What the documents look like
Creator sales campaigns depend on external creator performance and commission settlement, so copy drafting starts from source material that can be checked line by line. The working documents are creator cooperation agreements, commission setting sheets, sample shipment sheets, and livestream performance reports. These materials update per session or per week.
A price, gift, specification, commission rate, or after-sales promise that appears in draft copy requires a matching field in the source material. When that field is absent, the sentence is unsupported. Version timing also matters: the same product can use different agreement or setting sheet versions across sessions, so drafting must confirm which version is being cited.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that share the same vector model. Its output is a citation array whose length may be zero; even when nothing is retrieved, the output path still completes. The minimum relevance value 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, and retrieval itself is not authorization. Hybrid retrieval merges multiple result paths through RRF. Reranking requires a rerank model and must be enabled by configuration. Image retrieval requires corresponding image input and model capability. Knowledge bases with different vector models cannot be associated together; the handling step is to check the identifiers first, exclude the non-matching ones, and record what was observed without running a cross-model retrieval for comparison.
Hypothetical paper exercise: the source material reads "The product for this session is Model A thermos, agreement number H-001, cooperation date March 8, 2025; the commission setting sheet states a commission rate of 15% for this session; the sample shipment sheet states a shipment quantity of 2 units." The draft sentence to be judged is "The commission rate for Model A thermos in this session is 15%, and the gift is 1 unit of the same thermos." On checking, the commission rate matches the same field in the commission setting sheet. The gift item has no corresponding field in the source material, so it is unsupported; the evidence gap must be recorded, and the sentence cannot be judged as a negation or a conflict.
Configuration and implementation checks
The table lists only documented settings or implementation checks relevant to this page's primary task. The values are starting points for this hypothetical exercise, chosen so that each item can be compared against the same field in the source material.
| Configuration or check item | Hypothetical starting point for this exercise | Comparison method and notes |
|---|---|---|
| Associated knowledge bases | Select the knowledge base containing the agreement, commission setting sheet, and sample shipment sheet | Check each knowledge base's vector model identifier first, and exclude those with a different vector model |
| Search mode | Hybrid retrieval | Semantic and full-text results are merged through RRF, so numbered content can be matched by the full-text path |
| Result reranking | Enabled per configuration | When a rerank model is available, candidates are reordered and the score is used for filtering |
| Minimum relevance | Hypothetical starting point 0.3 for this exercise | Compared against the rerank score; it acts only as a filter threshold and does not perform permission judgment |
| Citation limit | Hypothetical starting point 2000 tokens for this exercise | Citation content is trimmed by tokens and compared against the evidence volume the draft sentence requires |
| Citation output | Array, length may be zero | An empty array is handled by downstream design and is not treated as a workflow error |
Three common errors
- Treating the minimum relevance value as a permission or entity filter. The input is one retrieval request; the comparison object is the filter scope described in the documentation; the observable result is that this setting affects relevance filtering only and does not exclude unauthorized records.
- Treating a pending review or a quotation validity period as an already effective fact. The input is the draft sentence to be judged and the source material; the comparison is the value of the same field; the observable result is that a field not stated in the material is recorded as unsupported and is not written as a negation or a conflict.
- Treating an optional retrieval path as executed on every request. The input is one text query containing no image; the comparison object is the path conditions listed in the documentation; the observable result is that image retrieval and reranking are enabled under their respective conditions and produce no corresponding result when not enabled.
How to verify the setup
- Input the source material and draft sentence from the hypothetical exercise above, and compare the commission rate field. The observable result is that the sentence is supported when both values are 15%; if the source material does not state a gift, that part remains pending verification.
- Input one text query and observe the citation array, comparing the number of returned entries against the citation limit. The observable result is that the array length falls between zero and the limit, and the downstream path still completes when the array is empty.
- Input an explicit paper empty result [] and compare it against the downstream handling design. The observable result is that downstream processes it as an empty citation; this check does not claim that a fabricated rule retrieval necessarily returns an empty array.