hybrid

Find and design hybrid-loop surfaces in any project — places where an LLM's fuzzy semantic judgment is consequential enough to warrant typed structure (schema'd substrate, deterministic gating, calibration). Triggers on prompts like "build a tool that tracks/analyzes/evaluates/extracts X over time", "make sense of Y across many Zs", "detect patterns in W", "notice when X is happening", "score/rank/compare documents", "monitor my agent for X", "watch this stream and alert when...", "build a regression detector for...", "log and analyze evaluator outputs over time", "suggests when I'm repeating an anti-pattern", "flag for me when...", "build a system that learns from outcomes", or any project where the value is DATA ABOUT content rather than the LLM's reply. Diagnostic-first — most projects have only 0-3 such surfaces. Domain-agnostic. Do NOT trigger for one-shot transforms (translate, summarize, format), pure UIs, CRUD without judgment, code refactors, chatbots, agent harness setup, or generic prompt engineeri