What the documents look like
Price and display auditing takes place during ordinary trading periods. Supervisors compare what a store actually does against head office standards on a continuing basis, rather than during onboarding or launch coaching. The working materials are store visit photographs, price tag lists, display standards, and rectification notices, with update cycles measured in weeks or months.
These materials shape what retrieval receives. A price tag list carries identifiers and price fields, which suits full-text retrieval. A display standard carries natural-language description, which suits semantic retrieval. Whether a store visit photograph can be retrieved depends on whether an image description index or an image vector index was built when the material was stored, and on the capabilities of the current vector model and vision model. The knowledge base search node can associate one or more knowledge bases that use the same vector model. Knowledge bases built on different vector models cannot be associated together.
Constraints at the retrieval step
The retrieval flow combines the optional paths the documents list: question optimization, semantic retrieval, full-text retrieval, hybrid retrieval, image description retrieval, image vector retrieval, RRF fusion, ReRank reordering, similarity filtering, and reference cap trimming. Each path runs only when its input and model capability are present. Image retrieval requires an image input and model capability. Reordering requires a ReRank model and must be enabled in the settings. The reference cap is measured in tokens. The minimum relevance value takes effect only under semantic retrieval or when result reordering is used. Retrieval is a recall operation and does not perform authorization decisions. The authorization state of materials, operations, or requests during an operational handover must be confirmed separately by downstream business steps.
Hypothetical drill sample: a chain brand operations department intends to check a Store Price Tag Execution Check Sheet (hypothetical sample). The source text reads: "Store number A-017, 10 March 2025 to 16 March 2025, shelf price tag shows unit price 9.9 yuan, the POS price list for the same product in the same period shows unit price 9.9 yuan, the display standard requires this product to sit on the second layer of the end cap." One draft sentence awaits judgement: "In this period the store's price tag and POS price are consistent, and the display position meets the standard." That sentence covers three facts: price tag, POS, and display position. Each must be checked separately. The display position appears in the source text only as a standard requirement; the source does not record an actual photograph observation result.
Configuration and implementation checks
| Configuration item | Setting or check relevant to this page | Relation to the sample |
|---|---|---|
| Associated knowledge bases | Associate one or more knowledge bases that use the same vector model | Confirm the vector model identifier is identical across the price tag list, display standard, and store visit photographs |
| Search mode | Compare semantic, full-text, and hybrid retrieval against test questions | Price tag identifier queries go to full-text retrieval; display description queries go to semantic retrieval |
| Reference cap | Unit is tokens; test with materials of differing length to see whether content stays complete | A hypothetical trial starting value may be set to cover one price tag record plus one display standard passage, then compared across values to see whether references are truncated |
| Minimum relevance | Takes effect only under semantic retrieval or result reordering | Label relevant and irrelevant samples first, then adjust the threshold; it is not used to exclude unrelated businesses or permissions |
| Question optimization | Used for coreference resolution and question expansion | Compare the original question with the optimized question, and check whether store number and period were altered |
| Result reordering | Requires a ReRank model and must be enabled in settings | Reorders text results when candidates are numerous; image retrieval results are fused together with them |
Three common errors
- Treating retrieval output as an authorization basis. In the hypothetical sample, a matching price tag and POS price shows only that two field values agree. It does not show that the store is permitted to apply that price, and it does not show that rectification is complete. Retrieval does not perform authorization decisions.
- Associating knowledge bases that use different vector models. The documents require associated knowledge bases to share the same vector model. When identifiers differ, check and exclude the non-matching base first. Do not run cross-model retrieval to compare missed recalls.
- Writing a pending or unsigned state as completed. In the hypothetical sample, if a rectification notice is still pending review, it cannot be described as completed rectification. Signing, reaching cooperation, and contract effectiveness are three distinct facts.
How to verify the setup
- Input the hypothetical sample source text and the draft sentence. Compare "price tag and POS price are consistent" in the draft against the two field values in the source. Observe whether both fields concern the same subject in the same period and carry the same value. When those conditions hold, that clause is supported. The display position clause is recorded as awaiting verification, because the source does not state an actual observation result.
- Input the store number and period from the hypothetical sample. Compare the subject and time range in the retrieval terms before and after question optimization. Observe whether the store number and the start and end dates were rewritten. If they were rewritten, record it as a deviation introduced by question optimization and require manual checking.
- Input a store number that clearly does not exist. Observe whether the returned reference content is an empty array. The empty-result output path still completes. Downstream design can be checked against a clearly stated paper input of an empty array. Do not assert that a fictional rule query necessarily returns an empty array.
This page addresses only chain store price and display auditing / product material and selection questions / knowledge base retrieval and recall. Which materials, operations, or requests sit in what authorization state during an operational handover cannot be treated as authorized execution on the basis of retrieval or a form.