Save the Foundry agent version before running its evaluation
Short answer
D Save Agent1 as a new version so the updated instructions become a versioned target for the synthetic-dataset evaluation while remaining unpublished to the dev team
Question
You open Agent1 in the playground and update the instructions.
You need to run a full evaluation against the updated instructions. The solution must meet the following requirements:
Test the changes by using a synthetic dataset.
Ensure that the changes are available only for the development team that has access to Agent1.
What should you do first?
Key Takeaways
The question tests the save-then-evaluate-then-publish sequence; the trap is that publishing sounds like the natural way to 'release' updated instructions but violates the requirement that only the dev team sees them.
To evaluate updated instructions in Microsoft Foundry, you must first save Agent1 as a new version, creating a versioned target for the synthetic-dataset evaluation while keeping the change unpublished and restricted to the development team (D).
Common mistake: Option A is the most plausible wrong answer: publishing Agent1 as a new version does expose the updated instructions, but through its own endpoint and identity to consumers beyond the development team, breaking the second requirement. Option B fails because evaluating before saving runs against the old instructions.
Expert Analysis
Why the version must exist first Evaluation runs in Microsoft Foundry are bound to a concrete, saved artifact of the agent. Playground instruction edits are unsaved working state, so there is nothing stable for a full evaluation — synthetic dataset or otherwise — to execute against. Saving Agent1 as a new version (option D) freezes the updated instructions into a versioned target you can point an evaluation at, which is the required first step in the sequence.
Why not publish (A) or preview (C) Publishing is the adjacent decision here, and it is what the second requirement rules out. A published version is exposed through its own endpoint and identity to consumers beyond the people who hold access to Agent1, so publishing would leak draft instructions to a wider audience than the development team — the stated comment captures exactly this reasoning. Preview (C) is a manual chat smoke test inside the playground: useful for a quick sanity check, but it is not a full evaluation and it does not create a versioned evaluation target, so it fails the synthetic-dataset requirement.
Why not create the evaluation first (B) Option B is the tempting trap because it sounds like the main task. But an evaluation created before the instructions are versioned would run against the previously saved state, producing results that silently test the old agent. The correct order is save the new version, then attach the evaluation to it — and only publish if the scenario later says the change must reach external consumers.
Consequence of getting it wrong Choosing A over-exposes draft instructions operationally; choosing B yields misleading evaluation scores. Keeping the new version unpublished confines the change to the development team that already has access to Agent1, which is exactly the governance requirement.
Exam Strategy
When a question asks what to do 'first' with a 'must also' governance constraint, separate the sequencing answer from the exposure answer: version-creation steps usually precede evaluation, and anything that creates a public endpoint is suspect. On Foundry agent questions, treat publish, save-as-version, preview and create-evaluation as an ordered pipeline (save → evaluate → publish) and pick the earliest stage that satisfies every stated requirement. If a requirement says changes must reach only a named internal audience, eliminate any option that implies an endpoint, identity, or consumer-facing rollout.
Frequently Asked Questions
Why must I save Agent1 as a new version before running an evaluation?
What is the difference between saving and publishing an agent version?
Can I use Preview in the playground instead of a full evaluation?
Community Votes
100% of anonymous learners picked answer D. Votes are pick records left by other test-takers — they are not the verified answer.
Discussion 1 contributions
Compiled from public community threads and reader submissions, lightly edited for clarity. Handles and vote tallies are shown as recorded, and comments reflect the views of individual contributors.
From public community threads 1
-
ExamProAzure 👍 1 Answered DWhy not publish Publishing makes the agent available to other consumers through its own endpoint and identity. That goes beyond the development team and isn’t needed to evaluate.
Reader comments & corrections
No reader corrections yet — spotted an error or have a note? Share it below.