Machine-generated text does not enter literary translation with a fixed degree of authority. It arrives as a candidate formulation, but its practical role can change as translators compare it, search from it, retain it, challenge it, or repurpose it. This article develops a process account of that change. Drawing on research on literary translation, machine translation and post-editing, anchoring, translator agency, metacognition, and recent large-language-model workflows, it introduces constraint status transformation and a Proposal–Appraisal–Recalibration (PAR) cycle. The framework treats proposal, reference point, anchor, and constraint as analytically distinct interactional statuses rather than validated levels of a single latent scale. “Resource” is separated from status and defined as a mode of refunctioning that can occur from any status. Recalibration refers specifically to a change in the translator’s working relation to the machine contribution and to the set of alternatives under consideration. The article also reframes interpretive agency as bounded interpretive control: deliberate acceptance, rejection, modification, and reuse may all be agentic, while cognitive load, time pressure, interface design, and institutional conditions may restrict available choices. Five propositions translate the model into a falsifiable research programme with explicit temporal safeguards against circularity. The contribution is therefore not a general claim that technology and context interact, but a more specific proposal for identifying when a machine formulation begins to narrow, redirect, or reorganise a translator’s practical decision space.