Latest live status is maintained in docs/STATUS_2026-02-11.md.
Short summary:
- End-to-end GPU inference now produces intelligible speech.
- Voice cloning and all emotion control pathways run and generate WAV outputs.
- Major rumble source in BigVGAN path has been fixed.
- Remaining gap is GPT cached decode parity drift after step0.
See docs/HANDOFF_2026-02-11.md for full continuation context.