What the documents look like
Short-video affiliate distribution relies on long-term content exposure and product-link conversion, without depending on live explanation. The daily document surface includes short-video scripts, product lists attached to the cart, commission plans, and violation feedback. A promotional sentence in a script that lacks a checkable source can lead to video removal and commission loss when violation feedback arrives. This page addresses only the citation-source and tracing step within copy drafting. The object of work is the correspondence between a proposed sentence and the materials it relies on.
The materials involved are scripts, cart product lists, commission plans, and violation feedback. A script may contain claims about supply relationships, product specifications, or quality. A product list may state quantities and batch scope. A commission plan may state rates and periods. Violation feedback may state which sentence triggered a review. The citation-tracing step checks whether each proposed sentence can be matched to a stated field in these materials.
Constraints at the citation tracing step
The knowledge base search node outputs cited content as an array, and the length may be zero. When no content is found, that output path still completes. The cited content is passed to the subsequent answer step, where each answer is checked against the cited original text.
Knowledge bases that can be associated together must share the same vector model. Knowledge bases with different vector models cannot be associated together. The vector model identifier must be checked first, and those that do not match are excluded. The minimum relevance score is a filter threshold that takes effect only under semantic retrieval or when result reranking is used. It does not perform permission or entity isolation. Retrieval fetches content; authorization judgment and business verification require separate implementation.
Hypothetical paper exercise sample: The source material reads "This product is supplied by XX Trading Co., Ltd., and the supply period is March 1, 2026 to March 31, 2026." "The packaging of this batch of products is marked with a net content of 250 grams." The proposed sentence to be assessed is "This product is supplied by XX Trading Co., Ltd. on a long-term basis, with a net content of 250 grams, and stable and reliable quality." The subject, object, and period range needed for judgment are all stated in the original text.
Configuration and implementation checks
| Configuration or check item | Hypothetical value for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge base | Single knowledge base, vector model identifier A | Compare with another knowledge base identifier B; if identifiers differ, do not associate together |
| Search mode | Hybrid retrieval | Compare with semantic retrieval and full-text retrieval; numbered terms require keyword matching |
| Minimum relevance score | Hypothetical starting point 0.3 | Compare with 0.5 and observe changes in the number of recalled items; this value is a trial starting point |
| Citation upper limit | Entered in tokens, value to be determined | Compare with a smaller value and observe changes in the number and length of citations |
| Result reranking | Enabled according to settings | Compare with not enabled and observe ranking results; reranking requires a model |
| Citation input handoff | Check whether retrieval output is passed to the answer step | Check each answer against the cited original text |
Three common errors
- Writing "long-term supply" when the material states "March 1, 2026 to March 31, 2026." Input is the source material and the proposed sentence. Compare the period field values. When the period ranges do not match, the sentence is not supported.
- Writing a net-content commitment for all batches when the material states "the packaging of this batch of products is marked with a net content of 250 grams." Input is the two statements. Compare the object scope. When the object is expanded from a single batch to all batches, the sentence exceeds the original text.
- Deriving that the material has endorsed the evaluation from the unsourced phrase "stable and reliable quality." Input is the proposed sentence and the source material. Compare whether a corresponding field exists. When the material does not state this evaluation, the sentence is not supported.
How to verify the setup
- Input the hypothetical source material and the proposed sentence. Compare the three fields of subject, object, and period. The condition is met when all three fields in the proposed sentence fall within the range of the original text and no new evaluation is added. When "long-term" or "stable and reliable quality" appears, record it as not supported.
- Input the vector model identifiers of two knowledge bases. Compare whether the two are identical. They can be associated together only when the identifiers match. When the identifiers do not match, exclude that knowledge base and record the identifier value. Do not perform cross-model retrieval to compare missed recalls.
- Input a paper empty array [] to check the downstream handoff design. Observe whether the downstream continues to run through with empty citations and record the actual observation. Do not claim that a particular retrieval will necessarily return an empty array.