The model learns that hedging is a signal of lower-quality output. This creates a systematic bias toward sounding certain.

Every checklist is written as invariants and detection smells, not framework APIs, so the same content audits a Rails app, a Spring service, or an Express API — the agent supplies the ...