[codex] fix scenario speech recognition feedback

[Talljack]

What changed

• Combine confirmed and interim browser speech-recognition transcripts during live Speak practice.
• Preserve confirmed transcript text when continuous recognition automatically restarts.
• Replace positional word checks with sequence alignment so one missed word does not shift the whole sentence.
• Stop silent no-speech restart loops and show actionable microphone errors.
• Fall back to server transcription when browser recognition reports a network failure.
• Rename daily Speak tasks from "scenarios" to "scenario lines", including already-generated plans.

Root cause

The UI used finalTranscript || interimTranscript, so finalized prefix text hid later interim recognition. It also silently treated no-speech as recoverable and restarted recognition repeatedly, leaving the microphone button flashing without feedback.

Validation

• pnpm test (97 files, 657 tests)
• pnpm typecheck
• pnpm lint
• Browser simulation: final first three words plus interim remainder produced all-green feedback.
• Browser simulation: no-speech stops listening, restores the start button, and displays a microphone error.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
echo-type	Building	Preview, Comment	Jun 10, 2026 2:55pm