What the documents look like
Contract and transaction material comparison in supermarket checkout and loyalty promotion work centres on four document types: promotional price tags, loyalty points rules, checkout daily reports, and abnormal refund slips. These materials come from the price tag system, the member CRM, the POS system, and the payment platform. Updates occur daily or per shift, so any comparison must first lock the material scope to the same date or the same shift. Prices and member benefits tie directly to the amount a customer pays at settlement. When the citation relationship between materials breaks, the practical consequences include price complaints, cash overages and shortages, and disputes over member benefits.
This page addresses only the citation source and tracing step within contract and transaction material comparison. That step confirms which material, which version, and which field a given comparison conclusion draws its supporting text from. A conclusion about a promotional price must point back to the price tag record that carries the effective date. A conclusion about points accrual must point back to the loyalty rules text that states the applicable threshold. A conclusion about settlement must point back to the checkout daily report. A conclusion about a refund must point back to the refund slip and its recorded status.
Constraints at the citation tracing step
Citation source and tracing requires that every comparison conclusion can be traced to a specific position in a specific material. The effective period of a promotional price tag, the applicable threshold in loyalty points rules, the actual received amount in a checkout daily report, and the approval status on an abnormal refund slip each belong to different fields. These fields cannot substitute for one another. Only explicitly mutually exclusive values within the same field constitute a contradiction. A value that is not stated, or for which only related-stage evidence exists, remains unsupported or pending verification.
The search node outputs cited content as an array, and the length may be zero. An empty array means the output path runs through without matching content; it does not mean the workflow reported an error or terminated. Knowledge base search can associate only knowledge bases that use the same vector model. Knowledge bases with different vector models cannot be associated together. The minimum relevance score is a filtering threshold. It does not perform permission isolation or entity isolation. Whether a retrieved result is usable still requires manual review against the valid version.
A pending status on a refund slip is not approval. An offered option is not execution. An unsigned termination is not a completed termination. Retrieval is not authorization. These distinctions matter because the citation tracing step checks what the source text actually states, not what a downstream reader might assume.
Configuration and implementation checks
The following configuration items support citation tracing verification for this page's sample. Hypothetical drill sample: a store shift on 10 March 2025. The promotional price tag text reads: "Member day selected items 10% off, effective 10 March 2025 to 12 March 2025." The loyalty points rules text reads: "Member day spending earns double points, applicable to items marked 'member day'." The checkout daily report text reads: "10 March actual received amount total 12,450 yuan." The abnormal refund slip text reads: "Refund slip number R20250310003, status is pending review." The draft sentence to be assessed reads: "The 10 March member day promotion was executed at 10% off, double points took effect, and refund slip R20250310003 has completed the refund."
| Configuration item | Hypothetical trial starting point | Comparison method and reason |
|---|---|---|
| Associated knowledge bases | Promotional price tag base, loyalty points rules base | Check whether the two vector model identifiers are the same; only identical identifiers allow joint association |
| Search mode | Hybrid search | Price tag numbers and refund slip numbers are exact queries; hybrid search covers both semantic and keyword matching |
| Citation upper limit | Hypothetical starting point of 3000 tokens for this trial | Observe whether cited content length covers the key fields of all four materials; compare with the length of the daily report received-amount passage |
| Minimum relevance score | Hypothetical starting point of 0.3 for this trial | Observe whether low-relevance fragments are filtered; compare with the score distribution after reranking is enabled |
| Result reranking | Enabled | Secondary sorting when candidate results are numerous; compare with the ordering before reranking |
| Citation content handoff | Search output passed to the answer step | Check each answer sentence against the corresponding field in the cited source text |
Three common errors
- Treating a retrieval hit as authorization confirmation. Retrieving the refund slip text "status is pending review" does not permit the inference that the refund has been executed. Pending review is not approval. The draft sentence's claim that the refund "has completed the refund" lacks support.
- Treating values from different fields as a contradiction. The effective date on the promotional price tag and the received amount in the checkout daily report belong to different fields. A difference between them is not a contradiction. Only mutually exclusive values within the same field constitute a contradiction.
- Treating an empty citation array as a system failure. The search node outputs a citation array whose length may be zero. An empty array means no content was matched; the workflow still runs through. It cannot be used to conclude that retrieval failed or that data is missing.
How to verify the setup
- Input the four material texts and the draft sentence from the hypothetical sample. Compare the three assertions in the draft sentence — "executed at 10% off," "double points took effect," and "has completed the refund" — against their respective source fields. Satisfied condition: the first two assertions find support in the price tag and loyalty rules texts with the same subject, same period, and same field; the third assertion is marked unsupported because the refund slip status reads "pending review."
- Input the vector model identifiers for the promotional price tag base and the loyalty points rules base from the hypothetical sample. Compare whether the two identifiers are the same. Satisfied condition: identical identifiers allow joint association to be recorded; different identifiers exclude the non-conforming base and record the actual observation, without performing cross-model retrieval to compare missed recalls.
- Input an explicit paper-based empty array [] as the search output. Compare whether the downstream answer step design accepts an empty result. Satisfied condition: the downstream design can handle the empty array and output a no-match notice, without claiming that a fictional rule retrieval necessarily returns [].
This feature has verified public technical facts (for identifying relevant content in documents; the prose above is written independently and does not require copying a fixed five-row table): [["Associated knowledge bases", "The knowledge base search scope first checks the selected knowledge bases and material versions; only the same vector model allows joint association."], ["Citation content output", "The search node outputs a citation array whose length may be zero; an empty array does not mean the workflow has reported an error or terminated."], ["Citation input handoff", "Check whether the retrieval output is passed to the answer step, and verify each answer against the cited source text."], ["Citation display check", "Use a traceable sample to verify source display; do not substitute thinking process or answer fluency for source evidence."], ["Material timeliness check", "Test sources with current and historical document versions; whether a citation is adopted still requires review against the valid version."]]