Primary-source library
Read past the summary.
144 distinct source pages underpin this edition. Links open the original standard, regulator, research, or vendor documentation—not a search result.
NIST — AI Risk Management Framework ↗
Used in What a chatbot actually is, The support agent operating model, Scope and refusal contracts: make a useful no repeatable, Conversation quality review: use an anchored rubric, not a vibes score, Model-change evaluation: compare a changed system without a live leaderboard.
W3C — Natural Language Interface Accessibility User Requirements ↗
Used in The channel is part of the system, Accessible conversation is more than screen-reader support, Multilingual service is not a translation switch, Accessibility of embodied voice interfaces: never make the character the only path.
NIST — Generative AI Profile ↗
Used in Ground answers in knowledge users can inspect, Evaluate conversations before launch, Latency, resilience, and cost are one design problem, The support agent operating model, Embodied safety: disclosure, likeness, and dependency, Prompt injection at retrieval and tool boundaries.
WAI-ARIA 1.2 ↗
Used in Avatar behavior state machines: make animation answer to service state, Conversational onboarding: preview the work before asking for trust, Editable confirmation summaries: corrections must reopen the right decisions, Accessible streaming responses: make partial output interruptible and readable, Kiosk session reset and privacy: make the next visitor truly new, Character content boundaries: keep canon separate from session memory.
WCAG 2.2 Focus Order ↗
Used in Conversational onboarding: preview the work before asking for trust, Accessible streaming responses: make partial output interruptible and readable, Multilingual evaluation fixtures: test the interface, not only translated sentences, Kiosk session reset and privacy: make the next visitor truly new, Embodied training scenarios: design roleplay that can be reviewed and transferred carefully, Character content boundaries: keep canon separate from session memory, Avatar fallback quality ladder: degrade presence without losing the task.
W3C Trace Context ↗
Used in Latency trace budget: measure the first usable response, Streaming tool-call assembly: execute only a finished call, Webhook delivery reconciliation: separate receipt from business outcome, Conversation cache invalidation: freshness is scoped, Multi-user avatar turn ownership: the room needs an authority.