Multi-Model AI Translation
Use fake mode by default. Use provable mode only with explicit user approval for provider cost and a configured environment.
- Confirm source, targets, scope, placeholders, and whether reverse verification is required.
- Submit a batch once through existing scripts; never manually simulate provider outputs or send secrets through chat.
- Preserve structured data and placeholders. Choose exact consensus or majority output; flag no-consensus and semantic-drift entries.
- Verify only the translated batch, then report files, counts, cost when applicable, and exception keys.
Do not emit full model outputs, proof payloads, or per-key success tables unless the user requests an audit artifact.
