What the documents look like
Tmall flagship store daily operations for product information and selection Q&A draw on four kinds of material: the product publishing page, the brand authorization letter, the delivery commitment record, and the after-sales ticket. These materials come from the Tmall merchant backend, the Tmall rules center, the ERP order table, and customer service chat records. The product publishing page carries category qualifications and selection parameters. The brand authorization letter is used to check flagship store identity and authorization scope. The delivery commitment record and after-sales ticket carry fulfillment and after-sales status. A citation in selection Q&A must be traceable to a specific field in these materials, and the material version and validity range directly determine whether the citation holds.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that share the same vector model, and it outputs cited content as an array whose length may be zero. When the citation array is empty, the output path still runs through; this alone does not establish that the workflow errored out or terminated. Retrieval output must be passed to the answering step, and the answer must be checked item by item against the cited source text. A similarity score reflects only how related the content is to the question; it does not perform permission filtering or entity isolation. The citation limit is measured in tokens, which is a different parameter from the thinking budget and cannot be mixed with it. Hybrid retrieval merges multiple result paths through RRF; reranking requires a model and must be enabled by configuration; image retrieval requires image input and model capability. These paths are not guaranteed to execute on every request.
Hypothetical drill sample: the original text of a Tmall flagship store product publishing page states "This store's brand authorization letter number is A-2025-031, the authorization validity period ends December 31, 2025, and the covered category is home appliances." The original text of the brand authorization letter states "The authorized party is the Tmall flagship store, the authorization period is January 1, 2025 to December 31, 2025, and the authorized category is home appliances." A draft sentence to be judged reads "The flagship store's brand authorization already covers home appliance category selection Q&A for 2026." Checking the same field: the authorization period ends December 31, 2025, while the draft sentence claims 2026 coverage. The two take mutually exclusive values on the same authorization period field, so the draft sentence is not supported.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this trial | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Select the product publishing page and the brand authorization letter as two knowledge bases | Check that both vector model identifiers match before associating them together; exclude any whose identifier does not match |
| Search mode | Hybrid retrieval | Selection Q&A contains model numbers and proper nouns; compare keyword hit results against pure semantic retrieval |
| Citation limit | Set to 3000 tokens | This is a hypothetical starting point for this trial; compare citation count and truncation against a 1500-token result |
| Minimum relevance | Set to 0.5 | This is a hypothetical starting point for this trial; compare whether low-relevance fragments are filtered against a 0.3 result |
| Question optimization | Enabled for multi-turn follow-ups | Compare against the disabled state to see whether the retrieval query is completed in continuous conversation |
| Citation output handling | Connect the citation array to the answering node | Check whether the answer corresponds item by item to the cited source text |
Three common errors
- Treating a similarity score as permission filtering. The input is a selection question outside the authorized scope; compare the retrieval score against the authorization field. A high score does not mean the authorization covers it, and the valid range of the authorization letter must be checked manually.
- Treating an empty citation array as a workflow failure. The input is a query with no matching content in the knowledge base; compare the node output against the downstream design. The empty array still runs through the output path and must be handled as zero length.
- Treating a pending-review status as already effective. The input is a pending-review compensation request in an after-sales ticket; compare the ticket status field against the compensation completion record. Pending review does not establish that compensation has been executed.
How to verify the setup
- Input the original text of the product publishing page and the brand authorization letter from the hypothetical sample above, and compare the authorization period field against the 2026 coverage claimed in the draft sentence. If the same field takes mutually exclusive values, the draft sentence is not supported, and the evidence gap must be recorded.
- Input the original text of the brand authorization letter from the hypothetical sample above, and compare the authorized category field against the category involved in the selection Q&A. If the categories match and the date is within the validity period, the citation holds; if it is beyond the validity period, the citation does not hold.
- Input a query with no matching content in the knowledge base, and compare the node output against the downstream answering node design. If the output is a zero-length array and the downstream handles it normally, the empty-result path meets expectations; if the downstream errors, the handling logic must be checked.