Vocence locks miner code, tightens competition rules

Vocence is implementing stricter anti-cheat measures by locking miner.py and restricting miners to fine-tune only the Qwen3 voice model weights, blocking code exploits that circumvent instruction-based voice synthesis. A staging branch PR will merge at 5 PM EDT with enforcement starting immediately; miners must review and update accordingly. The change targets discovered workarounds where miners inject hardcoded speaker embeddings instead of generating voices from instructions.

• •miner.py now locked; only Qwen3 voice model .safetensors fine-tuning permitted
• •Staging PR merging ~5 PM EDT; cutover block set to 8270310
• •Exploit: top miners using speaker lookup tables to bypass instruction synthesis
• •Governance challenge: manual blacklists insufficient; automated validation needed