A team wants to improve how it evaluates Copilot coding agent work across issue assignments. The team has seen pull requests where tests passed, but reviewers later found requirement misunderstandings. Which four actions should you perform in order? 1. Require the agent to write a self-evaluation summary. 2. Define acceptance criteria and measurable task outcomes. 3. Map reviewer judgments to qualitative evaluation labels. 4. Configure workflows to collect quantitative validation signals. 5. Treat the Copilot PR summary as final success evidence. 6. Tune prompts after reviewing failed evaluation cases.

The correct sequence prioritizes establishing clear expectations and robust evaluation mechanisms. First, Define acceptance criteria and measurable task outcomes (2) ensures the agent and reviewers understand what constitutes success, preventing misunderstandings. Next, Map reviewer judgments to qualitative evaluation labels (3) standardizes feedback, making it actionable and consistent. Then, Configure workflows to collect quantitative validation signals (4) gathers objective data (e.g., test coverage, performance metrics) to complement qualitative feedback. Finally, Tune prompts after reviewing failed evaluation cases (6) uses the collected data and feedback to iteratively improve the agent's performance. Option 1 → 4 → 2 → 6 is incorrect because requiring a self-evaluation summary (1) before defining clear outcomes (2) or collecting data (4) makes the summary less reliable. Option 4 → 2 → 5 → 6 is incorrect because treating the Copilot PR summary as final success evidence (5) is a critical mistake, as the problem statement highlights that tests passing doesn't guarantee requirement understanding. Option 3 → 1 → 5 → 4 is incorrect for similar reasons, prematurely trusting agent summaries (5) and misordering foundational steps.