Glossary — AI Product Deployment Specification (AIPDS)
For Product-family and semantic terms, see the AI Product Glossary and AIPROD Glossary.
Deployment Environment
The operational environment in which an AI Product version executes or is accessed, including cloud, on-premises, edge, embedded/physical-control, hybrid, or agent runtimes.
Execution Specification
The AIPDS declaration of supported execution modes and operational characteristics such as latency, throughput, statefulness, deadlines, action boundaries, or control-loop requirements.
Execution Mode
How intelligence is operationally invoked or runs, aligned to AIPROD MO values such as Batch, Real-Time/Event-Driven, Synchronous API, Interactive, Embedded, Continuous, or Agentic/Goal-Driven.
Inference
Execution of a learned or other intelligence mechanism to produce an intelligence effect. Model inference is one specialization of AI Product execution, not the universal AIPDS abstraction.
Learning Mode
How, if at all, the deployed Product updates intelligence through learning. AIPDS supports NONE, PERIODIC_RETRAINING, TRIGGERED_RETRAINING, ONLINE_LEARNING, and HUMAN_GOVERNED_LEARNING.
Evolution Mechanism
A governed way the Product realization may change, including model retraining/replacement, rule update, knowledge update, prompt/policy change, tool update, upstream dependency update, or software/runtime change.
Training Orchestration
An optional governed activity used only where model training/fine-tuning is part of the Product's lifecycle. It is not mandatory for all AI Products.
Observability Binding
Reference to runtime monitoring/evidence configuration appropriate to the Product's operational, intelligence, behavioral, security, and safety characteristics.
Safety Binding
Deployment control or evidence binding for embodied/actuating or otherwise safety-relevant AI operation.
Runtime Dependency
A versioned operational dependency required by the deployed Product, for example an upstream AI Product, Model Product, AI Primitive, Data Product, knowledge service, tool, or runtime component.
Risk Tier
The canonical AIPS effective risk tier R0–R4. AIPDS operationalizes controls required by the tier; it does not define a separate risk taxonomy.
Composability
Operational ability to integrate or orchestrate AI Products. Composition becomes a new Derived AI Product only when AIPROD/Core establish a distinct governed intelligence proposition.
Lifecycle State
Operational maturity state such as Experimental, Development, Production, Deprecated, or Retired.
Core distinction: Evolution is universal; retraining is conditional. Execution is universal; model inference is conditional.
Governance Obligation Profile
The version-specific computed set of obligations produced by Core/AIPROD risk and governance analysis. AIPDS binds this profile through governanceProfileRef.
Governance Control Binding
An AIPDS mapping from one governance obligation (for example ActionAuthorization or EmergencyStopSafeState) to the operational control that realizes it in a deployment.