SALT19 develops EvoMind as both software and a research program. The public record tracks the architecture, memory integrity, semantic state change, experience compounding, capability composition, and the validation evidence used to test those ideas.
Research boundary: EvoMind is an experimental AGI-oriented governed cognitive architecture under active validation. SALT19 does not present it as externally certified AGI or as proof of unrestricted general intelligence.
Each item is listed by its deposited type and DOI. Public descriptions stay close to what the record is intended to establish rather than treating publication itself as capability proof.
The public software record for EvoMind as a governed cognitive architecture spanning reasoning, planning, memory, and desktop-agent operation.
A separately deposited architecture research object focused on persistence, verification, governed execution, and experience-driven agentic autonomy.
Introduces a research direction for belief-gated state transitions and repair-invariant memory so long-horizon agent state changes can remain explicit, inspectable, and resistant to silent corruption.
Positions governed experience compounding as a systems hypothesis: persistent deployment experience can become reusable capability only when evidence, authorization, verification, and revocation remain explicit.
Documents governed semantic capability composition: interpreting user intent into required outcomes, resolving available capabilities, producing work products, and grounding the result in verification and artifact identity.
EvoMind research asks how reasoning, state, memory, planning, action, verification, and learning can behave as one persistent governed system rather than as disconnected model calls.
Maintaining goals, beliefs, uncertainty, conversation context, and task state across longer-lived work.
Preserving useful experience without allowing repair, contradiction, or state mutation to silently rewrite what the system believes happened.
Converting high-level goals into required outcomes and capability graphs while keeping planning separate from execution authority.
Separating cognition, permission, execution, verification, rollback, and audit so capability does not automatically imply authority.
Testing whether repeated success, recovery, correction, and bounded transfer can become reusable capability through explicit evidence and revocable promotion.
Moving beyond text generation into desktop environments where intent must bind to actual applications, controls, artifacts, and observable outcomes.
SALT19 separates architecture hypotheses from software demonstrations and internal acceptance evidence. A paper can define a mechanism or systems thesis; the validation record documents what has actually been exercised in the runtime.
EvoMind investigates AGI-relevant systems problems such as persistent cognition, autonomous reasoning, durable memory, planning, semantic capability composition, desktop action, verification, experience reuse, and governance. The current evidence remains bounded internal engineering evidence. It does not establish unrestricted transfer, universal task competence, independent external replication, or certified AGI.
Open questions and unresolved authority paths remain part of the published validation posture rather than being hidden behind a product label.