ARCHITECTURE OF AN OBSESSIVE AI

DESCRIPTION This article presents a formal architecture for goal-oriented artificial intelligence systems, based on the separation between persistent search, verifiable recognition of the objective, mission termination, and return to the initial state. The construction is based on a state machine in which the system remains in search mode while the arrival criterion has not been satisfied: C(Xt,G)=0.C(X_t,G)=0. When the objective is recognized according to the conditions established by the architecture, the transition occurs: C(Xt,G)=1,C(X_t,G)=1, followed necessarily by the protocol: TARGET→STOP→RETURN→CAPSULE→AMBER→GREEN→RED→END.\text{TARGET} \rightarrow \text{STOP} \rightarrow \text{RETURN} \rightarrow \text{CAPSULE} \rightarrow \text{AMBER} \rightarrow \text{GREEN} \rightarrow \text{RED} \rightarrow \text{END}. The architecture therefore establishes a formal distinction between searching, finding, confirming, and terminating. The work integrates components previously developed within the research program, including the Primitive Architecture, the Unified Transduction Architecture, the Autonomous Transduction Machine (ATM), and mechanisms for memory, coverage, refinement, branching, recovery, and auditing, together with falsification and verification procedures. The previous architecture had already established the sequence: LATENCY→STRUCTURE→GENERATION→STATE→REFINEMENT→MEMORY→COVERAGE→BRANCHING→RECOVERY→AUDIT.\text{LATENCY} \rightarrow \text{STRUCTURE} \rightarrow \text{GENERATION} \rightarrow \text{STATE} \rightarrow \text{REFINEMENT} \rightarrow \text{MEMORY} \rightarrow \text{COVERAGE} \rightarrow \text{BRANCHING} \rightarrow \text{RECOVERY} \rightarrow \text{AUDIT}. The present formulation adds a specific mission-control layer, responsible for determining when exploration must cease and when the system must enter the terminal protocol. The arrival criterion is refined through an Arrival Certificate, in which the produced result must be associated with its corresponding evidence, history, verification, and audit. The architecture also incorporates a falsification stage before a solution is accepted as a completed result. Thus, completion is no longer defined solely by the existence of a candidate result. It depends on a structured chain: OBSERVATION→HYPOTHESIS→FALSIFICATION→REPRODUCTION→VERIFICATION→AUDIT.\text{OBSERVATION} \rightarrow \text{HYPOTHESIS} \rightarrow \text{FALSIFICATION} \rightarrow \text{REPRODUCTION} \rightarrow \text{VERIFICATION} \rightarrow \text{AUDIT}. The article also formalizes properties of terminal safety, conditional liveness, non-return to search after completion, monotonic termination, and absorption of the final state. In particular, the following property is established: C(Xt,G)=1⇒no new search action within the same mission.C(X_t,G)=1 \Rightarrow \text{no new search action within the same mission}. The work further distinguishes the mission cycle from the agent cycle. Mission termination does not imply elimination of the agent. After the terminal protocol, the agent returns to the capsule state and remains available for a new mission: MISSION→END,\text{MISSION}\rightarrow\text{END}, while: AGENT→CAPSULE.\text{AGENT}\rightarrow\text{CAPSULE}. This separation makes it possible to represent persistent intelligence without transforming persistence into indefinite execution of the same task. The architecture is presented as a theoretical-computational construction, with formal properties that can be implemented and subjected to experimental testing. Computational validation is intended to verify the machine transitions, the behavior of the arrival criterion, the integrity of the history, search coverage, the effectiveness of falsification, and, above all, the absence of search actions after completion has been declared. The resulting architecture therefore treats persistence not as an end in itself. Persistence remains active while the objective has not been confirmed and is necessarily replaced by termination when the arrival condition is satisfied. The central formulation can be summarized as: SEARCH→FIND→CONFIRM→TERMINATE\boxed{ \text{SEARCH} \rightarrow \text{FIND} \rightarrow \text{CONFIRM} \rightarrow \text{TERMINATE} } followed by: STOP→RETURN→CAPSULE→END.\boxed{ \text{STOP} \rightarrow \text{RETURN} \rightarrow \text{CAPSULE} \rightarrow \text{END}. } The work therefore proposes an artificial intelligence architecture in which freedom of search, a verifiable arrival criterion, formal termination, and operational return constitute components of a single system. KEYWORDS Artificial Intelligence; Autonomous Intelligence; Transductive Architecture; Autonomous Transduction Machine; State Machine; Persistent Search; Arrival Criterion; Verification; Falsification; Audit; Memory; Coverage; Mission Control; Termination; Return; Autonomous Systems. CONTRIBUTION TYPE Theoretical-computational construction; formal architecture; state-machine model; mission verification and termination protocol; proposal for implementation and experimental validation. STATUS Preprint / Research Article. The architecture presented constitutes a formal and computationally testable formulation. The structural properties derived from the state machine are presented as formal results of the construction; the practical effectiveness of the system depends on implementation, controlled experimentation, independent reproduction, and validation across different classes of tasks.