacotherapy Digital-Twin Theory, an original non-empirical account in which the twin is understood as a partial, task-specific, and revisable model assembled through structured interaction among clinical evidence, computational inference, and patient-reported knowledge. The proposed architecture comprises three inspectable model layers, a negotiation workspace, a disagreement register, an intended-use contract, a permission and human-control layer, provenance-preserving updates, and conditional representations of alternative futures. Negotiation does not require consensus or imply that every representation is equally credible. Instead, it makes assumptions, conflicts, uncertainties, and authority boundaries visible before a model is accepted for a specified medication decision. Evaluation would require technical verification, external and temporal validation, calibration and uncertainty assessment, subgroup analysis, human-factors testing, workflow evaluation, patient-concordance assessment, clinical-utility studies, and postdeployment monitoring. The theory does not establish predictive superiority, safer prescribing, improved outcomes, regulatory acceptability, or deployment readiness. Its original contribution is to replace the ontologically strong replica metaphor with a governed model of representation in which the patient remains distinct from—and able to contest—the digital account constructed for pharmacotherapy.