Multi-Part Prompt with multi-step instructions

I have structured long prompts to llama 3.1 models with headings # Your Role Details on role # Content Background Content # Content (that is to be transformed) Content # Example of transformation # Rules on transformation #Instructions Following YOUR ROLE, transform the CONTENT while following the RULES OF TRANSFORMATION so that the output is similar to the EXAMPLE OF TRANSFORMATION. Your output should consider the CONTENT BACKGROUND. This seems to work for the most part. Sometimes the output contains the example and I have to restart it. Do any of you use a similar process or a better one? Have you found ways to do this type of thing better? In my experience, the output won’t perfectly conform and so I’ll ask it to evaluate how it could do better in the context of each of these sections. It gives good advice and I then tell it to implant the changes… then it doesn’t. Any advice here? #