What the documents look like
Retail checkout and loyalty materials centre on price and membership entitlements. They do not deal with short-shelf-life stock. Everyday documents include promotional price tags, membership points rules, end-of-day checkout reports and abnormal refund slips. Sources include the POS system, the membership CRM, payment platforms and the price tag system. These materials update daily or per shift. Errors can produce price complaints, cash variances and disputes over membership entitlements.
Promotional price tags and membership points rules are rule-type materials. Checkout reports and abnormal refund slips are process-type materials. The two types serve different retrieval purposes. Rule-type materials answer whether a given promotion is in effect and how points are calculated for a membership tier. Process-type materials answer the variance amount for a shift and the current state of a refund. Operations authorisation and handover management requires retrieval results to present only what the materials record. A retrieval hit does not function as an authorisation basis.
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 with different vector models cannot be associated together. The vector model identifier should be checked first, and non-matching bases excluded. Actual observations are recorded. Cross-model retrieval is not executed to compare missed recalls. Search modes include semantic search, full-text search and hybrid search. Hybrid search merges multiple result paths through RRF. RRF is a merge process, not a standalone switch. The citation limit uses tokens as its unit. Minimum relevance takes effect only under semantic search or when result reranking is used. Question optimisation performs coreference resolution and question expansion, and it makes an additional model call. Citation content is output as an array whose length can be zero, and the downstream path still runs when nothing is retrieved. Retrieval and similarity scores do not perform permission or entity isolation.
Hypothetical drill sample: a store price tag system record reads "1 March 2025 to 31 March 2025, member price 9.9 yuan, non-member price 12.5 yuan, applies to directly operated stores in the East China region." A membership CRM rule reads "points accrue at 1 point per 1 yuan actually paid; corresponding points are deducted on refund." A checkout report reads "12 March 2025, shift A, checkout variance -37.60 yuan, pending handover supervisor review." A draft sentence awaiting assessment reads "the checkout variance for shift A on 12 March at this store has been reviewed and approved by the supervisor." The material shows only "pending handover supervisor review" and does not show that review was completed. The state asserted in the draft sentence is not supported.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this trial | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the promotion rules base using the same vector model | Check the vector model identifier and exclude bases whose identifier does not match |
| Search mode | Test semantic search and full-text search separately first | Compare the two result paths on the same test question, then decide whether to use hybrid search |
| Citation limit | Set to cover one complete points rule | Compare citation completeness using actual rules of different lengths |
| Minimum relevance | Annotate relevant and irrelevant samples before choosing a value | Compare the distribution of annotated samples on either side of the threshold; no fixed value is preset |
| Question optimisation | Enable only for multi-turn follow-ups or short questions | Compare the original question with the optimised question and check whether the subject was altered |
| Result reranking | Enable according to settings when a rerank model is available | Compare candidate ordering before and after reranking and observe relevance score changes |
Values in the table are hypothetical starting points for this trial and are not official defaults. Image retrieval requires corresponding image input and model capability. Reranking requires a model and must be enabled according to settings. Optional paths shown in the overall document flow diagram do not mean every request executes all of them.
Three common errors
- Treating a retrieval hit as an authorisation basis. The input is a checkout report retrieval result. Compare the retrieved content with the authorisation state in the handover request. If the retrieved content records only "pending review," it cannot support a determination of "reviewed and approved." Retrieval and similarity do not perform permission isolation.
- Associating knowledge bases that use different vector models. The input is the vector model identifiers of two knowledge bases. Compare whether the two identifiers match. A base whose identifier does not match should be excluded and the observation recorded. Cross-model retrieval is not executed to infer missed recalls or redundant results.
- Treating RRF after hybrid search as a standalone switch that can be turned on. The input is the search mode setting. Compare the configuration items for semantic search, full-text search and hybrid search. RRF appears only when hybrid search merges multiple result paths. There is no separate enable item.
How to verify the setup
- For the draft sentence in the hypothetical sample, "the checkout variance for shift A on 12 March at this store has been reviewed and approved by the supervisor," input the checkout report text "pending handover supervisor review." Compare what the two texts record for the same shift and the same field. If the source text shows only a pending-review state, the draft sentence is not supported and should be recorded as pending verification.
- For the price tag text and the membership points rule text in the hypothetical sample, input the question "March member price and points accrual rules." Compare the citation content returned by semantic search and by full-text search. Observe whether the two result paths cover the price tag validity period and the points calculation field. A first-hit match or full recall is not guaranteed.
- For empty-result continuation, input the explicit paper input []. Compare whether the downstream path still runs. Observe whether the downstream design handles an empty citation array as empty references when its length is zero. Do not claim that a fictional rule retrieval necessarily returns [].
This page addresses one scope only: retail checkout and loyalty / operations authorisation and handover management / knowledge base retrieval and recall. Which materials, operations and requests are in what authorisation state during an operations handover cannot be treated as authorised execution on the basis of retrieval or a form.