What this is: LLMs often give contradictory answers to the same question when the wording changes slightly. Verity detects this automatically by generating semantic variations of your prompt, running them through the model, and flagging conclusion flips, reasoning contradictions, and adversarial failures. Like unit tests but for AI reasoning.