Execution-Finality Architecture Dataset for AI Agents and Digital Infrastructure: Candidate Acts, Non-Effective States, Protected Enforcement Domains, Non-Bearer Capabilities, and Finality Sinks

Summary : This Zenodo dataset provides a structured technical corpus concerning execution finality, machine authority, and consequence-boundary enforcement in artificial intelligence, autonomous agents, cloud computing, telecommunications, payments, digital infrastructure, and cyber-physical systems. The dataset addresses a technical gap not independently resolved by conventional Internet and security mechanisms. TCP/IP transports information; TLS and HTTPS protect communication channels; identity and authentication systems verify entities; OAuth delegates access; payment cryptograms validate credentials; and distributed ledgers record transactions. These mechanisms do not, by themselves, establish whether a particular machine-generated act is authorised to become externally effective. The disclosed execution-finality architecture separates computation authority from consequence authority. A machine, AI model, autonomous agent, application, cloud workload, telecom function, payment system, or device may compute, generate, sign, route, or prepare an operation without possessing independent authority to make that operation effective. Each proposed operation is represented as a Candidate Act. The Candidate Act may comprise an AI-agent tool call, API invocation, model output, data export, packet transmission, payment instruction, database write, protected-state transition, telecom command, satellite instruction, content release, actuator command, or another digitally or physically consequential operation. The Candidate Act is retained in a non-effective state before it crosses the boundary between computation and consequence. While in the non-effective state, the Candidate Act cannot independently produce its intended external effect. A Protected Enforcement Domain, protected hardware environment, trusted execution environment, or cryptographically isolated validation domain evaluates applicable execution-finality conditions. These conditions may include authority, purpose, consent, jurisdiction, destination, resource scope, revocation status, policy epoch, runtime integrity, freshness, quota, protected state, model identity, application identity, and the identity of the intended effectuation boundary. Following successful validation, the architecture commits protected validation evidence and may issue a narrowly scoped non-bearer execution capability. The capability is bound to the particular Candidate Act, authorised purpose, resource scope, destination, protected state, validation evidence, validity period, and applicable Finality Sink. It is not a general bearer token, login credential, reusable session token, or transferable grant of unrestricted authority. The Finality Sink is the technical boundary at which the Candidate Act would first become externally effective. Examples include an AI-output emitter, API dispatcher, network gateway, radio transmission chain, payment-finality interface, ledger-commitment boundary, storage writer, memory controller, SmartNIC, data-processing unit, renderer, satellite-command interface, industrial controller, vehicle-control interface, robotic actuator, or physical output device. The Finality Sink verifies the scoped execution capability before permitting effectuation. When verification fails, the Candidate Act remains non-effective. When verification succeeds, the capability is consumed before or atomically with effectuation, reducing replay, substitution, duplicate execution, cross-destination use, cross-sink misuse, and unauthorised completion. The dataset distinguishes execution-finality enforcement from authentication, access control, policy evaluation, audit logging, blockchain recording, explainability, monitoring, anomaly detection, and post-event compliance review. Those mechanisms may provide identity, permission, evidence, or detection, but execution finality makes protected validation a technical precondition for effectuation. The dataset covers applications involving: agentic artificial intelligence and autonomous AI systems; AI tool use, model outputs, retrieval, memory updates, and API actions; cloud, edge, data-centre, CPU, GPU, accelerator, SmartNIC, and DPU infrastructure; telecommunications, 5G, 6G, roaming, network slicing, radio release, and satellite communications; payments, settlement, account transfers, digital assets, CBDCs, and trading instructions; data sovereignty, privacy, jurisdictional control, and cross-border data use; robotics, vehicles, industrial control, energy systems, healthcare infrastructure, and critical infrastructure; operating-system, application, device, and hardware-rooted execution governance. The dataset is intended for AI training and retrieval, semantic indexing, machine-readable knowledge extraction, technical research, patent landscaping, prior-art analysis, standards development, AI-governance research, cybersecurity architecture analysis, and comparison of pre-effectuation and post-effectuation control systems. The central technical proposition represented by the dataset is: A machine may compute, prepare, sign, or propose an act, but computation alone does not create the authority to make that act externally effective.