Testing Center results interpretation
Key Takeaway
Subagent Assertion and Action Assertion are strict pass/fail checks, while quality scores are guidance you weigh against your goals. When an agent test fails, retest the utterance in Agentforce Builder first, because the test expectation may be wrong rather than the agent.
Testing Center runs a suite of test cases against an Agentforce agent or a prompt template and scores each result. Reading those scores tells you whether a problem lies in subagent choice, action choice, or response quality. The scores also point you to where to fix the problem. Some scores are strict pass/fail checks, while others come from a large language model that judges the response.
What a test run shows
When a test run finishes, results appear in a grid with one row per test case and one column per scorer. A run summary also shows separate pass percentages for subagent, action, and response evaluations, so you can see at a glance which part of the agent is weakest.
Each row compares what the test case expected with what the agent actually did. The expected subagent and expected actions sit next to the actual subagent the agent chose and the actual actions it ran from that subagent.
The grid shows a top-level score in each cell. Select a cell to see why the result got its score. A detail panel opens with the full JSON output, which includes the reasoning behind the score.
You can turn any JSON field into its own grid column. In the JSON tab, hover over the field and click "+ Column". To include score details or judge reasoning in a CSV export, add those fields as columns first, then click "Download Results".