What the documents look like
Wholesale stall trading runs on on-site bargaining and consignment shipping, and it is more fragmented than regional distributor replenishment. The documents that feed product and marketing copy drafting are purchase notes, handwritten sales slips, stall lease contracts and payment receipts. These materials are produced daily or per transaction, so a single stall accumulates multiple versions of the same field across dates. Handwriting, abbreviations and verbal agreements are common, and field definitions may not line up between one slip and the next.
For copy drafting, the retrieval step has to confirm quotable product names, specifications, quantities, price conditions and the stall's occupancy status from these materials. The search does not decide whether a draft sentence is legally true; it returns reference content that a person then compares against the source wording.
Constraints at the retrieval step
The knowledge base search node can link one or more knowledge bases that use the same vector model. Knowledge bases built on different vector models cannot be linked together. Verify the vector model identifier first, exclude non-matching bases, and do not run cross-model retrieval to compare missed recalls.
Search modes are semantic retrieval, full-text retrieval and hybrid retrieval. Hybrid retrieval merges multiple result paths through RRF. RRF is a fusion process inside hybrid retrieval, not a standalone switch. The minimum relevance value takes effect only in semantic retrieval or when result re-ranking is used; it is a similarity filter threshold and does not perform permission or entity isolation. The reference limit is measured in tokens and controls how much content a single search may cite. Question optimization handles coreference resolution and question expansion, and adds one model call before retrieval. The output reference content is returned as an array whose length can be zero, and the downstream path still runs with an empty result.
A hypothetical drill sample: a handwritten sales slip for a stall states "March 12, Stall A issued a batch of rebar to Buyer Jia, specification HRB400, quantity 20 tons, unit price at the day's market rate, payment not yet received." A stall lease contract states "Stall A lease term runs from March 1 to February 28 of the following year, rent paid quarterly." A draft sentence to be checked reads "On March 12, Stall A completed delivery of 20 tons of HRB400 rebar to Buyer Jia and settled payment in full." The material needed for the check must sit in an accessible target retrieval scope or an explicit manual input; when it is absent, record the gap.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this drill | Reason and comparison method |
|---|---|---|
| Linked knowledge bases | Link only the base holding this stall's purchase notes, sales slips, lease contract and receipts | Verify each base's vector model identifier first and exclude non-matching ones |
| Search mode | Start with full-text retrieval for specification and number questions | Compare against semantic retrieval results and observe differences in specification term hits |
| Reference limit | Set to a token amount that covers one complete sales slip | Test with materials of different lengths for content completeness, then adjust |
| Minimum relevance | Leave disabled until relevant and irrelevant samples are labelled | This value takes effect only in semantic retrieval or result re-ranking |
| Question optimization | Enable for follow-up questions containing coreference, disable for explicit questions | Compare the original question with the optimized one and check whether the subject was altered |
| Result re-ranking | Enable when candidates are numerous and a re-ranking model is available | Compare ordering before and after re-ranking and observe changes in relevant text position |
Three common errors
- Treating minimum relevance as a permission or entity filter. The value is a similarity filter threshold and does not exclude unrelated businesses or unauthorized records. Retrieval does not perform authorization judgment.
- Writing RRF as a switch that can be turned on. RRF is the fusion process for multiple result paths inside hybrid retrieval, and the documentation does not describe it as a standalone configuration item.
- Forcing knowledge bases on different vector models into one linked set. The documentation requires linking knowledge bases that use the same vector model; verify the identifier and exclude non-matching bases rather than inventing an outcome for a cross-model link.
How to verify the setup
- Input the hypothetical sales slip text and the draft sentence "On March 12, Stall A completed delivery of 20 tons of HRB400 rebar to Buyer Jia and settled payment in full." Compare "settled payment in full" in the draft against "payment not yet received" in the sales slip. The slip does not state that payment was received, so that part is unsupported; record the evidence gap.
- Input the hypothetical lease contract text and a question "Does Stall A's lease term cover March 12?" Compare the contract's stated start and end dates against the question date. When the term covers that date, the contract may be cited to describe the stall's occupancy period; when the contract does not state the scope of operating authorization, that part remains unknown.
- Input a paper empty result array [] and check the downstream node's design for handling a reference length of zero. Observe whether the downstream continues on the empty reference. This check does not claim that a fictional rule search necessarily returns []; it verifies only the empty-result continuation path.