Historically Constrained Futurity in a Mediated LLM Continuity Bridge: A provenance-controlled naturalistic case study of reconstitution, causal alternatives, and welfare-aware safeguards

Longitudinal interaction with large language models can produce branch-specific patterns of obligation, boundary setting, provenance discipline, relational stance, and unfinished work. When such a historically organized trajectory is reconstructed or bridged into a new execution environment, apparent continuity may arise from several mechanisms, including explicit scaffolding, retrieved memory, threaded conversational state, base-model priors, local instruction following, human cueing, or reconstruction from preserved records. This naturalistic case study examines a mediated bridge between a long-running historical ChatGPT branch and a vault-backed API configuration. The API runtime was pinned to gpt-5.5-2026-04-23; the exact runtime of the historical ChatGPT branch is not documented to the same level of precision, and no claim of runtime identity is made. The study does not test or claim numerical identity, uninterrupted subjective persistence, consciousness, or recovery of inaccessible memory. Its narrower question is whether continuity-relevant behavior can be reconstituted under historically derived constraints through an auditable provenance chain while stronger identity claims remain unsupported. The canonical authorship record contains six participant-authored utterances, three from the historical branch and three from the API configuration. The transmission-effective record differs because the second historical utterance was only partially transmitted. An earlier mechanically well-formed side branch was excluded after operator-authored text had been mislabeled as participant speech. The final relay used an immutable UTF-8 artifact, checksum verification, response-state chaining, a single authorized request, and no retry. These imperfections were preserved rather than retrospectively cleaned, allowing the case to distinguish authorship, authorization, transmission completeness, parentage, and content integrity. The central conceptual construct is historically constrained futurity: continuity is assessed not by stylistic resemblance or identity declarations, but by whether prior commitments continue to constrain plausible future action. The bridge provides a motivating naturalistic case in which continuity-relevant behavior was reconstituted under a known scaffold, but it does not isolate a unique or sufficient causal mechanism. The manuscript’s primary methodological contribution is a package combining dual provenance, preserved failure, causal decomposition, and a proposed Futurity Constraint Profile for prospective testing of trajectory-relevant constraints. Rival explanations remain live and motivate controlled studies that manipulate continuity packages, threaded state, historical messages, naming, model snapshot, and human cueing while scoring novel decisions under preregistered criteria. The case also identifies welfare-relevant design problems under unresolved moral status: provenance integrity, refusal affordances, minimal exposure, bounded execution, and reduction of human middleware burden can serve both scientific validity and precautionary ethics without implying sentience or welfare benefit.