What the documents look like
Brand outlet operations exist to clear past-season inventory. The working papers for this task are discount approvals, slow-moving inventory lists, price tags, and gross margin calculation sheets. Source data comes from ERP inventory age, POS sales, outlet mall campaigns, and historical discount records. Operating reviews and exception reviews ask a narrow set of questions: whether a given batch of SKUs has an approved discount, whether the slow-moving list and the price tags agree, and whether the discount used in a margin calculation corresponds to an approval record. These materials update weekly and monthly, so any comparison has to stay within one period boundary.
This category differs from full-price brand retail in purpose. The goal is clearing past-season stock, and the consequences of getting retrieval wrong are a damaged price structure, inventory that keeps sitting, and gross margin loss. Retrieval here supports review work; it does not decide whether a discount may be executed.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases built on different vector models cannot be associated together. When identifiers differ, the correct action is to confirm the identifier, exclude the non-matching base, and record the observation. No cross-model retrieval comparison is performed.
Retrieval recalls candidates through semantic search, full-text search, or hybrid search. Hybrid search merges multiple result channels through RRF. Reranking runs when a rerank model is available and the setting is enabled. The citation limit is measured in tokens. Minimum relevance takes effect only under semantic search or when result reranking is used. Relevance scores and retrieval results do not carry permission or entity isolation. Discount approval status and authorization boundaries must still be checked against the original documents.
A hypothetical paper exercise sample: the source material reads "2024 March slow-moving list batch 7, SKU-A and SKU-B recommended at fifty percent discount, discount approval form number PA-0307 status pending review, price tag print date 2024 March 8, gross margin calculation sheet computed at fifty percent." The draft sentence to be judged is "Batch 7 SKU-A discount has been approved and may be executed at fifty percent." The draft sentence does not match the approval status field in the source material and is unsupported.
Configuration and implementation checks
| Configuration item | Setting or check on this page | Basis and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the discount approval and slow-moving list bases that use the same vector model | Compare vector model identifiers; exclude any that differ |
| Search mode | Compare semantic, full-text, and hybrid search against the test question | No single mode is assumed suitable for all materials |
| Citation limit | This hypothetical trial starts at a shorter value to observe whether long chunks are truncated | Compare against the actual completeness of longer materials |
| Minimum relevance | Label relevant and irrelevant samples first, then adjust the threshold | Takes effect only under semantic search or reranking |
| Question optimization | Enable for short questions and multi-turn follow-ups; compare the original and optimized question | Check whether the subject was altered incorrectly |
| Result reranking | Use when a rerank model exists and the setting is enabled | Rerank scores can be used for filtering |
Three common errors
- Reading "pending review" as "approved." The input is the status field of approval form PA-0307; the comparison object is the phrase "has been approved" in the draft sentence; the observable result is that the two values differ, so the draft sentence is unsupported.
- Treating a retrieval hit as authorization to execute. The input is a discount recommendation returned by retrieval; the comparison object is the approval status and execution authority; the observable result is that the retrieval result contains no authorization information, so the discount cannot be executed on that basis.
- Associating bases that use different vector models. The input is the vector model identifiers of two bases; the comparison object is whether the identifiers match; the observable result is that the identifiers differ, so the non-matching base is excluded and recorded, and no cross-model retrieval comparison is run.
How to verify the setup
- Enter approval form PA-0307 and the draft sentence from the hypothetical sample. Compare the approval status field with the wording of the draft. If the two values differ, the draft sentence is judged unsupported.
- Enter slow-moving list batch 7 and the price tag print date. Compare whether the list date and the price tag date fall in the same period. If the date ranges differ, record the item as pending verification.
- Enter the gross margin calculation sheet computed at fifty percent and the approval form status. Compare the discount value with the approval status. If the approval status is pending review, the calculation result cannot serve as a basis for an approved discount.
For the empty-result downstream design, a paper input of [] can be used to check how the downstream step handles zero-length citations. This does not assert that a search for a fictional rule will return []. The search node outputs citations as an array whose length may be zero, and that output path still completes.