Canonical research terminology
Type: Reference
Status: Active
Canonical for: Shared definitions, correctness terms, evidence language and ownership boundaries
Depends on: Reconciled research evidence; definitions do not prescribe a production contract format or PDS API
Core terms
Component contract
The logical, version-linked set of governed facts needed to understand, select, configure, compose, implement and verify a component.
“Logical” means the contract is the coherent agreement those facts form. It is not necessarily one file, record, schema, repository or source of truth. Authoritative facts may remain distributed across design guidance, Figma, code, Storybook, tokens, tests and other governed artifacts.
AI-exposed for an access path
A component is AI-exposed for a stated access path when an agent can reliably discover, retrieve and inspect current governed artifacts describing its design or implementation. Those artifacts need not make selection meaning, usage rules, relationships, constraints or decision logic explicit, so the agent may still need to infer governed meaning.
AI-exposure is an access and inspectability state. It does not establish semantic completeness, correct selection, correct composition, valid implementation or measured agent performance.
An AI-exposure claim should identify the supported client or access path, artifact scope and relevant version or freshness mechanism. A component appearing in model training data, an unversioned webpage or an undiscoverable repository file is not sufficient by itself. Unlike the informal term machine-consumable, AI-exposed is a governed claim about reliable access to current artifacts.
AI-ready for a capability
A component is AI-ready for a stated capability and context when its governed knowledge is explicit, addressable, current, retrievable and checkable enough that an agent need not invent design-system rules. Remaining ambiguity must support clarification, rejection or escalation.
AI-readiness is not an unqualified binary certification and does not prove that agents perform well.
AI-optimised under tested conditions
A component is AI-optimised under tested conditions when agent performance has been measured against an exact contract and artifact version, task set, model, harness and retrieval setup, and meets defined capability-specific criteria. Any claimed improvement must identify the change and its measured effect.
Avoid a permanent or context-free label such as AI-optimised: true.
Machine-consumable
Use only as an informal description of material an agent can access or parse. It is not a separate readiness state and does not imply evaluation.
Relationship among the three AI states
- AI-exposed means the relevant current artifacts can be found and inspected through a stated access path.
- AI-ready means those artifacts and their governed connections make enough meaning explicit for a stated capability and context.
- AI-optimised means performance against an exact version of that knowledge has been measured and meets stated criteria under defined test conditions.
This is a conceptual progression, not a universal maturity score. AI-ready necessarily depends on exposure of the knowledge required for that capability, but AI-exposure alone does not imply readiness. An optimisation result remains limited to the capabilities and conditions actually evaluated.
Meaning and decision terms
Component purpose
The user-facing job for which a component exists.
Interaction/function category — provisional
A provisional category for the kind of interaction performed, such as action, navigation, selection, disclosure or toggle.
The final taxonomy and term remain unsettled. This category is narrower than user or product intent. When describing SAP evidence, retain SAP’s first-party term functional role.
Supported user or product intent
The task or outcome a component can serve in a stated context. It must not be collapsed into interaction/function category, visual hierarchy or an API property named intent without a local definition.
Accessible semantic role
The native or accessibility-platform role exposed to assistive technology. Do not shorten this to “semantic role” where it could be confused with component purpose or product meaning.
Context
Decision-time facts outside the component’s enduring meaning, such as workflow stage, action group, platform, parent, product policy or risk. Contextual fit is the match between those facts and a component’s selection conditions.
State
A current condition such as loading, disabled, selected, expanded or error. State is distinct from enduring purpose, intent, appearance and consequence.
Variant
A system-defined API or design-tool mechanism for alternatives. The term does not name a stable semantic dimension. Its local values must be interpreted from the system’s own property model and guidance.
Tone, impact and severity
System-defined dimensions that require local definitions. Prefer impact or consequence for action risk and status or severity for feedback when that distinction matches the source system. Tone, impact, severity, color and prominence are not universal synonyms.
Relationship
A typed link between concepts or artifacts, such as parent/child, peer, target/control, alternative, replacement or implementation mapping. A mapping establishes correspondence, not selection appropriateness. A composition rule constrains one or more relationships.
Naming levels
Component or tool name
The name of the top-level candidate an agent may select or invoke. Naming has direct causal evidence in evaluated tool/API-method selection tasks. Transfer to design-system component selection remains adjacent.
Property or parameter name
The key under which a value is supplied, such as variant, tone or appearance. Parameter-name and schema-key effects have narrower direct evidence in their tested tasks. Effects on design-system property interpretation remain adjacent.
Enum-value name
One legal symbol inside a finite property, such as primary, secondary or danger. The current research has no verified direct experiment that isolates design-system enum-value names while holding property name, descriptions, behavior, tasks and available concepts fixed.
Semantic naming
The study of how component/tool identifiers, property/parameter identifiers and enum-value identifiers express or obscure their governed meaning. These identifier classes must be analyzed separately.
Correctness terms
Structural validity
The output conforms to declared syntax, type, enum or schema. Structural validity does not establish that the correct legal option was chosen.
Semantic correctness
The selected component and configuration fit the governed design rule and supplied context.
Behavioral correctness
The implementation performs the intended action, navigation, state transition or side effect according to its implementation contract.
Visual fidelity
The rendered result resembles the intended visual output. It is separate from structural, semantic, behavioral and accessibility correctness.
Ownership boundary
- The component contract owns design-system meaning and general rules.
- The implementation source owns actual API and runtime guarantees.
- The agent-access layer discovers and delivers a relevant projection of governed knowledge.
- The product supplies current task context and product policy.
- The platform/runtime owns enforcement such as authorization, audit and rollback.
- Validation and tests provide evidence only for the rules and cases they cover.
A component contract may link or represent an external condition without becoming authoritative for its truth.
Evidence language
- Demonstrated: verified public implementation or direct empirical result in the stated domain.
- Documented: first-party guidance or organizational claim.
- Empirical adjacent evidence: a relevant experiment that does not directly test the design-system task.
- Inferred: a conclusion drawn from combined evidence, not directly measured.
- Not established: current evidence is insufficient.
Standards demonstrate representability and conformance models. They do not, without an agent experiment, demonstrate improved LLM performance.