Posture is generated, not held. Multiple independent research programs in pain science, postural neuroscience, and trauma-informed clinical practice have recently converged on a framing that treats postural and persistent symptom phenomena as products of distorted predictive processing rather than of structural pathology. The Fit-for-Purpose model (Wand et al. 2023) proposes that chronic nonspecific low back pain represents a state in which patients hold intransient internal models of a damaged back. Two randomized clinical trials published in JAMA in 2022 provide RCT-level evidence for non-pharmacological back pain interventions from different theoretical traditions: the RESOLVE trial (Bagg et al. 2022) for graded sensorimotor retraining (1.0-point between-group pain reduction at 18 weeks, meeting the minimum clinically important difference threshold), and the SPINE CARE trial (Choudhry et al. 2022) for Individualized Postural Therapy via the Egoscue Method (-4.3 Oswestry Disability Index points at 3 months and-6.7 points at 12 months, n=2,971). A parallel research program in mild traumatic brain injury (Kontos et al. 2026, JAMA Network Open) has produced similar evidence at the randomized trial level. We propose the Generative Posture Framework (GPF), a unified computational account under which these convergent findings become instances of a common principle: the body schema generates postural output as the precision-weighted prediction of an active inference process operating on whatever sensory precision the nervous system currently has available. This framework offers the first conceptual bridge we are aware of between Porges's polyvagal theory and Friston's active inference framework, integrating autonomic state dependencies into the precision-weighting architecture as a testable proposal rather than a formal mathematical derivation. Within GPF, the family of phenomena the framework addresses is referred to as inference problems: conditions in which the body schema generates a distorted postural prediction from degraded sensory input. Postural conditions including adolescent idiopathic scoliosis, kyphosis, chronic pain-associated postures, post-traumatic postural freezing, age-related postural collapse, and post-concussive postural change represent different instantiations of the same computational architecture under different sensory precision and autonomic state configurations. A set of operative conditions governs when the body schema revises its postural prediction, organized into three functional categories: five prerequisite principles that must be substantively present for slow-loop update to occur, one interference mechanism (efference copy suppression) that actively blocks update when triggered by high-effort voluntary correction (Blakemore et al. 2000; Shergill et al. 2003), and two amplifiers – co-regulated relational context (Porges, Feldman, Dumas, Goldstein) and Interoceptive Attention – that scale update magnitude when available without being strictly necessary for update to occur at all. The fifth prerequisite, Interoceptive Attention, is dual-classified as both prerequisite (4.5) and amplifier (4.6b): as prerequisite, it gates cortical plasticity through the nucleus basalis cholinergic mechanism, since unattended repetition produces negligible representational change (Recanzone et al. 1992; Kilgard and Merzenich 1998); as amplifier, it scales the magnitude of change when present, with attended training producing 1.5- to 3-fold cortical representational expansion beyond the baseline enabled by the other prerequisites alone. Ten testable predictions with proposed study designs are included. The framework is offered as a complement to the clinical traditions of the past century rather than as a replacement.