# Lifecycle, events and execution Understand how an organization starts, waits, resumes and preserves responsibility between turns. Concept ID: axiomao:lifecycle-events-execution Language: en Status: reference Reference edition: 0.1 Canonical: https://axiomao.org/en/lifecycle-events-execution/ Published: 2026-10-14 ## A conversation is one part of a lifecycle An organization can exist while no model session is active. Its identity, outstanding commitments and pending events remain in durable state. Distinguish initialization, management ownership, an individual working turn, waiting, pause and termination. Opening a repository or reading its instructions does not initialize or launch an AxiomAO instance. Similarly, an idle launcher and a stopped organization are different operational conditions that should be visible to an operator. ## Delivery is not completion An event records something the organization may need to evaluate: a message, deadline or result. Receiving it does not prove that the corresponding responsibility has been fulfilled. In AxiomAO’s batch protocol, receive, start, result and done are distinct transitions. The saved result identifies all included events and evidence of actions. If execution stops between result and done, the next manager reconciles the saved outcome before acknowledging the same batch. ```text Documented batch protocol; placeholders are not executable wake receive wake start wake result --event-ids --actions wake done ``` ## Choose a wake-up mechanism deliberately Interactive operation depends on a user or client returning control. A service launcher can wait for events and start bounded native turns. Scheduled work needs absolute deadlines and a policy for missed intervals. AxiomAO documents a separate service process and local ownership rules; it does not imply that every model remains continuously active. A useful status view distinguishes queue depth, active management, running workers and the launcher’s ability to deliver another turn. ## Pause requires process awareness Pausing new runtime transitions is not automatically equivalent to stopping every native process. An operator must know which workers are still running, what effects they may produce and whether shutdown was confirmed. The current launcher documents process identity and reconciliation mechanisms for its managed sessions. These local controls do not establish distributed fencing or tenant isolation. Resume only after outstanding actions and ownership are understood, not merely after clearing an error flag. ## Sources - [Axiom AO — source documentation excerpts (2026-10-14)](https://axiomao.org/reference/implementation-evidence.md) ## Related - https://axiomao.org/en/recovery-observability-audit/ - https://axiomao.org/en/operating-an-organization/ - https://axiomao.org/en/implementation-and-validation/