Near-expiry markdowns: citation tracing for procurement

Review citation tracing for procurement in Near-expiry markdowns. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

Near-expiry markdown handling applies to shelf inventory batches, not to order-by-order weighing and sorting. The operational records are shelf-life data from the point-of-sale system, store inspection notes, enterprise resource planning inventory, and quality inspection records. For the inventory procurement and supplier records task, the documents to be checked are the near-expiry list, the markdown approval, the damage report, and the promotional price tag. Citation tracing here means that any batch number, supplier, discount percentage, or status appearing in an answer must be traceable to the original position in those documents. Data updates occur daily or per shift.

Constraints at the citation tracing step

The knowledge base search node selects one or more knowledge bases that share the same vector model and outputs an array of cited content whose length may be zero. When the array is empty, the output path still runs. Search results must be passed to the answering step, and each answer statement must be checked against the cited original text. The minimum relevance value is a filtering threshold that takes effect only in semantic search or when result reranking is used. That number does not perform permission or entity isolation, and retrieval itself does not constitute authorization.

Hypothetical paper exercise sample: the following are paper materials and do not come from any real entity. Material one (near-expiry list): "Store A, batch A-0417, leafy vegetables, expiry date 2025-05-02, current status is pending markdown." Material two (markdown approval): "Batch A-0417, requested markdown 30%, approval status is pending review, applicant: Store A, date 2025-04-28." Material three (damage report): "Batch A-0417, damage quantity 2 kg, reason: wilted leaves, date 2025-04-29." Draft sentence to be assessed: "Batch A-0417 has received approval for a 30% markdown and may proceed to markdown sale." The field needed for judgment is approval status. Material two states "pending review," which is mutually exclusive with the draft sentence's "has received approval" on the same subject, same period, and same field. That draft sentence is not supported.

Configuration and implementation checks

Configuration or check itemValue or action on this pageBasis and boundary
Linked knowledge basesHypothetical exercise starting point: 1, containing the three materials aboveOnly knowledge bases with the same vector model may be linked together; for different models, verify identifiers and exclude them
Search modeHybrid search with result rerankingFull-text search suits exact batch and number queries; reranking requires an available model and must be enabled by configuration
Minimum relevanceHypothetical exercise starting point: 0.4, used to compare missed-recall behavior against 0.2 and 0.6This value takes effect only in semantic search or with result reranking and does not act as a permission filter
Citation limitMeasured in tokens; hypothetical exercise starting point: 2000Controlled per document in tokens, not converted into a fixed number of items
Cited content outputAccepts an array, which may be emptyCheck downstream design against the paper input []
Citation display checkCompare the answer line by line against the three original materialsDo not substitute answer fluency for source evidence

Three common errors

  1. Treating approval status "pending review" as "has received approval." Check: take the approval status field from material two and compare it with "has received approval" in the draft sentence; when the two are mutually exclusive, that sentence is not supported.
  2. Treating a retrieval hit as authorized execution. Check: take the batch and markdown percentage cited in the answer and compare them with the original markdown approval; returned retrieval content does not change approval status, and executing a markdown requires a separate authorization boundary.
  3. Merging damage quantity and markdown quantity into one field. Check: take the damage quantity of 2 kg from material three and compare it with the markdown scope stated in the draft sentence; when the fields differ, they cannot support each other.

How to verify the setup

  1. Input the original text of material two and the draft sentence to be assessed, then compare "pending review" with "has received approval." When mutually exclusive values appear for the same subject, same period, and same field, the sentence is judged unsupported, and the evidence gap is recorded.
  2. Input the paper empty array []. Compare whether the downstream node continues to run with an empty citation. The condition is met when the output path runs, reports no error, and fabricates no source; an empty result is one possible outcome and is not assured for every query.
  3. Input material one and material three, then compare batch number, date, and status fields. The check condition is met when batch numbers match, dates fall within the valid range, and each status field is traceable; missing fields are recorded as pending verification.