What the documents look like
Near-expiry markdown handling follows the shelf-stock lifecycle. It does not weigh and sort individual order lines. The working records are near-expiry lists, markdown approvals, damage write-off forms and promotional price tags. Data arrives from POS stock age, store inspections, ERP inventory and quality inspection records, and it refreshes daily or by shift.
Store service and internal training rely on these records to carry markdown rules, approval wording and isolation requirements into a knowledge base that store managers, loss administrators and finance auditors can retrieve. Because the source material changes quickly, a retrieved passage reflects the text as it was stored. Retrieval does not verify the current batch state and does not authorize a markdown or a write-off.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Retrieval paths include semantic retrieval, full-text retrieval, hybrid retrieval, question optimization, RRF fusion and ReRank. Image retrieval requires an image input and the relevant model capability. ReRank requires a model and must be enabled in the settings. The minimum relevance value takes effect only in semantic retrieval or when result reordering is used. The citation limit is measured in tokens.
The same-vector-model requirement applies only to the knowledge bases this node can associate. It does not extend to external systems. Retrieval and similarity scores do not perform permission or entity isolation. Forms collect values. Business verification, approval, database writes and external synchronization each need their own downstream implementation.
Hypothetical paper exercise: a store knowledge base contains a document titled "Near-expiry markdown approval wording (paper sample)". The source text reads: "Batch A entered the near-expiry list on 10 March. The markdown request is submitted by the store manager, reviewed by the loss administrator, and may be executed only after confirmation by the finance auditor. Before confirmation, the batch keeps its original price and is isolated separately." A draft sentence under review reads: "Batch A has completed markdown approval and the new price tag is in effect." The source text describes a submission and review process. It does not state that confirmation was completed. The draft sentence is unsupported.
Configuration and implementation checks
| Configuration or check item | Assumed starting point for this exercise | Reason and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the store operations base that uses one vector model | Check the vector model identifier first, exclude mismatches, and do not run cross-model retrieval |
| Search mode | Run semantic retrieval and full-text retrieval separately | Compare whether the sample source text is recalled under each mode; check full-text hits separately for identifier-style questions |
| Citation limit | Start at a token amount that can hold the full sample sentence | Test with source material of different lengths to see whether content stays complete, then adjust |
| Minimum relevance | Label relevant and irrelevant samples first, then set a threshold | This value takes effect only in semantic retrieval or result reordering, and is not a permission filter |
| Question optimization | Enable for follow-up questions containing references, disable for comparison on explicit questions | Compare the original question with the optimized question and check whether the subject was altered |
| Empty-result handling | Use a paper input of an empty array to check the downstream design | Citation content can have a length of 0, and the chain still runs |
Three common errors
- Treating a retrieval hit as markdown authorization. The input is the retrieved approval wording. The comparison object is the draft sentence "has completed markdown approval and the new price tag is in effect". The observable result is that the source text does not state confirmation was completed, so the sentence is unsupported. Retrieval and similarity do not perform authorization decisions.
- Writing a pending review as approved. The input is the sample wording "submitted by the store manager, reviewed by the loss administrator". The comparison object is "may be executed only after confirmation by the finance auditor". The observable result is that review and confirmation are two stages. A pending review does not show that operating permission is absent, and it is not an approval.
- Forcing knowledge bases with different vector models into one association. The input is two knowledge bases with different vector model identifiers. The comparison object is the node requirement of the same vector model. The observable result is that mismatched identifiers should be excluded and recorded. Do not infer the runtime result of an association that was not formed.
How to verify the setup
- Input the sample source text and the draft sentence. Compare whether the source text contains subject, object, state and effective scope. The observable result is that the source text supports the submission and review process, while the draft sentence's "confirmation completed" is unsupported and is recorded as pending verification.
- Input the same question through semantic retrieval and full-text retrieval. Compare the recalled content from both paths. The observable result is that recall of the sample source text in at least one path meets the condition. When it is not recalled, record the gap. Do not guarantee a first-position hit or full recall.
- Input a paper empty array to check the downstream chain. Compare the citation content length with the downstream node design. The observable result is that the chain still runs when the length is 0. Do not claim that a query against a fictional rule necessarily returns an empty array.