What the documents look like
Category access happens before listing or onboarding, and its core materials are entity qualifications and the authorization chain. In store service and internal training, marketplace development staff, onboarding specialists, and brand channel managers work with four document types: business licenses, trademark registration certificates, brand authorization chains, and industry licenses. These materials come from marketplace recruitment pages, the National Enterprise Credit Information Publicity System, trademark office data, and license systems. Updates are irregular, and collection restarts with each onboarding.
Document versions directly affect whether a citation holds. The same entity may hold certificates in different states across periods. A citation needs the material name, the subject, the status, and the valid scope. Missing any one of these is an evidence gap.
Assume a paper exercise sample. Material text one: "XX City XX Trading Co., Ltd., Unified Social Credit Code 00000000000000000X, business license registration status: active, operating period from January 1, 2024 to long-term, business scope includes daily goods sales." Material text two: "Trademark Registration Certificate No. 00000000, registrant XX City XX Trading Co., Ltd., approved goods Class 21, exclusive right period from March 1, 2024 to February 28, 2034." The draft sentence to be assessed: "This store has obtained brand authorization and can immediately list all Class 21 goods." The draft sentence involves only the business license and the trademark registration certificate. It does not state a brand authorization chain document. Authorization status is unsupported and cannot be judged as authorized.
Constraints at the citation tracing step
The knowledge base search node can select one or more knowledge bases that share the same vector model, for vector search. Output is an array of citations, and its length can be zero. Even when nothing is found, the output path still runs through. Citation content must be passed to the answer step, where the answer is checked line by line against the cited source text.
The minimum relevance score is a filter threshold. It takes effect only in semantic search or when result reranking is used. It does not perform permission or entity isolation. Retrieval itself is not authorization. Hybrid search combines multiple result paths through RRF. RRF is a fusion process, not a switch that can be turned on. Image retrieval requires corresponding image input and model capability. Reranking requires a rerank model and must be enabled by configuration. Optional paths shown in the document flow diagram do not mean every request executes all of them.
A relevance score is not a category, entity, or permission filter. A similarity setting does not exclude unrelated businesses or unauthorized records. Authorization, approval, database writes, and external synchronization require their stated downstream implementation.
Configuration and implementation checks
| Configuration or check item | Assumed starting point for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Select one knowledge base containing qualification materials | Check the vector model identifier of each knowledge base; only matching identifiers can be associated together |
| Search mode | Hybrid search | Compare recalled content against semantic search and full-text search separately |
| Citation limit | Assumed starting point 2000 tokens for this exercise | Compare citation count and truncation against 1000 and 4000 |
| Minimum relevance | Assumed starting point 0.4 for this exercise | Compare remaining count after filtering against 0.2 and 0.6 |
| Result reranking | Enabled | Compare ranking changes against disabled; a rerank model must be available |
| Citation content handoff | Check answer against source text line by line | Inspect whether the answer corresponds to cited source text item by item |
The citation limit is expressed in tokens, the documented unit. It is not a guessed item count. The starting values above are this exercise's assumptions, not official defaults. Each needs a reason and a comparison method, as shown.
Three common errors
- Treating a retrieval hit as established authorization. Input: the draft sentence "can immediately list all Class 21 goods." Compare whether the source material contains a brand authorization chain document. The material contains only a business license and a trademark registration certificate. Observable result: authorization status is unsupported.
- Treating the minimum relevance score as a permission filter. Input: a low-relevance qualification record belonging to another entity. Compare whether that record is excluded by the threshold. Observable result: the threshold affects relevance ranking only; it does not exclude records of other entities.
- Associating knowledge bases with different vector models. Input: two knowledge bases with different vector model identifiers. Compare whether the identifiers match. Observable result: identifiers differ; exclude the non-matching one and record it; do not perform cross-model retrieval.
How to verify the setup
- Input the business license source text and the draft sentence from the assumed sample. Compare whether the authorization claimed in the draft sentence has a corresponding field in the source text. Observe that the source text does not state a brand authorization chain. Judge as unsupported.
- Input an explicit paper empty array []. Compare whether the downstream design continues through the empty citation path. Observe that the output path does not break on the empty array. Judge the handoff design as satisfied.
- Input two knowledge bases with different vector model identifiers. Compare whether the identifiers match. Observe that they differ. Exclude the non-matching one and record it. Do not proceed to compare missed recalls.
A pending request is not approval. An offered option is not execution. An unsigned termination is not a completed termination. Missing evidence remains missing. Where a document field is not provided, it is unknown.