1 match across blog posts, projects, case studies and series.
Keep your prompts in a versioned JSON library, run them against fixed test cases with a small Python harness, and grade the output with rule checks or an LLM j…