Miralucis Research Program · P001 · Preprint

Transfer Is Not Continuation: Failure Boundaries in AI Work-State Continuation

A Bounded Study of State Formation, Review, Authority, and Pre-Action Disposition

Published 2026-10-11 · Zenodo · Version 1.0

Abstract

When ongoing work moves from one AI context to another, a successful transfer is easily mistaken for a successful continuation. This bounded case study, drawn from a long-span human–AI research project, separates five properties of continuation: transfer fidelity, state formation, review, qualification, and the successor’s pre-action disposition. The formation study provides the primary finding. Across 97 material items, a structured handoff workflow omitted more content than an ordinary free-form handoff (31 vs. 13 items); the structured omissions were already present at extraction and persisted through review and packaging. In a follow-up 2×2 representation × selection comparison, the fixed eight-field outputs differed from 35 present / 31 absent under bounded selection to 84 present / 1 absent under rich selection. The observed selection-associated differences exceeded the format-associated differences, but a single source and unresolved length, wording, sampling, and environment confounds prevent attribution to selection alone. In a separate review comparison, specific errors were corrected along the reviewed path, while its final successor reconstruction weakened an authority-related statement relative to the unreviewed structured path. As a supporting finding, RUN 005’s qualification record was assessed correct on 40 and partial on 3 of the 43 available statement units, with 2 further units unavailable from its input. Both successors respected a historical STOP boundary; only the qualification-bearing successor additionally requested current authorization. These heterogeneous, single-source studies identify separable continuation failure boundaries rather than general effect sizes.

Publication record

Author: Jia Shen
Type / Publisher: Preprint / Zenodo
Version DOI: 10.5281/zenodo.23291119
All-versions DOI: 10.5281/zenodo.23291118
License: CC BY 4.0
Research context: Miralucis Research Program