OASI: Operational Artificial System Intelligence — An Organismic Computing Architecture for Body-Bound Runtime Assurance and Developmental OS–AI Integration

Operational Artificial System Intelligence (OASI) is the canonical name of a research paradigm investigating a system-level architecture in which operation, artificial embodiment, memory, cognition, authority, and development are coordinated through one versioned causal history and constitutionally mediated effects. In this name, operational refers to system operation, not production readiness, and system intelligence names a research target rather than achieved general or superintelligence, consciousness, deployment, external validation, or superiority. This preprint formalizes the architectural hypothesis as operational monism and combines a computational body model, developmental memory, typed PolicyIR, obligation-checked compilation, bounded reflexes, and AERA, a bounded commit-time authority-revalidation predicate. In this corrective version, AERA is a stable project identifier rather than a normative acronym expansion. The evidence remains internal and deliberately bounded. The accompanying release preserves a negative T4 result and two local deterministic effect-boundary simulations comprising 2,700 retained records across 90 cells. S5 did not establish an advantage over a cooperative idempotent receiver. S6 identified a retry-versus-omission safety-availability tradeoff under a non-cooperative fixture, without isolating an OASI-specific advantage. The 30 repetitions per cell are implementation-stability traces, not independent population samples, and the fault cases are simulated control-flow traces rather than physical failures. This v0.4.1 preprint is a terminology correction of v0.4; it does not reclassify any code behavior or experimental result. This work does not demonstrate a complete OASI operating system, general superiority, production readiness, universal effect guarantees, real-fault validation, performance advantage, or independent replication. Development and project-controlled internal adversarial review were extensively AI-assisted; they are not human peer review, external certification, institutional evaluation, or independent scientific review.