WFGY

Why we included this project

When an AI workflow keeps failing in ways that are hard to explain, such as hallucinated citations or a RAG pipeline quietly returning plausible but wrong answers, this project is built around putting those failures on a map you can actually work from. The Problem Map and Global Debug Card are structured troubleshooting guides that walk you from symptom to likely cause for reasoning and retrieval pipelines, including agent runs that stall midway. The Polaris Goal Compiler is the more forward-looking piece: a portable protocol you can paste into an assistant, an agent rule file, or a workflow policy, so goals get compiled and verified before the AI claims the job is done. Teams that value evidence will find experiment packages with verdicts, audits, and token records, alongside the evaluation discipline laid out in the WFGY 4.0 material. This is a practical debugging and verification toolkit for people building serious LLM-based systems, not another demo or prompt collection.