Transcription baseline verification
Date: September 16, 2026
Base: 71fee5e03192dffe48056dedc0efacc94925bb62 (fresh origin/main worktree)
Scope: existing HudsonVoice tests; no proposed adapter implementation.
Command
swift test --scratch-path "$HOME/Library/Caches/codex-builds/hudson-transcription-baseline" --filter HudsonVoiceTestsResult
Exit status 0. Build completed; 27 tests in five suites passed:
- HudDictation capture start gate
- HudPendingUtteranceStore
- HudsonVoicePreferences
- HudSpeech
- HudSpeechPlayback
The checks cover duplicate starts and cancellation, durable queued audio and context, preferences, and existing speech credential/playback behavior. Dependencies were resolved from the current package graph during the run; this is a point-in-time baseline, not a claim about every future dependency revision.
Limits
No MAI, Gemini, ElevenLabs, or WhisperKit adapter under the proposed contract exists in this change. No remote transcription request, microphone capture, accuracy benchmark, speaker-label evaluation, or native settings interaction was performed. Existing voice tests do not establish those outcomes.
Next: implement the smallest contract and one reference adapter, then run the proposed contract fixtures before live audio acceptance. See the design.