What the documents look like
Marketplace category qualification access happens before store onboarding or product listing. The core materials are the business license, trademark registration certificate, brand authorization chain, and industry license. These materials form an evidence chain for the operating entity and the scope of authorization. When any link is missing or inconsistent, the result can be onboarding rejection or the inability to publish within a category. For operations permission and handover management, the key question is what authorization state a given material or operation request occupies at the moment of handover.
Material sources include platform merchant recruitment pages, the national enterprise credit information publicity system, trademark office data, and license systems. These sources update irregularly, and each onboarding round requires fresh verification. Citation tracing addresses a specific problem: when operations staff hand over, they need to trace which material, which version, and which passage supported a qualification judgment.
Constraints at the citation tracing step
The knowledge base search node outputs cited content as an array, and the array length may be zero. Even when no content is found, the output path still completes. This means an empty citation array is a valid output. Downstream handover steps must handle empty results, and an empty array does not equal a system error.
The search node can associate one or more knowledge bases that share the same vector model. Knowledge bases with different vector models cannot be associated together. During verification, check the vector model identifier first, exclude non-matching bases, and record the actual observation. The minimum relevance score is a filtering threshold that takes effect only in semantic retrieval or when reranking is used. That value does not perform permission filtering or entity isolation. Whether a retrieved result is adopted still depends on the valid version of the material and human review.
Hypothetical paper test case: A brand channel manager submits a scanned copy of a Brand Authorization Letter to a store onboarding specialist. The original text states: "The authorizing party is A Brand Management Co., Ltd.; the authorized party is B E-Commerce Co., Ltd.; the authorized category is household appliances; the authorization period runs from January 1, 2025 to December 31, 2025." In the same batch, a business license states the entity is B E-Commerce Co., Ltd., with a business scope of daily necessities sales, and household appliances are not listed. The draft sentence to be assessed is: "Company B has obtained brand authorization for the household appliances category and can directly submit the onboarding application."
Configuration and implementation checks
| Configuration item | Hypothetical starting value for this test | Explanation and comparison method |
|---|---|---|
| Associated knowledge base | Select one qualification materials base | Check the vector model identifier; only bases with the same identifier can be associated together |
| Search mode | Hybrid retrieval | Uses semantic recall and full-text recall together, merged through RRF |
| Citation limit | Set to 2000 tokens | Hypothetical starting point because qualification passages are long; compare citation completeness against 1000 tokens |
| Minimum relevance | Set to 0.5 | Hypothetical starting point, effective only in semantic retrieval or reranking; compare recall range against 0.3 |
| Question optimization | Enabled | Handover questions often contain references such as "this authorization," which need the subject filled in |
| Result reranking | Enabled | Secondary ordering when candidates are numerous; the score is used for filtering |
The citation limit is controlled by tokens, not by top k. The hypothetical starting value of 2000 tokens is chosen because one authorization letter plus one business license totals roughly 1500 to 1800 characters, leaving margin. This value is not an official default. Adjust it after comparing with the actual length of cited content.
Three common errors
- Treating a retrieval hit as authorization confirmation. The input is the authorization letter text and the business license text from the hypothetical sample. Retrieval may recall both materials. Compare the entity and category fields in both materials: the authorization letter authorizes household appliances, while the business license scope does not list household appliances. The observed result is that the two materials disagree on the category field, and the draft sentence "has obtained brand authorization for the household appliances category" is not supported. A retrieval hit does not constitute valid authorization.
- Treating a pending-review state as approved. The input is the handover request "submit onboarding application." If the request is marked as pending review in the handover record, compare the request status field against the approved value. The observed result is that the status is pending review, and the submission action has not been executed. Pending review does not demonstrate a lack of operating permission, and it does not equal approval.
- Treating an empty citation array as retrieval failure. The input is the search node output. Observe whether the output is an array of length zero. If the downstream design does not handle an empty array, the handover record may lose the fact that no material was retrieved. An empty array is a valid output and must be distinguished from a system error.
How to verify the setup
- Input the authorization letter text and business license text from the hypothetical sample. Compare the entity fields and category fields in both materials. Observed result: the entity is consistent, and the category field is inconsistent. The condition is met when the draft sentence is marked as unsupported and the handover record retains both material texts and the field difference.
- Input the handover request "submit onboarding application" from the hypothetical sample. Compare the request status field against the approved value. Observed result: the status is pending review. The condition is met when the submission action has not been executed, the record is marked pending review, and it is not marked approved or rejected.
- Input the search node output array. Compare the array length against the downstream handling logic. Observed result: when the length is zero, check whether the downstream path completes. The condition is met when the downstream path executes normally, the handover record notes that no material was retrieved, and no system error assertion is produced.