What the documents look like
Export tax rebate filing happens after goods leave the country, and the work centers on matching tax vouchers. The documents involved are the export tax exemption and rebate filing form, the export customs declaration, purchase invoices, and foreign exchange receipt vouchers. Finance staff, tax accountants, and documentation clerks at trading companies handle these materials monthly or batch by batch. The central relationship in inventory procurement and supplier records is the correspondence between purchase invoices and customs declarations, together with whether supplier identity, goods name, quantity, and amount fields align across documents. At the retrieval step, the question is whether a given document contains source text that supports a particular field value. This page addresses the knowledge base retrieval and recall step for that material set. It does not describe batch synchronization mechanisms.
Constraints at the retrieval step
The knowledge base search node can be linked to one or more knowledge bases that use the same vector model. Knowledge bases built on different vector models cannot be linked together. Retrieval recalls candidates through semantic search, full-text search, or hybrid search; hybrid search merges multiple result paths through RRF. Result reranking requires a rerank model and must be enabled according to configuration. Image retrieval requires corresponding image input and model capability. The reference limit is measured in tokens. The minimum relevance threshold takes effect only under semantic search or when result reranking is used. A similarity score expresses the degree of content relevance and does not perform permission or entity isolation; a retrieval result does not constitute authorization.
Hypothetical paper exercise sample: a trading company loads a batch of supplier A records into a knowledge base. The source text reads: "Supplier A issued a purchase invoice on the hypothetical date 10 March 2025, goods name cotton men's shirts, quantity 1,200 pieces, corresponding to customs declaration number hypothetical identifier BG-2025-0310." A draft sentence awaits judgment: "Supplier A's batch has completed export tax rebate filing and been approved." The retrieval request is: within the inventory procurement and supplier records knowledge base, find source text that supports this draft sentence.
Configuration and implementation checks
The table below is tied to the sample above and lists only documented settings or implementation checks relevant to this feature and primary task.
| Configuration item | Hypothetical starting point for this exercise | Reason for the choice and comparison object |
|---|---|---|
| Linked knowledge bases | Link only the inventory procurement and supplier records knowledge base | First check each knowledge base's vector model identifier and exclude those with different models; compare against the material scope |
| Search mode | Start with full-text search for identifier-type questions | The customs declaration number is an exact query; compare against semantic and hybrid mode results |
| Reference limit | Start at a token amount sufficient to cover one complete invoice record | Test with actual materials of varying length to see whether content stays complete, then adjust |
| Minimum relevance | Start at a middle value, effective only under semantic search or reranking | First label relevant and irrelevant samples, then compare results at different thresholds |
| Question optimization | Start disabled, testing with complete questions | Compare the original question with the optimized question and check whether the subject was altered incorrectly |
| Result reranking | Not enabled for now, pending confirmation of a model | Reranking requires a model and must be enabled according to configuration; compare against unreranked results |
Three common errors
- Treating a similarity score as a permission or entity filter. The input is the retrieval request; the comparison object is the returned content against the authorized scope. If records without access rights are returned, the filtering is not implemented. A similarity setting does not exclude unrelated businesses or unauthorized records.
- Treating a retrieval hit as a business fact. The input is the draft sentence "has completed export tax rebate filing and been approved"; the comparison object is the sample source text. The source text states only invoice and customs declaration information and does not state filing or approval status, so the draft sentence remains unsupported.
- Forcing knowledge bases with different vector models into a joint link. The input is the vector model identifiers of multiple knowledge bases; the comparison object is whether the identifiers match. If they do not match, they should be excluded. Cross-model retrieval is not executed to compare missed recalls.
How to verify the setup
- Input the sample source text and the draft sentence, then compare whether the source text contains the fields "filing" or "approved." If the source text has no such wording, judge the draft sentence unsupported and preserve the evidence gap.
- Input the customs declaration number BG-2025-0310 and compare the results returned by full-text search and semantic search. If full-text search can locate the record containing that number, the exact-query expectation is met; a first-position hit is not guaranteed.
- Input an empty reference array [] and compare whether the downstream node accepts the empty result as designed. If the chain still proceeds, the empty-result handling expectation is met. No claim is made that a fabricated rule retrieval must return [].
This page addresses only export tax rebate filing / inventory procurement and supplier records / knowledge base retrieval and recall. At operational handover, the authorization status of materials, actions, and requests cannot be inferred from retrieval or from a form. Retrieval is not authorization.