A team has drafted two different system message variants for a Foundry-hosted support assistant and wants to decide which one to ship, backed by evidence rather than opinion. What should they do before choosing?
Select an answer to reveal the explanation.
Short Explanation
When you've got two system message variants, don't guess—run both through Foundry's evaluation tools against the same test question set and compare metrics like groundedness and relevance. That gives you evidence instead of opinion.
Full Explanation
Running both system message variants through Foundry's evaluation tools against the same set of test questions produces comparable metrics such as groundedness and relevance, giving the team objective evidence about which variant actually performs better before either one reaches real users. Shipping both variants simultaneously to all users and seeing which gets more complaints is incorrect because it exposes every user to whichever variant they happen to get with no controlled comparison, and complaint volume is a noisy, delayed signal rather than a direct quality measurement. Picking the variant written by the most senior engineer is incorrect because choosing based on the author's seniority says nothing about how well the prompt actually performs; a more experienced engineer can still write a weaker system message than a less experienced one. Choosing the shorter system message, since shorter prompts always perform better, is incorrect because prompt length by itself does not determine quality; a shorter system message can omit necessary grounding instructions just as easily as a longer one can include useful detail, so length is not a reliable stand-in for measured performance.