Agent Genesis: lineage, sandbox and authority-free probation

Begin with an evidenced gap in the available view

The Agent Genesis reference starts from an evidenced need, not from a general desire to create more agents. The candidate search has a declared scope.

For an illustrative specialist request, record what capability is missing and which candidates were considered. “No match in this bounded view” does not prove that no suitable agent exists anywhere. That distinction keeps creation decisions tied to inspectable evidence.

A candidate profile is not productive authority

The documented lifecycle separates a synthesized attenuated profile, reproducible adversarial sandbox, probation, review and later admission. Sandbox and probation do not grant membership, work or action authority.

In an operator interface, show a candidate as under evaluation until the relevant admission transitions occur. Avoid displaying a probation result as if the new lineage already owned a production task. Evidence about the candidate and permission to use it are separate records.

Retain lineage and bounded creation conditions

Genesis requires parent and ancestry references, bounded depth and population budgets, evidence for capability additions and authority attenuation. Independent certification and explicit review conditions are also part of the documented boundary.

The profile and operator reference explains the relationship between catalog, derived and synthesized profiles. A profile remains input to governed creation; it is not itself an identity or a grant of permissions.

Cross the owning subsystems explicitly

Factory creation, membership, attestation and work handoff remain identifiable parts of admission. The governance reference keeps those responsibilities with their existing owners.

Trace an approved candidate through the actual receipts. If creation succeeds but another required boundary does not, retain the partial state and follow the configured compensation or retirement path. Do not compress that sequence into an optimistic “agent ready” label.

Evaluate the new lineage without promising improvement

• Record the capability gap and searched candidate scope.
• Inspect ancestry, ceilings and evaluation evidence.
• Verify that probation has no productive authority.
• Follow review into the separate admission boundaries.
• Measure later work under a declared deployment scenario.

A candidate passing the governed process does not establish universal performance superiority. The operational value of its later work still needs evidence from the environment where it is used.