What the documents look like
Brand member private-domain operations work from CRM systems, enterprise messaging, POS purchase records and mini-program behavior data. Daily documents include member tag tables, outreach plans, coupon rules and privacy authorization records. Inventory procurement and supplier materials form one category inside this work: supplier quotations, purchase requests, inbound records and supply status notes. These materials update daily or weekly, and retrieval needs concentrate on checking one supplier's quotation, supply status and authorization boundary within one period. Member private-domain operations do not handle store scheduling, so retrieval targets lean toward supplier entity, material object, procurement stage and effective scope.
The primary task on this page is knowledge-base retrieval and recall for inventory procurement and supplier materials. Retrieval locates content. It does not approve a supplier, release a purchase, or confirm that a quotation has passed review.
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; check the model identifier and exclude any base whose identifier does not match. Search modes are semantic retrieval, full-text retrieval and hybrid retrieval. Hybrid retrieval fuses multiple result paths through RRF, and RRF is a fusion process rather than a standalone switch. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used; it is a filter threshold and does not perform permission or entity isolation. The reference limit is measured in tokens and controls the maximum content referenced per search. Question optimization performs coreference resolution and question expansion, and adds one model call before formal retrieval. The search output is an array of references whose length can be 0, and the output path still completes when nothing is found. Retrieval itself does not grant authorization; whether procurement staff may use supplier materials depends on the permission implementation in downstream business systems.
Hypothetical paper exercise sample: Supplier A submitted a quotation on 1 March 2025, material number M-100, quotation valid until 31 March 2025, pending review by the procurement supervisor. Draft sentence under review: "Supplier A has passed review and the March supply price is locked." Fields to check: review status, supply period, basis for the price lock. This sample is a paper exercise and does not represent any real entity or actual approval.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the supplier materials base, and confirm all associated bases use the same vector model | Bases on different vector models cannot be associated together; check identifiers first and exclude mismatches |
| Search mode | Start with full-text retrieval for material number M-100 | Number-type queries depend on keyword hits, so compare against the semantic retrieval result |
| Reference limit | Start at a token amount that covers one complete quotation | Test content completeness with quotations of different lengths, then adjust the limit |
| Minimum relevance | Label related and unrelated samples first, then set the threshold | This value takes effect only under semantic retrieval or result reranking |
| Question optimization | Enable for continuous follow-up questions and compare the original question with the optimized question | Check whether the subject is altered, then evaluate short-question scenarios |
| Result reranking | Enable when candidates are numerous and filter by reranking score | Reranking scores run from 0 to 1; compare with vector scores before deciding |
Three common errors
- Treating a retrieval hit as an approval. Input "Supplier A quotation" and recall the document, then compare the review status field in the document and observe whether it still reads "pending review by the procurement supervisor." A retrieval result does not change an approval state.
- Treating the quotation validity period as the supply period. Input the sample source text and compare "quotation valid until 31 March 2025" with the draft sentence "the March supply price is locked," then observe whether the source states a supply period. A field that is not stated remains unsupported.
- Associating knowledge bases that use different vector models. Check the vector model identifier of each base, exclude those whose identifiers do not match, and record the actual observation. Do not run a cross-model search to compare missed recalls.
How to verify the setup
- Input the hypothetical sample source text and the draft sentence, compare "pending review by the procurement supervisor" with "has passed review," and observe whether the draft sentence contradicts the status field in the source. Expected: the draft sentence is not supported by the source text.
- Input material number M-100 and run full-text retrieval and semantic retrieval separately, compare whether each returned path contains the quotation for that number, and observe whether full-text retrieval hits the number field. Expected: number-type queries rely on full-text retrieval.
- Input an explicit paper empty result [] and check whether the downstream design handles array length 0, then observe whether that output path continues to complete. Expected: an empty result does not interrupt the flow, and this does not mean the retrieval rule necessarily returns an empty array.