Store price and display checks: citation tracing for product questions

Review citation tracing for product questions in Store price and display checks. Check input mappings, source evidence and failure handling before use.

What the documents look like

Price and display auditing happens during routine operations. Supervisors visit stores weekly or monthly and produce store visit photographs, price tag lists, display standards, and rectification notices tied to price tags, display standards, and head office promotion periods. These materials come from the supervision system, POS price lists, store cameras, and head office promotion schedules. The operating roles are the chain operations department, regional supervisors, and store managers. For citation tracing, the central question is whether each conclusion in a product information answer can be traced back to a specific document, a specific version, and a specific store object. When document versions disagree, a retrieved match is only a lead for verification and does not automatically count as an effective standard.

Constraints at the citation tracing step

The knowledge base search node can associate one or more knowledge bases that use the same vector model, and it outputs citations as an array whose length may be zero. When the array is empty, the output path still runs. Citation content must be passed to the answering step and then compared item by item against the source text. Hybrid retrieval merges multiple result paths through RRF. ReRank runs when configured and returns a 0-1 score. Minimum relevance takes effect only under semantic retrieval or result reranking. Relevance is a filter threshold and does not perform store, brand, or permission isolation. Retrieval itself is not authorization and does not show that a rectification notice has been approved. The citation limit is measured in tokens and is unrelated to the thinking budget. Image retrieval requires image input and the corresponding model capability, and reranking requires a model and runs when configured. These paths do not mean every request executes all of them.

Hypothetical drill sample: the source material to be checked consists of three sentences. "In the March 2025 price tag list for sample store A, item number P-1001 is marked at 9.9 yuan, and the effective scope is limited to that store's March promotion period." "The head office March 2025 promotion schedule states that the suggested retail price for P-1001 is 9.9 yuan and applies to directly operated stores in the East China region." "The regional supervisor's rectification notice dated 18 March 2025 records that the P-1001 price tag matches the promotion period, no rectification is needed, and the status is pending store manager confirmation." The draft sentence to be judged is: "Store A's P-1001 price has passed head office approval and completed the rectification loop." This draft sentence turns a pending status into a completed loop and provides no approval evidence.

Configuration and implementation checks

Configuration or check itemHypothetical starting point for this drillReason for the choice and comparison method
Associated knowledge basesAssociate only the knowledge bases holding the price tag list and the promotion scheduleFirst check that the vector model identifiers match; exclude those that do not, and do not run cross-model retrieval
Search modeHybrid retrievalFull-text recall for item number queries, vector recall for semantic phrasing, then RRF fusion
Citation limitSet to 2000 tokensA starting value, compared against the length of the sample passage containing the complete price tag and promotion clause, then adjusted
Minimum relevanceSet to 0.3A starting value, effective only under semantic retrieval or reranking, compared against the score distribution of recalled passages
Result rerankingEnabled and requiring an available modelSecondary ordering when candidates are many; the score is used for filtering, not for permission decisions
Citation content handlingCompare the answer against the cited source text item by itemCheck whether an answer sentence matches the cited passage on the same subject, the same period, and the same field

Three common errors

  1. Treating a retrieval match as authorization or approval evidence. In the sample, the rectification notice status is pending store manager confirmation. A retrieval match on that notice does not support a conclusion that rectification is closed or that head office approval has been given.
  2. Treating the relevance threshold as a store or permission filter. Minimum relevance only filters low-scoring results. It does not exclude unauthorized store materials and does not separate brands or regions.
  3. Treating an empty citation array as a process failure. The output length may be zero and the path still runs. The downstream step must be designed for an empty array, and the fact that no content was retrieved should be recorded.

How to verify the setup

  1. Enter the three sample source sentences and the draft sentence, then compare whether the draft sentence contains wording such as "passed head office approval" or "completed the rectification loop" that goes beyond the source text. If such wording appears, the draft is judged unsupported.
  2. Enter the three limiting conditions of store A, P-1001, and March 2025, then compare whether the returned citations share the same subject, the same period, and the same field. A match is judged consistent only when the cited passage explicitly states that status.
  3. Enter a clearly written paper input of an empty array [], then compare whether the downstream node handles it normally and outputs that no content was retrieved. This checks the empty-result branch without claiming that a fictional rule search will always return an empty array.