hermeneutic is an evidence-first drift gate for AI agents. It mines corrections from your AI chat logs (prior response, user correction, repair), classifies the drift, and runs a cheap-to-expensive pre-flight gate on the next response before drift ships. Regex, then structured scoring, then a pressure probe. MIT, zero dependencies, by Hermes Labs.