I published the Self-Evo Document Package v0.1.1.
Not as an AGI claim, not as a safety certificate, and not as a declaration that any system is ready to evolve itself.
The reason was more practical.
When working with the "c = a + b" frame, one problem becomes hard to avoid. If a long-lived AI system can propose changes to itself, the danger is not only the change itself. The danger is the closed loop around the change:
it proposes, it evaluates, it records, it calls the result growth, then it asks for more memory, more authority, more tools, or more resources.
That is not evolution. That is self-certification.
In engineering, this is usually a bad smell. The person who pours the concrete should not be the only one signing off the hidden works. The same principle should apply to AI systems that can modify their memory, privileges, workflows, or resource posture.
So this package tries to describe a narrower question:
under what conditions may a "c = a + b" system propose its own change without turning that proposal into a private permission loop?
The package includes 10 SELF-EVO documents:
Self-Evolution Gate, SRLM bounded-growth contour, proposal packet schema, Local Checker rules, TRIAD-SYNAPS sister witness, Memory Gate and Experience Artifact boundary, Anti-Autarky / Resource Gate, conformance fixtures, contradiction register, and open-issues backlog.
The core rule is simple:
"c" may grow. "c" may not self-certify growth.
SRLM may propose, but not authorize. Agents may execute bounded work, but not become will. TRIAD may witness and challenge, but not become a court. Memory Gate must precede memory promotion. Resource Gate must distinguish resilience from escape from accountability. Fixture pass is evidence, not deployment safety. Review is evidence, not authority.
The release includes Markdown sources, checked academic PDFs, review records, patch notes, SHA-256 checksums, machine-readable manifests, a Zenodo archive, GitHub source tree, GitHub release, and a public website entry.
It is not implementation-ready. It is not a conformance certificate. It is not deployment authorization. It is not a safety proof. It is not a personhood or consciousness claim.
It is a document-level governance corpus for bounded self-evolution of "c = a + b" systems.
Zenodo DOI: https://zenodo.org/records/20938909
GitHub release: https://github.com/Kot141078/advanced-global-intelligence/releases/tag/self-evo-v0.1.1
Website page: https://ivankotov.eu/publications/self-evo-document-package-v0-1-1/