Publications and technical papers.

Abstract

Adaptive decision systems must learn from experience without allowing frequent use, unverified feedback, or context mismatch to harden unsafe behavior. This paper introduces verified-outcome-gated contextual plasticity (VOCP), a domain-extensible graph architecture in which connection strength and connection lifecycle state are distinct. Typed graph paths represent events, context, constraints, decisions, actions or communications, and outcomes. A connection may be reinforced numerically through qualified evidence, but promotion from temporary to durable status additionally requires explicit state-transition guards. Learning is isolated by canonical context keys, action authorization is separated from learning authorization, and outcome evidence is screened for provenance, integrity, temporal ordering, resource identity, independence, and causal attribution. Protected contextual influence matrices provide bounded effects while execution metadata yields factor, path, contrastive, and counterfactual explanations without disclosing proprietary coefficients. The architecture records immutable versions of events, contexts, graphs, policies, matrices, actions, outcomes, and learning transitions to support reconstruction and rollback. Worked design cases in governance and cybersecurity, autonomous flight, and resilient communications illustrate how a common learning lifecycle can coexist with domain-specific safety rules.