What the documents look like
Returned-goods inspection handles reverse parcels and assigns a condition grade to each item. It does not handle outbound picking or packing for new orders. The materials come from the WMS returns module, platform after-sales records, carrier return shipments, and the inspection bench. The daily documents are the return warehouse receipt, inspection photographs, the grade determination table, and the resale listing form. Updates occur per item and per day, so the same product may carry different grade conclusions on different dates. Retrieval must therefore lock its scope by document date and document number rather than by product name alone.
The material forms span structured fields, photographs, and determination tables. A retrieval path needs to cover both textual descriptions and image indexes. Whether an image index was created at the ingestion stage directly determines whether that image can later be retrieved. This is a property of the stored data, not of the query alone.
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. Search modes are semantic, full-text, and hybrid. Full-text retrieval suits precise queries such as numbers and model codes. Semantic retrieval suits natural-language questions. The minimum relevance value takes effect only under semantic retrieval or when result reranking is used. The citation cap is measured in tokens, not in a fixed item count. Image retrieval depends on image description indexes or image vector indexes generated at ingestion, together with a vision model or a vector model that supports image input. Retrieval results do not perform authorization checks. Whether an item may be resold or refunded is decided by downstream business review.
Hypothetical drill sample: the source material reads, "Return warehouse receipt R-2024-0517, product A-type filter element, returned quantity 12 pieces, inspection bench determined surface scratches, grade determination table marked Grade B, resale listing form status pending review." The draft sentence to be judged is, "This batch of A-type filter elements has completed resale listing." The comparison point is that the resale listing form status is pending review, while the draft sentence states completion. The draft sentence is unsupported by the material.
Configuration and implementation checks
| Configuration item | Hypothetical starting point for this page | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only knowledge bases holding returned-goods inspection material that share one vector model | Verify that the vector model identifiers of each base match before associating; exclude those that do not match |
| Search mode | Hybrid retrieval | Run the same question through semantic and full-text separately, then compare hit differences between code-type and natural-language queries |
| Citation cap | Hypothetical starting point of 2000 tokens | Test with receipts and determination tables of differing lengths to see whether content is truncated, then adjust up or down |
| Minimum relevance | Hypothetical starting point of 0.4 | Label relevant and irrelevant samples first, then observe whether key rows of the determination table survive filtering |
| Question optimization | Enabled | Compare the original question with the optimized question, and check whether the subject and document number were altered incorrectly |
| Result reranking | Enabled | When a reranking model is available, enable it per settings and compare candidate ordering before and after reranking |
The values above are hypothetical trial starting points for this page and are not official defaults.
Three common errors
- Treating a retrieval hit as resale authorization. In the sample, the resale listing form is pending review. Retrieving that document does not support a conclusion that listing is complete, and the pending-verification state must be recorded.
- Forcing an association across knowledge bases with different vector models. When two bases carry different vector model identifiers, verify the identifiers and exclude the mismatched one. Do not run a cross-model retrieval to compare missed recalls.
- Treating minimum relevance as a permission or entity filter. That value only filters low-relevance results. It does not exclude unrelated businesses or unauthorized records. Authorization requires a downstream implementation.
How to verify the setup
- Input the sample source text and the draft sentence, compare the two status fields "pending review" and "completed," and observe whether the system retains the pending-verification judgment rather than outputting a completion conclusion.
- Input the same question through semantic retrieval and full-text retrieval separately, compare the hit differences between code-type queries and natural-language queries, and observe whether hybrid retrieval fuses the two result paths.
- Input an explicit paper-form empty result
[]and check whether that output path still completes. Do not claim that a fictional rule retrieval necessarily returns an empty array.
The verification steps above describe expected observations. No test, deployment, approval, refund, or other business action is claimed to have occurred. Where the material does not state a field, that field remains unknown.