Single-tool LLM analysis produces reports that look authoritative but aren't. A serial consensus pipeline catches artifacts and hallucinations at source.