What the documents look like
Presale group buying collects orders first and then organizes procurement and fulfillment. The working materials for this page are the presale product table, the order aggregation result, the procurement plan, and the cut-off notice. Platform operations, procurement planning, and supply chain scheduling all read these materials, and retrieval output shapes downstream purchasing and sorting arrangements. The materials update daily or per session, so retrieval configuration has to serve the product and supplier information of one batch.
A presale product table for a cut-off on 12 June may state that a supplier can offer a quantity of leafy vegetables before the cut-off and give a quotation validity date. A cut-off notice may state the exact time when order collection stops and when the procurement plan is generated. These two documents carry different statuses: availability, quotation validity, order stop, and plan generation. Retrieval returns text; it does not confirm supply, approve a plan, or admit a supplier.
Constraints at the retrieval step
The knowledge base search node can be linked to one or more knowledge bases that use the same vector model. Search mode can be semantic retrieval, full-text retrieval, or hybrid retrieval, and hybrid retrieval merges multiple result paths through RRF. The citation limit is measured in tokens. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used. Question optimization performs coreference resolution and question expansion and adds one model call before retrieval. Retrieval and recall do not constitute authorization, and values collected in a form are not procurement approval or supplier admission.
A hypothetical paper exercise sample: on 10 June 2025 a platform operator prepares the presale product table for the "East China region, 12 June cut-off". The source text reads "Supplier A can supply 500 boxes of leafy vegetables before the 12 June cut-off; the quotation is valid until 11 June." A separate cut-off notice reads "order collection stops at 18:00 on 12 June; the procurement plan is generated after the cut-off." The draft sentence to be assessed reads "Supplier A has confirmed delivery of 500 boxes on 12 June, and the procurement plan has been approved." The review scope is limited to the presale document library that this operator is entitled to access.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this exercise | Reason and comparison method |
|---|---|---|
| Linked knowledge bases | Link only the libraries holding the presale product table and the cut-off notice | Both libraries use the same vector model; verify the library identifier and exclude libraries with other models |
| Search mode | Try hybrid retrieval first | Run the same set of test questions under semantic, full-text, and hybrid retrieval, and compare hits for codes and natural-language questions |
| Citation limit | Start at a token amount that covers one complete cut-off notice | Test with actual materials of different lengths to see whether content is truncated, then adjust |
| Minimum relevance | Start at 0.3 | Label relevant and irrelevant samples first, then compare whether supplier and date fields survive filtering |
| Question optimization | Enable for multi-turn follow-up scenarios | Compare the original question with the optimized question and check whether the subject was altered |
| Result reranking | Enable when candidates are numerous | Compare ordering before and after reranking and confirm whether supplier and cut-off date stay near the top |
Three common errors
- Treating a retrieval hit as confirmed supply. The draft sentence says "has confirmed delivery", while the source says "can supply", and the quotation is valid until 11 June, which does not cover 12 June. Executable check: input the draft sentence and the source text, compare the field values of "can supply" and "has confirmed", and observe whether a pending verification is written as a settled fact.
- Treating the cut-off notice as an approved procurement plan. The source says the procurement plan "is generated after the cut-off"; generation and approval are different facts. Executable check: input the cut-off notice and the draft sentence, compare "generated" and "approved", and observe whether an approval status was added.
- Linking knowledge bases across different vector models. The documents require linking knowledge bases that use the same vector model. Executable check: verify the vector model identifier of each library, exclude those that do not match, and record the actual observation without running cross-model retrieval to compare missed recalls.
How to verify the setup
- Input the hypothetical sample source text and the draft sentence, compare the field values of "can supply" and "has confirmed delivery", and observe whether the system marks the unsupported confirmation status as pending verification. The expectation is that the evidence gap is preserved and the status is not judged as confirmed.
- Input the cut-off notice and the draft sentence, compare "generated" and "approved", and observe whether the procurement plan status is marked as unsupported. The expectation is that no approval conclusion is added.
- Input a question that explicitly asks for the "12 June cut-off time", compare whether the returned content contains "18:00" and "order collection stops", and observe whether the citation limit truncates that sentence. The expectation is that the field is returned in full; if the return is empty, check the downstream design against the paper input [] without claiming that a fictional rule necessarily returns an empty array.