What the documents look like
Near-expiry discount handling follows the shelf-life cycle of store inventory. The working rhythm differs from piece-by-piece weighing and sorting, where each order line is measured individually. For markdown decisions, the relevant materials are generated at store level and refreshed daily or per shift: near-expiry lists from POS stock-age data, discount approvals, damage write-off forms, and promotional shelf labels. Store inspections, ERP inventory, and quality inspection records feed into the same set of documents. Each material carries a date, a batch identity, a quantity, and an authorization state. Those fields determine whether a product-information question can be answered with a traceable citation. This page does not derive batch synchronization, full synchronization, real-time triggers, or transaction-triggered mechanisms from the stated update frequency.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that share the same vector model, and it outputs cited content as an array whose length may be zero. When the array is empty, the output path still completes. A similarity score is a retrieval ranking signal; it does not carry permission or entity-isolation duties, and the retrieval action itself does not constitute authorization. Hybrid retrieval merges multiple result paths through RRF. Result re-ranking requires a re-ranking model and must be enabled in settings. Image retrieval requires image input and corresponding model capability. These optional paths do not indicate that every request executes all of them.
Hypothetical paper exercise sample: the source material reads, "On 10 June 2025, the near-expiry list for Store A records batch L20250605 yogurt, 12 boxes, status pending discount approval; the discount approval form of the same date records that the discount application for this batch has been submitted, and the approver signature field is blank." The draft sentence to be assessed reads, "This batch discount has been approved and may be sold at the discount shelf-label price." The assessment must use the same subject, the same period, and the same field. The approval form signature field is blank, so the approved status asserted by the draft sentence is not supported and must be recorded as pending verification.
Configuration and implementation checks
| Configuration or check item | Purpose on this page | Value and comparison method |
|---|---|---|
| Associated knowledge bases | Limit retrieval scope | Verify the selected bases and material versions; exclude bases whose vector model differs |
| Search mode | Determine recall path | Choose semantic, full-text, or hybrid according to material type; hybrid merges through RRF |
| Citation limit | Control total cited tokens | For this hypothetical exercise, start at a value comparable to one near-expiry list and compare against the actual returned tokens |
| Minimum relevance | Filter low-relevance results | Effective only under semantic retrieval or result re-ranking; compare against each returned score |
| Result re-ranking | Secondary ordering | When a re-ranking model exists and is enabled in settings, compare the re-ranking scores |
| Cited content output | Hand off downstream | Check whether the citation array is passed to the answering step; treat an empty array as length zero |
Three common errors
- Treating a retrieval hit as approval authorization. Input the draft sentence and the hypothetical sample; compare the approval form signature field. When the signature field is blank, the approved status asserted by the draft sentence does not hold.
- Associating knowledge bases that use different vector models. Input the selected base identifiers; compare the vector model identifier of each base. Bases whose identifiers differ must be excluded. Do not run cross-model retrieval to compare missed recalls.
- Inferring permission or entity scope from a similarity score. Input the returned score and the record ownership; compare the score against the authorization field. The score does not serve as a basis for a permission decision.
How to verify the setup
- Input the hypothetical sample source text and the draft sentence; compare the signature field against the status asserted in the draft sentence. The approved determination is satisfied only when the signature field states approval for the same subject and the same period; otherwise record pending verification.
- Input the selected knowledge base identifiers; compare the vector model identifier of each base. The joint association condition is satisfied only when all identifiers match. When a mismatch exists, exclude the base and record the actual observation.
- Input an explicit paper citation array []; compare whether the downstream answering step continues at length zero. Observe whether the empty-result handoff design holds. Do not claim that a fictional rule retrieval necessarily returns an empty array.