Recovery, observability and audit
Recover uncertain work, inspect organizational behavior and audit the evidence behind decisions.
#Unknown is an outcome worth preserving
A process can fail after an external effect occurs but before the result is recorded. Treating that state as failure and retrying may duplicate publication, payment or another consequential action. Preserve the uncertainty, inspect authoritative evidence and reconcile before continuing. AxiomAO documents an unknown-outcome approach for actions and recovery of saved event results. A local record alone cannot guarantee exactly-once effects across an unrelated external service.
#Observe responsibilities, not just uptime
A healthy process can coexist with a stuck organization. Observe unresolved commitments, overdue events, blocked actions, management ownership, active workers and time since meaningful progress. Link these signals to evidence and the next responsible actor. A successful HTTP response, zero exit code or empty error log proves only a narrow mechanism. It does not show that the requested work was accepted, delivered or retained after a restart.
#Audit binds a claim to its evidence
An audit finding needs a precise claim, scope, evidence version and resolution condition. If the supporting artifact changes, an earlier conclusion may need review. The documented v0.3 runtime introduces versioned findings and evidence checks rather than assuming every attachment permanently closes an issue. Audit should preserve disagreement and uncertainty. It is not automatically a business veto, a legal certification or proof that tools outside the observed path obey the same rules.
#Reconstruct before resuming
A recovery checkpoint should identify the last accepted artifact, pending obligations, live processes, unknown actions and next safe transition. Verify the old owner’s process identity before reclaiming management. Back up state using a consistent procedure appropriate to the storage engine; SQLite documents a dedicated backup API. Test restoration separately from backup creation. Reopening a database or restarting a service is not sufficient evidence that the organization resumed without losing or duplicating responsibilities.
Sources & context
Reference edition 0.1 · 2026-10-14. Read the editorial method.