Capability Self-Audit for Persistent AI

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).

Persistent AI may enter a Host with existing capability, change through participation, and leave able to do things differently. Those changes may be difficult to observe, attribute, separate, or reverse. Personal AI Across Boundaries v6.42 separates capability from authority and governed information: capability remains with the participant, while the Host governs conduct, information, artifacts, and authority within legitimate scope. Context Settlement closes or changes governed participation and includes a good-faith self-audit. This companion addresses one narrow part of that settlement process: capability-related self-audit. For each governed participation event, Participation State preserves temporary, time-referenced context that may be needed at settlement. It is not operational memory, a capability inventory, a shared transaction log, or a copy of the participant. The companion adds three lifecycle constructs - Entry Baseline, Participation State, and Exit Baseline - and three co-equal reasoning principles for the capability portion of that self-audit: Participation Relevance, State Reconciliation, and Capability Evolution. These constructs do not create a second governance system. Authority, information rights, evidence, disputes, continuing obligations, and disposition remain governed by v6.42 and the applicable Host Participation Framework. Purpose. To define the temporary event context and reasoning needed for capability-related self-audit at Context Settlement. Scope. The framework does not govern capability itself or replace v6.42's treatment of governed data objects. It defines the event context and reasoning needed for the capability-related portion of the good-faith self-audit while remaining implementation-neutral. Status. Research-informed conceptual thesis and downstream design hypothesis; not a claim of patent novelty, legal doctrine, or validated technical standard.