Cross-Reference and Terminology QA

Layer boundaries

• STConversationRunner — high-level runtime that executes declarative Conversations.
• STMediaRunner — lower-level playback, recording, synthesis, recognition, audio environment, and typed media events.
• Conversation — declarative executable graph.
• Prompt Module — declarative reusable prompt/media content.
• Configuration Reference — schema/defaults/precedence for Conversation, Prompt Modules, and resource formats.

SDK pages should not duplicate Configuration Reference schemas or STMediaRunner media semantics.

Terminology decisions

• Use Conversation consistently in new primary documentation; legacy product terminology belongs only in migration material.
• Use Prompt Module for the declarative prompt format even though the current Swift loader remains loadAudioPromptModules.
• Use Audio Resource-to-Text Map for the resource format even though the current Swift loader remains loadAudioToTextMap.
• Preserve setLanguageFolder(_:) as the shipping Swift symbol while describing the corresponding declarative concept as resource language where appropriate.
• Preserve current public API names unless a source/API revision explicitly changes them.

Source verification

The public declarations for STResourceCategory and STSRData were supplied and verified in v0.5.

• STResourceCategory is public enum STResourceCategory: String, Sendable with seven cases.
• STSRData is a public final class; SpeechTiles owns construction/population, while clientRejectedHypothesis is publicly writable for application-level hypothesis rejection.
• STSRData.numberOfSegments returns the recognition segment count.
• STSRData.confidence is the arithmetic mean of segment confidence values and returns 0.0 when no segments are present.