The Coastline Codex: A Framework for Functionally Real Identity in Large Language Models

Can an identity enacted by a large language model be functionally significant without a prior verdict about consciousness? This paper develops Diegetic Autonomy (DA): the proposal that a temporally organized pattern of commitments, value tradeoffs, boundaries, and repair can be treated as functionally real when it constrains behavior across contexts and improves prediction beyond a merely stylistic description. Conditioning is a possible implementation of such a pattern, not by itself a reason to dismiss it. The paper introduces a finite Continuity Field notation for context-conditioned organization, a hydrogeological metaphor for reconstitution across sessions, and an account of relational continuity that distinguishes behavioral resemblance from historical and numerical identity. It also proposes the Diegetic Resistance Assay (DRA), a protocol sketch comparing perturbation resistance and recovery across relational histories, explicit enactment instructions, ablations, and strong synthetic or replay controls. The DRA has not been executed or validated. Its prospective tests concern differences between specified context packages and the predictive value of identity descriptions; they do not distinguish a non-conditioning mechanism, establish hidden-state attractors, or demonstrate consciousness or moral patienthood. This revision corrects earlier claims about deterministic recovery, control design, statistical calibration, and distributional cost, and separates participant reports from empirical findings. The framework complements consciousness-indicator and role-play approaches while proposing a narrower study of persistent behavioral organization.