What the documents look like
Short-video affiliate distribution relies on long-term content exposure and product-link conversion, with no dependence on live explanation. The working materials are short-video scripts, product lists for attached links, commission plans, and violation feedback, updated daily or weekly. Contract and transaction material comparison works on the subject, object, period, and authorization state carried in these materials.
At the retrieval step, the input is a verification question raised by operations staff, and the output is an array of cited content whose length can be zero. Retrieval does not decide authorization and does not replace manual verification. Whether material has been trained and whether an index exists determine whether retrieval can recall it. Form values, approval status, and business actions each require their own downstream implementation.
Constraints at the retrieval step
Contract and transaction material comparison requires the same subject, the same period, and the same field. A contradiction exists only when one field carries clearly mutually exclusive values. A field that is not stated, or that carries only related-stage evidence, counts as unsupported or pending verification. A pending review does not demonstrate a lack of operating permission. Signing, reaching cooperation, and contract effectiveness are three distinct facts. A quotation validity period does not indicate a supply period.
A knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases with different vector models cannot be associated together; the identifier must be checked first and non-matching bases excluded, without running cross-model retrieval to compare missed recalls. Semantic recall and relevance scores do not handle permission or entity isolation, and a similarity setting does not exclude unrelated businesses or unauthorized records.
Hybrid retrieval merges multiple result paths through RRF. RRF is a fusion process, not an on/off switch. Full-text retrieval suits exact queries such as numbers, models, and proper nouns. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used. The citation limit is measured in tokens. Question optimization handles reference resolution and question expansion and adds one model call. Image retrieval requires corresponding image input and model capability, and reranking requires a model and must be enabled by configuration. The optional paths shown in the overall document flow diagram do not mean every request executes all of them.
Hypothetical paper exercise sample: a content operator submits a product list for attached links. Two sentences of the source text read: "This list applies to short-video promotion with attached links from March 1, 2025 to March 31, 2025. The commission rate follows the rate displayed on the platform commission plan page on the day of execution." A draft sentence reads: "The commission rate in this list is fixed at ten percent for the entire month of March 2025." The handover request to be judged is: "Please confirm that this draft sentence can be published externally as is." The source text does not state a fixed rate, and it does not state publication authorization.
Configuration and implementation checks
| Configuration item | Starting point for this hypothetical exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only one knowledge base whose vector model identifier matches | Check each base's vector model identifier first; exclude non-matching bases and do not associate them together |
| Search mode | Run semantic retrieval and full-text retrieval once each | Compare how the two modes recall the date and rate fields in the sample, then decide whether to use hybrid retrieval |
| Citation limit | Set to the token volume needed to cover the two source sentences in the sample | This is a starting point for this exercise; test with materials of different lengths to see whether content stays complete, then adjust |
| Minimum relevance | Keep the current value first, then adjust against labeled samples | Takes effect only under semantic retrieval or result reranking; label relevant and irrelevant samples before adjusting the threshold |
| Question optimization | Enable for questions containing references, disable for already clear questions | Compare the original question with the optimized question and check whether the subject was altered |
| Result reranking | Enable when candidates are numerous | Requires a reranking model enabled by configuration; the reranking score can be used for filtering |
Three common errors
- Treating a retrieval hit as authorization confirmation. The input is the handover request in the sample; the comparison object is the cited content returned by retrieval against the authorization field. When the retrieval result contains no publication authorization, the request remains pending verification.
- Treating an unstated field as a contradiction. The input is the sample source text and the draft sentence; the comparison object is the commission rate field. The source text does not state a fixed rate, so the fixed rate in the draft sentence is unsupported and does not form a contradiction with the source.
- Associating knowledge bases with different vector models together. The input is the vector model identifiers of several knowledge bases; the comparison object is whether the identifiers match. When identifiers do not match, exclude those bases and do not run cross-model retrieval to compare missed recalls.
How to verify the setup
- Input the two source sentences and the draft sentence from the sample. Compare whether the fixed rate in the draft sentence has support in the source text. The observed result is that the source does not state that rate, so the draft sentence should be marked unsupported.
- Input the handover request. Compare whether the cited content returned by retrieval contains the publication authorization field. The observed result is that the cited content does not contain that field, so the request stays pending verification and does not enter a publication action.
- Input the vector model identifiers of the knowledge bases. Compare whether the identifiers match. The observed result is that non-matching bases are excluded, retrieval runs only against matching bases, and the actually observed recalled content is recorded.
An empty-result handoff can be checked against the downstream design using the explicit paper input [], without claiming that a fictional rule retrieval necessarily returns []. Retrieval checks record observations and failure branches only; they do not guarantee a first hit, full recall, or that an unrelated query returns nothing.
This page addresses only: short-video affiliate distribution operations / contract and transaction material comparison / knowledge base retrieval and recall. It covers which materials, operations, and requests sit at which authorization state during an operations handover, and it does not treat retrieval or a form as authorization execution.