Implementation and validation
Read AxiomAO capability claims through source versions, execution paths and bounded test evidence.
#Keep three kinds of claims separate
A proposed architecture states what should exist. An implementation claim identifies a mechanism in a particular version. A validation claim reports observed behavior in a defined environment. These claims cannot substitute for one another. Current project documentation describes the local native-session runtime as v0.3.0-dev. This page summarizes inspected documents and code; preparing the website did not rerun the organization, its native agents or historical validation scenarios.
#What the local prototype contains
The inspected profile records identity, sessions, tasks, messages, commitments and events in SQLite. Role memory uses applicability checks and epochs. A separate launcher delivers event-driven native turns. The documented v0.3 work adds grants, common file-tool receipts and versioned audit findings. These mechanisms have specific boundaries: a native tool outside the common API is not automatically constrained by that API, and a local session lock is not a tenant-isolation mechanism.
#Historical evidence belongs to its revision
VALIDATION.md records a real Codex governance scenario on source revision 8d68ba1, alongside earlier v0.2 service scenarios on other revisions. It describes separate workers and reviewers, granted file operations, retained receipts, accepted work and completed event batches. These are reported historical results, not fresh tests of the website or every later change. The document also states that real loading of new native hooks was not established in that run; synthetic hook-envelope tests have narrower scope.
#Limits determine the next experiment
The documented profile supplies no container image or automatic container deployment. Earlier fifteen-minute service observations do not prove day-long operation or host reboot recovery. Broad multi-engine equivalence, independent tenant isolation and enforcement across every native tool path require separate evidence. A useful next report states the source revision, engine version, environment, exact scenario, retained artifacts, failures and claims it does not support. A green test count alone cannot establish autonomous organizational effectiveness.
Sources & context
Reference edition 0.1 · 2026-10-14. Read the editorial method.