v0.2.0-beta.5.1.7 - Gemini TTS Audio Materialization

Gemini TTS completed-Interaction audio materialization

Fixes the Batch A failure where an accepted Gemini Interaction returned status=completed before audio was visible in the initial snapshot.

Runtime behavior

• completed/no-audio preserves the accepted Interaction ID;
• up to 6 bounded same-ID GET materialization polls;
• GET polling remains separate from the generation POST hard cap;
• generation POST hard cap remains 10;
• no duplicate generation POST after an Interaction ID is accepted;
• response_format remains type=audio with no delivery property;
• fail closed with TTS_INTERACTION_COMPLETED_WITHOUT_AUDIO after the bound.

Deterministic runtime proof

• success: completed/no-audio -> same-ID GET #1 no audio -> GET #2 audio -> PASS;
• failure: persistent completed/no-audio -> exactly 6 same-ID GETs -> fail closed;
• same ID, poll numbering, API-key forwarding and poll interval are asserted.

GitHub lifecycle

• Issue #66
• PR #67
• PR final head: 1a3ba20
• merge/release SHA: 30c75a0
• PR CI run: 31585178328
• main CI run: 31585940997
• Pages deploy run: 31585940957

Trigger evidence

• failed controlled Batch A run: 31572749438
• failed source tag: v0.2.0-beta.5.1.5
• failed source SHA: 686f5ff

Safety boundary

This release does not dispatch Batch A, does not modify production episode metadata/player/Piper audio, and does not publish media-dual-v0.2.0-beta.6. Batch A remains separately controlled under #53.