Iterative Refinement

A prompt is rarely right on the first try. The job is not to write the perfect prompt — it is to converge on a prompt good enough for the use case, predictably, by changing one thing at a time and measuring.

When to use

• You have a prompt that works ~70% of the time and need to push it higher.
• A working prompt regressed (model upgrade, schema change, new edge case).
• You're tempted to rewrite the prompt from scratch — stop and iterate instead.
• You're handing the prompt off and need a record of why it's structured the way it is.

The Loop

Write → Test → Analyze → Improve → (back to Test)