What the documents look like
Promotion enrollment and price management turn on price, stock and campaign rhythm. The material set for this category includes the campaign enrollment form, the promotional price approval sheet, the stock lock table and the campaign review table. The enrollment form records the enrolling entity, the campaign period and the enrollment status. The approval sheet records the proposed selling price, the approval node and the approval conclusion. The stock lock table records the locked quantity and the lock scope. The review table records actual transactions and exception notes. These materials are updated weekly or per campaign and feed operating report consolidation and exception review.
Retrieval for this task means recalling the fields that correspond to the current review question from those documents, so that a person can compare them afterwards. Price approval status, stock lock status and enrollment status are three separate facts. Retrieval output presents document text; it does not stand in for approval or authorization.
Constraints at the retrieval step
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. At implementation time, verify the vector model identifier first, exclude those that do not match, and record what was actually observed. Do not run a cross-model search in order to compare missed recalls.
Search mode can be semantic retrieval, full-text retrieval or hybrid retrieval. Semantic retrieval depends on vector distance. Full-text retrieval depends on keyword hits and suits numbers, model designations and proper nouns. Hybrid retrieval merges the two paths through RRF; RRF is a fusion process, not a separate switch that can be turned on. The reference cap controls, in tokens, the maximum amount of content cited per search. The minimum relevance value is a 0-1 filter threshold that takes effect only under semantic retrieval or when result reranking is used, and it does not perform permission or entity isolation. Question optimization performs coreference resolution and question expansion and adds one model call. Reference content is returned as an array whose length can be 0, and the downstream path still runs on an empty result. Retrieval is recall, not authorization, and not approval.
Hypothetical paper exercise sample: a brand sales planner submits a retrieval request for a campaign review. The source text reads: "Campaign enrollment form: enrolling entity is a store in the East China region; campaign period is 1 June 2025 to 7 June 2025; enrollment status is submitted." "Promotional price approval sheet: proposed selling price is 199 yuan; approval status is pending review; the approver has not signed." The draft sentence to be judged reads: "The store's promotional price has been approved and enrollment can proceed." The comparison puts the draft sentence against the approval sheet source text: the source approval status is pending review, and no approval-passed or signed record appears, so the draft sentence's claim of approval is not supported. Pending review does not constitute approval passed, and it does not demonstrate the absence of operating permission.
Configuration and implementation checks
The configuration below addresses the comparison needs of the sample and lists only settings already described in the official material, or implementation check items. The values are starting points for this hypothetical exercise, not official defaults.
| Configuration item | Hypothetical starting point for this exercise | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only knowledge bases using the same vector model as the sample material | Verify the vector model identifier first and exclude non-matching ones; do not run cross-model retrieval |
| Search mode | Hybrid retrieval | The sample contains the keyword "pending review" plus a natural-language question; compare semantic, full-text and hybrid recall paths |
| Reference cap | Hypothetical starting point of 2000 tokens for this exercise | The sample material is short; compare whether cited content stays complete at 1000 and at 2000 |
| Minimum relevance | Hypothetical starting point of 0.3 for this exercise | Label relevant and irrelevant fragments in the sample first, then compare filtering results at 0.2 and at 0.4 |
| Question optimization | Enabled for this hypothetical exercise | The sample question contains a reference; compare the original and optimized question to see whether the subject was altered |
| Result reranking | Enabled for this hypothetical exercise | When candidates are numerous, compare ordering before and after reranking and confirm the rerank score is used for filtering |
Three common errors
- Treating pending review as approval passed. In the sample, the approval sheet source text states the approval status is pending review, while the draft sentence says the price has been approved. Input the approval sheet source text and the draft sentence, compare the value of the approval status field, and observe that the source says pending review while the draft sentence claims approval; that observation identifies the error.
- Treating retrieval recall as authorized execution. In the sample, the retrieval request may recall the enrollment form and the approval sheet, but the recalled result does not constitute approval or enrollment authorization. Input the retrieval output array and the draft sentence, compare whether the draft sentence presents recalled content as executable, and observe that the draft sentence crosses the approval status to call it executable; that observation identifies the error.
- Associating knowledge bases built on different vector models. The sample requires associating knowledge bases with the same vector model. Input the vector model identifiers of the knowledge bases, compare whether the identifiers match, and observe that different identifiers were associated together; that observation identifies the error. Do not invent runtime results of such an association.
How to verify the setup
- For the sample draft sentence "The store's promotional price has been approved and enrollment can proceed," input the approval sheet source text and the draft sentence, compare the approval status field, and observe that the source states pending review with no signing record, so the draft sentence's claim of approval is not supported. Judge it as not supported.
- For the sample retrieval request, input the question and the associated knowledge bases, compare the semantic, full-text and hybrid recall results, and observe whether hybrid retrieval merges the two paths through RRF and whether the reference content array contains the sample material source text. First-place hits and complete recall are not guaranteed.
- For empty-result handling, input an explicit paper input of an empty array, compare whether the downstream design handles reference content of length 0, and observe whether the path runs through and record the gap. Do not claim that a fabricated rule search necessarily returns an empty array.