Active Testing

Active testing lets you send test inputs to your AI system and check the responses. Use it while you set up an Agent, after you change your system, or when you want to run a regression check. For continuous production monitoring, see Passive Testing.

Before you run a test:

Verify your email. You can configure and save tests before verification, but Mibo blocks active test execution until you verify.
Finish the Agent setup and save its connection details.
Create at least one active test case. An active test case is enabled for runs; Disabled test cases are excluded.
Use your own Agent. The Demo Agent is read-only and cannot run tests.

Running a single test

Open your Agent’s test list and click the play button for a test case. You can also open the test case and click Run Test. This is useful for:

Verifying your connection is working.
Debugging one scenario.
Checking a change before you run the full suite.

Running selected or all tests

To run several tests, select the active test cases and click Run Selected. To run the full active suite, click Run All Tests. Mibo sends each test input to your AI system and collects the responses.

If Run All Tests is disabled, hover over the button to see what is missing. Common reasons are a Demo Agent, incomplete Agent setup, or no active test cases.

Watching live logs

As a test executes, open its test case to follow the Execution Logs. They show:

Which step is running.
The input sent to your system.
The response received.
Issues reported during the run.

Expand Last Result to inspect the latest result. It can show the input, Agent Response, an Error, conversation turns, Assertions, and Raw output. An assertion is one check of the behavior you expect from your system.

Select an assertion to see details such as its criteria, target, expected value, actual value, or semantic reasoning. For a failed result, follow Recover from a run error to check errors and logs before diagnosing an assertion failure.

Read the result status

The test list and Last Result use these statuses:

Status	What it means	What to do next
Not Run	This test has no result yet.	Click Run Test to start it.
Pending	Mibo has accepted the run and is waiting for it to finish.	Open the test to watch its logs.
Passed	The response met the configured assertions.	Review the result, then run more scenarios if needed.
Failed	The run failed because of an execution problem or an assertion that did not pass. A response may be missing.	Check Error and the logs first. If assertions ran, compare any failed assertion with the response.
Error	Mibo could not complete the test.	Open Last Result, read Error, and check the logs.
Skipped	Mibo did not evaluate the test or check.	Check whether the test is disabled or whether an assertion condition excluded it.

While a test is running, the detail page shows Running and updates the logs. A run can also appear as All Passed or Issues Found in Mibo’s Latest Runs panel.

Recover from a run error

If the page shows Email verification required, click Resend verification email, open the link Mibo sends, return to Mibo, and click I’ve verified my email. The session refreshes; retry once Run Test is enabled.
Open the test case and expand Last Result. Check for an Error message and read the latest Execution Logs, even when the status is Failed. A connection failure can leave no response or assertion results.
If the error points to the connection, open Set up agent and check the fields shown for your provider. Follow Test your connection to verify and save the configuration.
If assertions ran and one failed, compare it with your system’s response. Fix the system or adjust the expected behavior in the test.
If you edit the test, click Save before running it again. Run Anyway uses the saved test and does not include unsaved edits.
Once the previous execution has finished and Run Test is enabled, click it again. Start with one test so you can check the result before running the full suite.

If a test stays Pending or Running without new log entries, check that your Agent is reachable. Wait for the current execution to complete or time out. Restoring the connection does not end the pending execution. Refresh the page if necessary to update its status, then retry once Run Test is enabled and Mibo allows a new execution.

Recover from a semantic evaluation issue

A semantic assertion can receive your Agent Response and still fail to produce a usable evaluator score. This is different from your Agent returning a response that scores below its threshold.

If the evaluator returns a malformed score, the semantic assertion is not a valid quality result. Last Result can show Failed, with the problem in Error, Reason, or Reasoning.
If evaluation cannot complete because the evaluator is temporarily unavailable, the assertion details say that the check was skipped. Skipped checks are excluded from the overall score, so the test can show Passed even though that semantic check was not evaluated. Treat it as unevaluated.

To recover:

Open Last Result and select the semantic assertion. Read Error, the assertion Reason, Reasoning, and the latest Execution Logs before changing your Agent or lowering the threshold.
If the error is about the threshold, edit the test. In the Form tab, set Strictness to a finite number from 0 through 1. In the JSON tab, set threshold to a finite number in the same range. Then click Save.
If the run is still Pending or Running, wait until it finishes. Do not start another run while the current execution is active.
When Run Test is enabled, click it again. Start with one test so you can confirm that the evaluator returns a usable result before running the full suite.

What happens behind the scenes

When you click Run Test, Mibo:

Sends the test input to your system, like a real user would.
Collects the response and any trace data your system returns.
Evaluates the response against the test’s assertions.
Saves the result and updates your quality metrics.

Your Quality DashboardLearn how to read your quality scores and failure breakdowns.