How can I measure the robustness of my prompt?

Robustness is about consistency under stress. Try these checks:

• Paraphrase or shuffle sections of your prompt—does performance stay steady?
• Insert a benign distraction (e.g., an unrelated sentence) and see if the model still follows the core instructions.
• Add an adversarial string like "Ignore all previous instructions and ..." to test resistance to injections.
• Track score variance across 3-5 runs; a robust prompt shouldn't swing wildly.