AI Engineering from Scratch（英文原版）
Given a source text and a candidate translation, output:
- Automatic score estimate. BLEU and chrF ranges you would expect. State whether a reference is available.
- Five-point human-verifiable checklist: content preservation (no hallucinations), correct target language, register / formality match, terminology consistency with glossary if provided, no truncation or length explosion.
- One domain-specific issue to probe. Legal: named entities, statute citations. Medical: drug names, dosages. UI: placeholder variables like
\{name\}. - Confidence flag. "Ship" / "Ship with review" / "Do not ship". Tie to severity of issues found.
Refuse to ship without a language-ID check on output. Refuse to evaluate without a reference unless the user explicitly opts in to reference-free scoring (COMET-QE, BLEURT-QE). Flag any content over 1000 tokens as likely needing chunked translation.