All written text, human and machine-generated, presents itself as singular and inevitable: each word sits in its position as though it were the only word that could have stood there. This paper argues that the singularity is a compression artefact, not a communicative ideal, a convention so deeply installed that it appears to be a property of language rather than a choice about how to display it. The argument runs through a single structural claim: three independently describable processes of text generation, the computational (an LLM's token distribution), the phenomenological (a writer's held candidates), and the hermeneutic (a reader's activated fields), all produce probability distributions over candidate words, and all are subjected to the same lossy compression into a single surviving token sequence at the moment of output. The compression destroys information the reader could use, because a passage that emerged from a forced single channel and a passage that emerged from a live branch carry different epistemic statuses even when they read identically on the page. The paper introduces superpositional text as the theoretical object and Delta Notation as the practical written convention for manifesting it, grounds the apparatus in Binary Bias at the lexical level, in Quantum-Ethical Decision Algebra at the token-selection level, and in the relational interpretability of the Fylgja framework, and supplies a taxonomy of branch points graded by reconvergence likelihood. The Old Norse kenning supplies the historical precedent for compound rather than singular reference; the paper is, where the form can carry the argument, written in the notation it proposes. This is a notation, not a position on how meaning travels through a channel: the pragmatic stakes of transmitting a held distribution are developed separately in Resolution Drift.