> For the complete documentation index, see [llms.txt](https://docs.cherryai.com.cn/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.cherryai.com.cn/docs/en-us/knowledge-base/recall-test.md).

# Check Materials and Retrieval

Recall testing directly checks whether the problem can find the correct chunk, without first being polished by the chat model. It helps you determine whether the issue lies in the materials, parsing, chunking, or retrieval settings.

{% hint style="info" %}
Prepare 3–5 real questions whose answers you already know, and reuse them each time the materials, model, or chunks are updated. Fixed questions make regressions easier to spot than ad hoc test questions.
{% endhint %}

## Prepare test questions

It is recommended to cover three types of questions at the same time:

* Exact facts, such as “What is the accommodation limit in first-tier domestic cities?”
* Conditional rules, such as “Under what circumstances can overseas car rentals be reimbursed?”
* Easily confused, such as “Who gives additional approval for business trips over 5,000 yuan?”

Do not test using only the material title or the full original text; that will overestimate the real-world effect.

## Complete a recall test

{% stepper %}
{% step %}

### 1. Open Recall Test

Open the left navigation \[Knowledge Base] → select a knowledge base → enter \[Recall Test].
{% endstep %}

{% step %}

### 2. Enter a real question

Enter a question with a clear answer and run the test. The question should be close to everyday wording; do not deliberately copy the original text.
{% endstep %}

{% step %}

### 3. Check the source and chunk

Confirm that the source is correct, and that the chunk contains both the conditions and the conclusion needed to answer.

<figure><img src="/files/c6f3bfaf5deb56fb89e9a46bf9e60343a52ba0e2" alt="召回测试中的命中来源、相关度、片段内容和排序"><figcaption><p>Do not just look at whether there is a result; also check the source, chunk completeness, and order.</p></figcaption></figure>
{% endstep %}

{% step %}

### 4. Identify the problem by comparing the symptoms

If there is no correct chunk at all, first check the materials, parsing, and chunking; if the correct chunk can appear but the order is unstable, then consider embeddings, reranking, or Top K.
{% endstep %}

{% step %}

### 5. Adjust one item at a time and retest

Change only one setting in each round. If parsing, chunking, or indexing is involved, first run \[Reindex], then retest with the same set of questions.
{% endstep %}
{% endstepper %}

## How to read the results

| Phenomenon                                                    | Explanation                                                                                                 | Next step                                                               |
| ------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------- |
| The correct source appears first, and the chunk is complete   | Recall is basically acceptable                                                                              | Test several different ways of asking again                             |
| No correct result at all                                      | The materials are not ready, content is missing, the wording differs too much, or the threshold is too high | Check the materials, body text, Chunks, and retrieval settings in order |
| The source is correct, but the chunk is missing key sentences | The parsing or chunk boundaries are not ideal                                                               | Check the Chunks, adjust them, and reindex                              |
| Old and new versions appear at the same time                  | All materials with the same name are retained                                                               | Delete the old entry or re-import using \[Replace]                      |
| The correct item often appears later                          | Too many candidates or unstable ranking                                                                     | Clean up the materials, and consider an embedding or reranking model    |
| Recall is correct but the chat answer is inaccurate           | The problem is more likely in the prompt or chat model                                                      | Keep the recall settings and adjust the question wording and chat model |

{% hint style="warning" %}
The chat model cannot make up for key information that was not found during recall. When recall results are not good enough, do not rely on repeatedly switching chat models to troubleshoot first.
{% endhint %}

## Tuning loop

<figure><img src="/files/8a987762d4b568b0fa1dd608f98a10abd15b021c" alt="用固定问题检查召回、定位问题、单项调整、重新索引并复测的质量调优闭环"><figcaption><p>Fixed questions → check results → locate the layer → adjust one item → reindex if necessary → retest.</p></figcaption></figure>

Recommended order:

1. Confirm the materials are correct, with no duplicates or outdated versions.
2. Check the parsed body text and Chunks.
3. When the wording differs greatly from the original, consider the embedding model.
4. When the candidates are roughly correct but the order is unstable, then consider a reranking model.
5. Reindex after adjustments, and repeat the same set of tests.

<figure><img src="/files/0359b26559db3bea6e323bf599d74c081ad91c93" alt="知识库高级设置中的智能分段、分隔符、分段大小和重叠大小"><figcaption><p>If chunks are incomplete, check the chunking settings again; changes only affect new materials, and old materials need to be reindexed.</p></figcaption></figure>

## Configuration notes

| Configuration item       | Product default                              | Recommended starting point      | Effect                                         | Applicable scenarios                         | Notes                                                                |
| ------------------------ | -------------------------------------------- | ------------------------------- | ---------------------------------------------- | -------------------------------------------- | -------------------------------------------------------------------- |
| Number of test questions | —                                            | 3–5                             | Establish a repeatable quality baseline        | All knowledge bases                          | Covers exact facts, conditional rules, and easily confused questions |
| Top K                    | 6, optional 1–50                             | Keep 6 for now                  | Control the number of final chunks             | Trade-off between coverage and noise         | Increasing it may consume more context                               |
| Similarity threshold     | 0.0, shown only when reranking is configured | Start from 0.0                  | Filter out low-scoring results after reranking | Noise still remains after reranking          | Setting it too high will remove correct chunks                       |
| Retest method            | —                                            | Change only one item each round | Determine where the setting change comes from  | Tuning, updating materials, or model changes | Reindex first after changing chunking or the model                   |

## Expected results

* The correct source appears stably in the first few results.
* The chunk contains the conditions and conclusion needed to answer the question.
* After rephrasing naturally, the result remains stable.
* After updating the materials or settings, the fixed questions do not regress significantly.

## User case

Xiaolin found that “accommodation expense standard” could be hit using the original text, but “how much can be reimbursed at most for a hotel stay” was unstable. He first confirmed that the materials and Chunks were normal, then configured an embedding model and retested. After the correct chunk appeared, it occasionally ranked lower, so he then added a reranking model.

The completion criterion is: the same accommodation standard can be found in the first few results for all three different ways of asking, and the chunk contains the applicable cities and the maximum amount.

## Common questions

<details>

<summary>If there are no correct chunks at all, should I increase Top K first?</summary>

First check the material body text and Chunks. If parsing or splitting is wrong, increasing Top K will only return more incorrect or incomplete chunks.

</details>

<details>

<summary>Why can’t I see the similarity threshold?</summary>

The \[Similarity threshold] is shown in the knowledge base settings only after a reranking model is selected.

</details>

<details>

<summary>What should I do if recall is correct but the chat answer is still inaccurate?</summary>

Keep the current recall settings and check the question wording, conversation context, and chat model. At this point, the problem is usually no longer in the retrieval layer.

</details>

## Continue reading

<table data-view="cards"><thead><tr><th></th><th></th><th data-hidden data-card-target data-type="content-ref"></th></tr></thead><tbody><tr><td><strong>Model and retrieval settings</strong></td><td>Adjust embeddings, reranking, Top K, and chunking.</td><td><a href="/pages/244bae136e1bc54a35816d43dc1b6b0e3b8c7e3e">/pages/244bae136e1bc54a35816d43dc1b6b0e3b8c7e3e</a></td></tr><tr><td><strong>Document parsing and OCR</strong></td><td>Handle missing body text, garbled text, and scanned content.</td><td><a href="/pages/7d495aecf9165aca96855bc0d74b3e925728dedc">/pages/7d495aecf9165aca96855bc0d74b3e925728dedc</a></td></tr><tr><td><strong>Use in conversations</strong></td><td>Use the knowledge base for questions after recall is satisfactory.</td><td><a href="/pages/b797f738b8c03492018a83b57869ba2f43a00a62">/pages/b797f738b8c03492018a83b57869ba2f43a00a62</a></td></tr></tbody></table>


---

# 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 dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.cherryai.com.cn/docs/en-us/knowledge-base/recall-test.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

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.