What the documents look like
Creator-exclusive sales campaigns run on external creator fulfilment and commission settlement. The working documents are creator cooperation agreements, commission setting sheets, sample shipment forms and live-stream performance reports, drawn from selected alliance records, creator marketplace records, creator historical data and shop transaction details. Materials are updated per session and weekly.
In cross-border work a single cooperation package may exist in Chinese and in another language at the same time. Retrieval therefore has to serve natural-language questions and exact queries for numbers, model designations and proper nouns. One similarity mechanism cannot cover both needs, so the choice of search mode becomes a substantive decision rather than a formality.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases built on different vector models cannot be associated together. The vector model identifier must be checked first and non-matching bases excluded.
Semantic retrieval depends on vector distance and suits semantically close content across languages. Full-text retrieval depends on keyword hits and suits numbers and proper nouns. Hybrid retrieval uses both and merges the two result sets through RRF. The minimum relevance value applies only under semantic retrieval or when result reranking is used. It is a filter threshold and does not perform permission or entity isolation. The citation limit is measured in tokens. Cited content is output as an array whose length may be zero, and that output path still runs when nothing is found.
Hypothetical paper exercise sample. A brand business record holds this cooperation agreement abstract: "Party A commissions Party B to run one exclusive live session on 15 March 2025; the slot fee is RMB 30,000; commission is settled at fifteen percent of actual transaction value; samples are shipped by Party A." The commission setting sheet abstract in the same batch reads: "Commission rate for this session is fifteen percent; the settlement cycle is fifteen working days after the live session ends." A handover request awaiting a decision states: "Please confirm that this creator's commission rate for the session is fifteen percent and that settlement has already begun."
Configuration and implementation checks
| Configuration or check item | Value or action in this page's sample | Basis and comparison method |
|---|---|---|
| Associated knowledge bases | Associate the agreement and commission-setting bases that share one vector model | Check that the vector model identifiers match; exclude non-matching bases |
| Search mode | Compare semantic, full-text and hybrid retrieval on the commission-rate question | Observe recall differences across test questions; do not presume one mode fits all material |
| Citation limit | For this hypothetical trial, start at a value covering one agreement abstract | Test with long and short material to see whether content stays complete, then adjust |
| Minimum relevance | Label relevant and irrelevant samples first, then tune the threshold | This value applies only under semantic retrieval or result reranking |
| Question optimization | Compare the original and optimized form of a short question such as the commission question | Check whether the subject was altered; do not treat it as on by default |
| Result reranking | Enable reranking when candidates are numerous, subject to configuration | A rerank model must exist; its score can be used for filtering |
Three common errors
- Treating a retrieval hit as an authorization result. In the sample, the commission setting sheet states only a rate and a settlement cycle; it does not state that approval occurred or that settlement was executed. Returning that record does not mean settlement happened and does not mean an operating permission exists. The check is to compare the retrieved fields with the facts the handover request asserts and see whether they concern the same subject, the same period and the same field.
- Forcing knowledge bases built on different vector models into one association. The sample requires checking the vector model identifier first and excluding non-matching bases. Do not run cross-model retrieval to compare missed recall, and do not infer any runtime outcome from such an association.
- Reading a pending review state or a quotation validity period as a settled conclusion. If a material is only under review, that state does not show an absence of operating permission. A quotation validity period and a supply period are different fields and cannot stand in for each other.
How to verify the setup
- Input the agreement abstract and the commission setting sheet abstract from the sample. Compare whether the commission-rate field in the two materials concerns the same subject, the same period and the same field. When both read fifteen percent, the field is consistent; only a mutually exclusive value in the same field counts as a contradiction.
- Input the handover request awaiting a decision. Compare its assertion that settlement has already begun with what the materials state. When the materials state only a settlement cycle and do not state that settlement was executed, the assertion is unsupported and is recorded as an evidence gap.
- Input a query with no corresponding material. Observe whether the cited content output is an array of length zero, and check whether the downstream design handles the empty-result branch. Do not claim that a query against a fictional rule necessarily returns an empty array.
This page addresses one scope: creator-exclusive sales campaigns, cross-border material and multilingual retrieval, knowledge base retrieval and recall. Whether a material, action or request sits in a given authorization state at handover cannot be resolved by retrieval or by a form. Retrieval and relevance scores do not carry permission or entity isolation, and the same-vector-model requirement for associated bases does not extend to any external system.