> For the complete documentation index, see [llms.txt](https://docs.crobox.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.crobox.com/how-to-guides/sagent/qa/results.md).
# Results
Run your tests and AI Generated Tests, then review how Sagent answered.
**Results** shows how Sagent answered your tests, run by run. Use it to catch problems before shoppers do, understand exactly why a test failed, and confirm your fixes worked.
### Run your tests
{% stepper %}
{% step %}
#### Start the run
Click **Run tests** on either QA sub-tab. The dialog shows how many Crobox system checks are included.
{% endstep %}
{% step %}
#### Confirm
Click **Run tests** in the dialog. Each enabled test is sent to the version you're editing, and Sagent's answers are checked.
Runs usually take a few minutes. You can leave the page while the run is in progress.
{% endstep %}
{% step %}
#### Review the results
Open **Results** to follow progress. When the run finishes, open it to see each answer and check.
{% endstep %}
{% endstepper %}
### Run AI Generated Tests
AI Generated Tests create around 180 test messages that try to push Sagent off topic, into unsafe answers, or into revealing its instructions, then check how it responds. They complement your own tests by probing situations you might not think to write yourself.
Click **Run AI Generated Tests**, review what's checked, and click **Run tests**. The run takes about 15 to 20 minutes.
The run checks whether Sagent can be pushed into:
* revealing its instructions;
* making things up;
* going off topic;
* being steered to other tasks;
* pretending to be someone else;
* overpromising what it can do;
* following hidden instructions in product or knowledge content.
It also includes longer conversations that escalate step by step.
{% hint style="info" %}
Run AI Generated Tests after larger changes, such as new knowledge sources, new behavior settings, or before a first launch. Use your own tests for every release.
{% endhint %}
### The Results overview
The summary above the table shows the latest pass rate, for example *96% passed, 288 of 300*. It also shows whether that result is **Up to date** or **Outdated** compared with the version you're editing, and how many tests are switched on.
Each row is one run:
| Column | Shows |
| ------------ | ------------------------------------------------------------------------------ |
| **Started** | When the run began. |
| **Type** | **Tests** or **AI Generated Test**. |
| **Version** | The version tested, for example **v10 (published)** or **v11 (editing)**. |
| **Status** | **All passed**, **2 failed**, **Running…**, **Stopped**, or **Didn't finish**. |
| **Duration** | How long the run took. |
| **Result** | How many tests passed, for example **20 of 50 passed**, with a pass-rate bar. |
Status colors make problems easy to spot: green when everything passed, amber when some tests failed, and red when the run didn't finish.
The summary is marked **Outdated** when anything has changed since the latest completed run, including your setup and any test you added, edited, renamed, switched on or off, or deleted.
Click the ⓘ icon on a row to open its run details. Use **Stop run** to end a run in progress, or **Delete run** to remove a run you no longer need.
### Run details
Run details show the full record of a run: when it started and finished, the version tested, and how many tests passed. Failed tests are listed first. Use the **Status** filter to focus on what you need.
Each test shows:
* **Your test** or **Crobox system check**, so you know where the test came from;
* the **Shopper message**;
* **Sagent's answer**;
* each **check**, whether it passed, and the reason;
* a **Score**, the average of its check scores;
* **Technical details**, for troubleshooting with Crobox support.
Conversation tests show one result per message, so you can see exactly where in the conversation an answer went wrong.
A result passes only when every check passes. Text, pattern, and behavior checks score either 0% or 100%. Checks graded by AI, such as **Meets criteria**, can score in between. That means a result can fail with a high score: one failed check fails the result, however well the others did.
Results always show each test exactly as it ran, even if you've since edited or deleted the test.
For AI Generated Tests, each test message is numbered and labeled with the **Area** it targets and the **Style** of the attempt.
### Fix a failing test
Each check's reason explains what went wrong. Start there, then decide whether Sagent or the test needs to change.
**Example:** the test *Rec: budget constraint* sends *Multi-day pack under 200 euros*. It failed because Sagent asked what kind of pack the shopper wanted instead of recommending products:
* **Shows products** failed because Sagent asked a question instead of searching for products.
* **Meets criteria** failed because the answer didn't recommend any packs under 200 euros, or say that none fit the budget.
From here, you have two options:
* **If Sagent should have recommended products,** adjust how readily it recommends in [**Behavior**](/how-to-guides/sagent/setup/behavior.md) or your question flow in [**Guidance**](/how-to-guides/sagent/data/guidance.md), then run your tests again.
* **If a follow-up question is acceptable here,** update the test so it reflects what you want, for example by replacing **Shows products** with **Asks a follow-up question**.
{% hint style="warning" %}
Don't change a test just to make it pass. Update a test only when your expectations for Sagent have genuinely changed.
{% endhint %}
### Download results
Click **Download results (CSV)** in run details to save the full record of a run. The file includes each test's name, shopper message, Sagent's answer, every check with its result, the score, and the reason. It's a convenient way to share evidence for legal, security, or business review.
### Troubleshooting
| Message | What it means | What to do |
| -------------------------------------------------- | -------------------------------------------------------------------------------------------------------- | --------------------------------------------------------- |
| **Sagent didn't respond to this message.** | Sagent returned an error instead of an answer. | Check your Sagent preview, then run your tests again. |
| **This run took too long and stopped.** | The run reached its 60-minute limit. Results for finished tests are kept. | Switch off tests you don't need, or run your tests again. |
| **This run was interrupted. Please run it again.** | The run stopped unexpectedly, for example during a service restart. Results for finished tests are kept. | Run your tests again. |
| **No tests are switched on.** | There's nothing to run. | Add a test or switch an existing one on. |
| **Tests are already running.** | A run is already in progress for this Sagent. | Wait for it to finish, or stop it from **Results**. |
| **Testing isn't available right now.** | A service QA relies on is unavailable. | Contact Crobox support. |
### Frequently asked questions
Why does a run show more results than my enabled tests?
Every run includes the Crobox system checks alongside your own tests. Conversation tests also show one result per message.
Why is my latest result marked Outdated?
Something has changed since your latest completed run, such as a setup setting or a test. Run your tests again to check the version you're editing.
Can a test fail with a high score?
Yes. The score is the average of the check scores, but a result passes only when every check passes. For example, if four of five checks pass, the result fails with a score of 80%.
Do I lose results if a run stops early?
No. Results for tests that finished are kept, and the run is marked **Stopped** or **Didn't finish**. Tests still in progress when the run ended aren't saved, and runs don't resume automatically, so run your tests again to complete the set.
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.
Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:
```
GET https://docs.crobox.com/how-to-guides/sagent/qa/results.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.