Intervention-Disclosure Visibility and Time-Bounded Selective Nondisclosure in Hierarchical Agent Systems

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

This paper treats intervention disclosure in artificial agents with persistent state as a contract over observers, fields, channels, and time. It distinguishes disclosure assignment, delivery, awareness of an intervention's existence, and knowledge of particular fields, and separates the effect of a disclosure policy from effects mediated by awareness using common external outcomes. For a decision problem with the same loss in which a more informative policy can imitate a restricted policy at no additional cost, a finite-case proof shows that optimal risk cannot increase. This result narrows the conditions under which a benefit from nondisclosure can be claimed. The framework connects leakage through combined observers and retained histories, cumulative duration and reclassification of stable families, review of active nondisclosure, and obligations at specified deadlines to the audit and execution lifecycle. It defines maintenance outcomes that account for resumed support and approval evaluations that use the relevant state, with competing hypotheses and falsification conditions. The results are conditional arguments and verification specifications; they do not establish actual safety or efficacy, a universally optimal disclosure mode, or the permissibility of interventions on humans. Publication role. Companion B (Version 1.0, dated 20 September 2026, about 9,100 words; unsubmitted) develops the disclosure and obligation branch of the Integrated Framework series on contract-preserving lower-to-upper recalibration (SHDA). The flagship and its Technical Supplement are archived separately, as are Companion A on residual genesis and dynamic feedback route attribution and Companion C on family-scoped capability control, typed lineage and atomic re-entry. The formal scope is artificial agent systems; the selection rules do not apply to undisclosed interventions on humans, which require separate consent, rights, and legal and ethical review. No empirical study or deployment result is reported.