What the documents look like
Returned-goods inspections at the receiving warehouse work from four document families: the return inbound receipt, inspection photographs, the grade determination table, and the resale listing sheet. These arrive from the WMS returns module, platform after-sales, carrier returns, and the inspection bench. The people handling them are after-sales warehouse staff, inspectors, and inventory administrators. Unlike in-warehouse picking and packing, this work handles reverse parcels and product grades rather than new-order outbound. The stated consequences of error are resale of defective goods, refund disputes, and inventory contamination. The stated update frequency is per item and daily.
For procurement and supplier records, retrieval has to line up the return receipt number, the supplier, the product condition, and the grade determination outcome inside one set of cited content. Whether those fields are recalled consistently depends on how the vector model and the indexes are configured. The knowledge bases that are linked together for retrieval must use the same vector model, so the first check is the actual model identifier and the input type it accepts. A single source data block can carry several indexes, and the index content is checked separately from the source content. Index content is built from complete and focused information, and retrieval coverage is weighed against index length using real samples. Image retrieval depends on whether an image description or image vector was generated at ingestion time, so checking the query side alone is not enough. After any model or index change, the same set of questions is compared against the cited source text; a vector score on its own does not establish an outcome. Semantic recall and relevance scores do not carry permission or entity isolation, and the same-vector-model requirement for linked knowledge bases does not extend to business-registration sites or other external systems.
Constraints at the vector indexes step
A hypothetical paper sample is used here for checking. Source text one reads: "Return inbound receipt R-2026-0312, supplier Hengyuan Components, item filter element assembly, inspection conclusion surface scratches, grade determination B-grade resaleable, resale listing sheet generated, date 12 March 2026." Source text two reads: "Grade determination table G-2026-0312, corresponding return receipt R-2026-0312, determination C-grade scrap, date 12 March 2026." The draft sentence under review reads: "The item from Hengyuan Components under return receipt R-2026-0312 has been resold and listed at B grade."
The comparison is made on the grade determination field under the same return receipt number, the same supplier, and the same date. B grade and C grade are mutually exclusive values of that one field. The resale listing conclusion in the draft sentence is not supported by the source text, so the sentence is marked as pending verification. The inspection photograph path is separate: image retrieval requires a corresponding image input and model capability, and it depends on whether an index was built when the photograph was ingested.
Configuration and implementation checks
| Configuration item | Content for this page | How to check |
|---|---|---|
| Vector model | Linked knowledge bases must use the same vector model | Check the actual model identifier and input type, and exclude bases that do not match |
| Index and source data | One source data block can carry several indexes | Check index content and source content separately, confirming the return receipt number and supplier fields are complete |
| Index content | Indexes are built from complete and focused information | Use the return inbound receipt and grade determination table as samples, weighing retrieval coverage against index length |
| Image index | Image retrieval depends on whether an image description or image vector was generated at ingestion | Check whether the inspection photograph has a corresponding index at ingestion; confirm query-side capability separately |
| Search mode | Semantic search, full-text search, or hybrid search | Use full-text or hybrid search for identifier-type queries, semantic or hybrid for natural-language queries |
| Result review | After a model or index change, compare the same set of questions against the cited source text | Record differences between cited source text and the draft sentence; a vector score alone does not establish an outcome |
A hypothetical trial starting point for this paper exercise: set the minimum relevance to 0.3. The reason for this choice is that it sits in the middle of the 0-1 range, which makes it easier to observe how cited entries change before and after filtering. The comparison method is to run the same set of return receipt number questions before and after the adjustment, view the cited source text in each case, and record which entries were filtered out. This value is a starting point for this paper exercise only and is not an official default.
Three common errors
- Linking knowledge bases that use different vector models. Input: two knowledge bases with different model identifiers. Compare the model identifier fields. The observable result is that the non-matching base is excluded and the actual observation is recorded; cross-model retrieval is not run to compare missed recall.
- Treating a retrieval hit as an authorization basis. Input: one return inbound receipt returned by retrieval. Compare the resale listing conclusion in the draft sentence against the grade determination field in the source text. The observable result is that when the source text does not state a completed resale status, the sentence is marked pending verification.
- Treating image retrieval as a default active path. Input: an inspection photograph with no image index generated. Compare the ingestion index record against the query-side input. The observable result is that image retrieval requires a corresponding image input and model capability, and that path does not take effect when no index was built.
How to verify the setup
- Input the two source texts and the draft sentence from the hypothetical sample. Compare the grade determination field values under the same return receipt number. Observe whether B grade and C grade form mutually exclusive values of that one field. When that condition holds, the draft sentence is marked pending verification.
- Input return receipt number R-2026-0312 as the search term. Compare whether the cited source text contains supplier Hengyuan Components and date 12 March 2026. Observe whether the cited entries cover those fields. A first-position hit or full recall is not guaranteed.
- Input an empty-result case, using an explicit paper input of an empty array to check the downstream design. Observe whether the output path runs through. No claim is made that a fictional rule search must return an empty array.
This page addresses only returned-goods receiving inspection, procurement and supplier records, and the vector model and index step. During operational handover, the authorization state of materials, operations, and requests is a separate matter; retrieval or a form does not constitute authorization to execute.