Naming taxonomy

Type: Comparative view
Status: Active derived view
Canonical for: Classification of the semantic information carried and omitted by naming categories
Depends on: Variant matrix, property-model matrix, and reconciliation

The last column records a semantic hypothesis from the words and surrounding standards. It is not a measured LLM result.

Category	Encodes	Leaves unstated	Absolute or contextual?	Hypothesized interpretation; DS effect untested
Appearance	rendering: filled, outline, ghost	purpose, consequence	mostly absolute within a renderer	may expose appearance while leaving use unstated
Relative hierarchy	primary, secondary, tertiary	compared with what; business purpose	contextual	familiar label whose local meaning still requires scope
Importance/emphasis	high, medium, low	why important; consequence	contextual	may expose the dimension while remaining relational
Functional/interaction category (term provisional)	submit, navigate, dismiss, toggle	product goal and consequence	comparatively stable	platform vocabulary may provide grounding; transfer is untested
Product/user intent	commit, approve, purchase	exact interaction and presentation	domain-dependent	may be precise locally and unfamiliar globally without definition
Consequence/tone	danger, warning, critical, positive	stage, reversibility, prominence	partly contextual	may signal consequence without completing the decision
Brand	brand, accent, discovery, rovo	purpose and hierarchy	system/context-dependent	may signal presentation or identity rather than use
State	selected, disabled, loading, expanded	why state occurred	stateful but generally stable	platform alignment may aid interpretation; DS effect is untested
Context	inverse, toolbar, compact, on-dark	product intent	explicitly contextual	depends on the named context being available to the agent
Historical/default	default, base, standard, regular	nearly all decision meaning	system/version-dependent	supplies little explicit canonical meaning without local definition

Taxonomy conclusions

• Appearance words can be excellent names for appearance axes and poor selectors for product intent.
• Relative names are not inherently defective; they require a comparison set and placement rule.
• Functional or interaction categories may receive additional grounding from HTML and accessibility vocabularies, but their transfer to design-system agent performance is untested. “Submit” still does not distinguish purchase, application or deletion.
• Intent names can overfit a product domain and multiply APIs. They should be treated as defined concepts, not assumed natural-language universals.
• Consequence can coexist with any emphasis level. Systems such as GitLab and GOV.UK document precisely this distinction.
• default is operationally useful for compatibility but semantically thin. A stable machine identifier can coexist with a human/default label.