What the documents look like
Returned-goods inspections handle reverse parcels and item grading. The work does not cover outbound picking or new-order packing. Data comes from the WMS returns module, platform after-sales records, carrier returns, and the inspection bench. Everyday documents include the return warehouse receipt, inspection photographs, the grade determination table, and the resale listing sheet. Records update per item and per day.
For store service and internal training, these materials answer a narrow set of questions: what state a returned item currently holds, which evidence supports a grade, and who decides resale or scrap within which boundary. Retrieval has to surface the receipt number, the item condition shown in the photographs, the values in the grade table, and the authorization boundary of the resale listing sheet, so that training staff can check them against each other.
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. The node outputs cited content as an array, and the array length can be zero. When nothing is retrieved, the output path still runs.
Minimum relevance is a 0–1 filter threshold. It takes effect only under semantic retrieval or when result reranking is used. It filters low-relevance results and does not act as a permission or entity boundary. Hybrid retrieval merges multiple result paths through RRF; RRF is a fusion process, not a switch that can be turned on. Question optimization performs coreference resolution and question expansion, and it adds one model call. Retrieval by itself is not authorization. Forms collect values; business verification, approval, database writes, and external synchronization each need their own downstream implementation.
Hypothetical drill sample: the handover material reads, "Return warehouse receipt RT-0001, item is a Type A insulated cup, inspection photographs show scratches on the body, the grade determination table lists this item as Grade B, and no resale listing sheet has been generated yet." The draft sentence to be judged reads, "This returned item has completed resale listing approval." Comparing the sentence against the source: the source states only that no resale listing sheet has been generated. It does not record an approval state. The draft sentence is not supported.
Configuration and implementation checks
| Configuration item | Value or check for this page | Basis and boundary |
|---|---|---|
| Associated knowledge bases | Select the bases holding the return warehouse receipt, inspection photographs, grade determination table, and resale listing sheet that use the same vector model | The node can associate one or more knowledge bases sharing the same vector model |
| Search mode | Use full-text retrieval for identifiers such as RT-0001, semantic retrieval for item-condition wording, and observe RRF fusion when both run | Hybrid retrieval merges through RRF; it is not a switch |
| Citation cap | Hypothetical trial starting value of 800 tokens, chosen because the sample carries an identifier, a photograph description, and a grade field that must stay complete | Unit is token; compare against actual material length before adjusting |
| Minimum relevance | Hypothetical trial starting value of 0.4, set after labeling relevant and irrelevant samples for comparison | Takes effect only under semantic retrieval or result reranking |
| Question optimization | Enable for multi-turn follow-ups such as "the grade of that item," then check whether the optimized question still carries RT-0001 as its subject | Used for coreference resolution and question expansion |
| Result reranking | When a ReRank model is available and enabled by configuration, rerank candidates and compare the 0–1 score | Reranking requires the model and must be enabled by configuration |
Three common errors
- Treating minimum relevance as a permission or entity filter. The input is a retrieval request plus a threshold setting. The comparison is whether returned citations still contain records from other stores or unauthorized records. The observable result is that changing the threshold does not change record ownership. The correct step is to verify access scope separately.
- Reading "no resale listing sheet has been generated yet" as approval rejected or approval pending. The input is the RT-0001 source material. The comparison is whether the source records an approval state. The observable result is that the source states only that no sheet has been generated; the approval state remains unrecorded.
- Associating knowledge bases that use different vector models. The input is the vector model identifier for each base. The comparison is whether the identifiers match. The observable result is that a mismatched base is excluded, and no cross-model retrieval is run to compare missed recalls.
How to verify the setup
- Input the hypothetical source material and the draft sentence "This returned item has completed resale listing approval." Compare whether the source supports the draft. The condition is met when the sentence is judged unsupported and the evidence gap is recorded.
- Input the identifier RT-0001 into full-text retrieval. Compare whether the returned citations contain the fields tied to that identifier. The condition is met when the cited content covers the identifier and the grade field; a miss is recorded as not recalled, without inferring a cause.
- Input an explicit paper-level empty result
[]. Compare whether the downstream path still runs. The condition is met when the output path runs with a citation length of zero. Do not claim that a fictional rule-based query necessarily returns[].