Problem-specific category page

Agent evaluation for drift, fake completion, and correction hold.

Use this page when the search intent is agent evaluation, drift verification, fake completion diagnosis, or LLM judge alternative.

VeriClaw 爪印 evaluates whether the model actually did the work, whether the correction held, and whether the next step is safe to trust.