What the documents look like
Price and display audits for chain stores take place during routine operations, with materials refreshed on a weekly and monthly rhythm. Store visit photographs, price tag lists, display standards and rectification notices form the main body of the knowledge base. These materials carry a subject (regional supervisor, store manager), an object (a specific shelf in a specific store), a status (pending rectification, reviewed) and a validity scope (a specific promotion period). When retrieval relies on semantic similarity alone, fragments from different stores or different promotion periods can be drawn together. Operations handover work requires confirming which authorization state a material sits in before deciding whether it may enter downstream handling.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Knowledge bases using different vector models cannot be associated together. Retrieval results are output as an array, and its length can be zero, which means the output path still runs even when nothing is found. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used; it is a filtering threshold and does not perform permission or entity isolation. The reference cap is measured in tokens, so materials with large length differences need separate observation for completeness. Question optimization performs coreference resolution and question expansion, and adds one model call.
Hypothetical drill sample: a regional supervisor submits a store visit record on 10 March 2025, with the text: "On 8 March, the beverage shelf price tag at Store A showed 5.5 yuan; the headquarters March promotion schedule marks this item at 5.0 yuan; photograph number P-0310-07; status pending store manager confirmation." In the same batch, a draft handover request reads: "Please convert this price tag discrepancy at Store A directly to rectification complete." The retrieval scope is limited to the price tag list base and the display standards base covering that store and promotion period. To judge this request, compare the store visit record text with the draft sentence: the original status is pending store manager confirmation, while the draft asks for conversion to rectification complete, so the two take different values on the same field. Pending confirmation does not constitute completed rectification, and the request is not supported by the original text.
Configuration and implementation checks
| Configuration item | Assumed starting point for this page | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Price tag list base and display standards base | Both bases use the same vector model; verify the identifier before associating |
| Search mode | Hybrid retrieval | Price tag numbers are exact queries, so full-text and semantic paths are compared |
| Reference cap | Assumed starting point of 800 tokens | Compare store visit records of different lengths to see whether content is truncated |
| Minimum relevance | Assumed starting point of 0.4 | Label relevant and irrelevant samples first, then compare filtering results |
| Question optimization | Enabled for multi-turn follow-up questions | Compare the original question with the optimized question and check whether the store subject was altered |
| Result reranking | Enabled when candidates are numerous | Rerank scores run 0-1; compare ordering changes by score |
Three common errors
- Treating a retrieval hit as authorization approval. The input "Store A price tag discrepancy" can recall the store visit record, whose status is pending store manager confirmation. Retrieval returns reference content only; it does not judge whether that status may be converted to rectification complete. Check: enter that query, compare the returned reference with the original status field, and observe whether the words "pending confirmation" appear.
- Treating a pending draft sentence as already effective. The draft sentence asks to "convert directly to rectification complete," while the original status is pending store manager confirmation. Check: place the draft sentence beside the original text, compare the value on the same field, and observe whether the two are mutually exclusive.
- Associating knowledge bases that use different vector models. If the price tag list base and another base carry different vector model identifiers, they cannot be associated together. Check: verify the vector model identifiers of both bases, exclude the non-matching one, and record the actual observation without comparing missed recall results.
How to verify the setup
- Enter the store visit record text from the hypothetical sample together with the query "Store A March price tag discrepancy," and compare whether the returned reference contains photograph number P-0310-07 and the status "pending store manager confirmation." The presence of that status wording satisfies the check condition; if it does not appear, record it as not recalled and do not infer a cause.
- Enter the draft sentence "Please convert this price tag discrepancy at Store A directly to rectification complete," and compare it with the original status field. The original reads pending store manager confirmation, the draft reads rectification complete, and the two take different values on the same field, so the request is not supported. Observing this difference satisfies the condition.
- Enter an explicit paper input of an empty array [], and check whether the downstream design accepts a reference output of length zero. The empty-result path running through satisfies the condition; do not claim that a fictional rule retrieval necessarily returns an empty array.