Quality tab overview

The Quality tab gives you automated quality scoring and runtime safeguards for every specialized agent execution in Optimizely Opal. Define your quality criteria, score each execution against them, and recover or block executions that deviate from expected behavior.

The Quality tab has the following two sub-tabs:

• Output Evaluation – Scores every run against criteria, examples, and a baseline evaluation score you define. See Configure Output Evaluation.
• Execution Guardrails – Detects executions that deviate from your agent's normal behavior. See Configure Execution Guardrails.

A third feature, Execution Advisor, intervenes during execution to handle anomalies, retry failed tool calls, and review sensitive tool calls. Execution Advisor does not have its own configuration screen. It executions automatically based on what Output Evaluation and Execution Guardrails detect. See Execution Advisor.

The execution details display two independent verdicts for each run. Output Evaluation calculates the Evaluation Score. Execution Guardrails calculate the Guardrail Status. A run sometimes passes one and fails the other.

When to use each feature

Each Quality tab feature targets a different aspect of agent quality. Use the following list to choose the right feature for your goal.

• Output Evaluation – Score every run against quality criteria you define.
• Execution Guardrails – Stop executions that behave unlike a normal run for the agent.
• Execution Advisor – Recover flagged executions and gate sensitive tool calls automatically.

Output Evaluation and Execution Guardrails work independently. Many agents start with Output Evaluation only, then add Execution Guardrails when they have established a stable behavior baseline.

Access the Quality tab

Open the Quality tab to configure quality scoring and runtime safeguards for a specialized agent.