What the documents look like
Near-expiry markdown handling follows the shelf-stock lifecycle rather than order-by-order weighing and sorting. The material set for this task is the near-expiry list, the markdown approval, the damage report, and the promotional price tag. Source records behind these documents include POS stock age, store inspection, ERP inventory, and quality inspection records. Stock age and inspection results determine whether a batch enters the near-expiry list. The markdown approval determines whether a price tag change is permitted. The damage report determines whether unsellable goods can leave the shelf. In an operations handover, these documents carry authorization state, so the handover pair must be able to see what state a material, an operation, or a request is actually in. Retrieval results and form entries are not approvals.
The update rhythm of this category is daily or per shift. That rhythm affects how often the handover pair must re-read the current near-expiry list against the prior shift copy, and it affects how quickly an unsigned approval can become stale. It does not by itself create any synchronization or trigger behavior.
Constraints at the citation tracing step
The knowledge base search node can associate one or more knowledge bases that share the same vector model. Its citation output is an array whose length may be zero; an empty array does not mean the workflow has errored or stopped, and the output path still runs. Retrieval returns citations. It does not perform authorization judgment. The minimum relevance value is a filter threshold that takes effect only in semantic retrieval or when result reranking is used; it is not a permission or entity isolation control. Hybrid retrieval merges multiple result paths through RRF, which is a fusion step in retrieval and not a switch. Reranking requires a rerank model and must be enabled per settings. Image retrieval requires corresponding image input and model capability. These optional paths do not all execute on every request. Knowledge bases with different vector models cannot be associated together; the identifiers should be checked first and non-matching ones excluded, with the actual observation recorded rather than an inferred runtime result.
Hypothetical drill sample: a store shift handover record dated 10 June 2025 states: "Near-expiry list item 12, object is cold cabinet A batch yogurt, 40 units, status pending markdown approval, inspector has signed, approval column blank." The draft handover sentence reads: "Item 12 has received markdown approval, price tag may be changed and the goods moved to unsellable isolation." Checking means taking the source text above and the draft sentence, then comparing the object, batch, quantity, status field, and approval column value. The source approval column is blank; the draft sentence claims approval was received. The two are mutually exclusive on the same field, so the draft sentence is unsupported. Sellable versus unsellable disposition requires an approval result. A handover request cannot substitute for approval.
Configuration and implementation checks
| Configuration or check item | Value or state on this page | Basis and note |
|---|---|---|
| Associated knowledge bases | Hypothetical starting point for this drill: 1 | Used to check the version of the near-expiry list and markdown approval material; only the same vector model may be associated together |
| Citation content output | Array, length may be zero | An empty array does not mean the workflow has errored or stopped; the downstream must accept an empty result |
| Minimum relevance | Does not carry permission filtering | This value takes effect only in semantic retrieval or result reranking and is a relevance threshold |
| Rerank | Enabled per settings | A rerank model is required; when not enabled this path is not taken |
| Citation limit | The document uses tokens as the control unit | Top k is not used, to avoid unstable results from differing chunk lengths |
| Handover check action | Manual comparison of source text and draft | Compare subject, object, and status fields item by item; retrieval does not produce authorization |
Three common errors
- Treating a retrieval hit as authorization. Input a markdown approval query, compare the returned citation against the actual approval column value, and observe whether the presence of a citation is misread as approval. Retrieval returns citation content only; approval state must follow the material field.
- Writing a pending review as approved. Input the handover material source text, compare "pending markdown approval" against the draft sentence "has received approval," and observe whether the same status field is mutually exclusive. A blank approval column is unsupported, and the evidence gap must be recorded.
- Treating an empty citation array as an error. Input a clearly paper-based empty array, compare whether the downstream design accepts it, and observe whether it is judged as process termination. The document states the length may be zero and the output path still runs.
How to verify the setup
- Input the hypothetical sample source text above together with the draft handover sentence, compare the approval column value against the draft sentence wording, and when the draft claims approval while the source approval column is blank, judge it unsupported and record it as pending verification.
- Input a paper-based empty array, compare whether the downstream node continues to accept it, and when the output path runs without error, this matches the documented description of citation content output.
- Input two knowledge base identifiers that use different vector models, compare whether the identifiers match, and when they do not match, exclude joint association, record only the check result, and do not run a cross-model retrieval comparison.