What the documents look like
Import declaration work in general trade centres on customs value, freight and insurance, and duty payment. The customs declaration, commercial invoice, packing list, bill of lading and contract have to line up item by item on transaction terms, amounts and quantities. Materials are updated per shipment, and the same consignment can appear in several different phrasings across the declaration, the invoice and the contract. Retrieval therefore works over these instrument texts and has to bring the parties, object, period and fields of one transaction together before any comparison is meaningful. Arrival notices from the freight forwarder and foreign exchange payment documents supply arrival and payment status; they belong to related stages and cannot on their own support a conclusion that transaction terms are confirmed.
Constraints at the retrieval step
The knowledge base search node can associate one or more knowledge bases that use the same vector model. Bases built on different vector models cannot be associated together; the identifier is checked first and non-matching bases are excluded. Quoted content is output as an array, and its length can be 0, meaning the output path still runs when nothing is retrieved. The minimum relevance value takes effect only under semantic retrieval or when result re-ranking is used; it is a filter threshold and does not perform permission or entity isolation. Hybrid retrieval merges multiple result paths through RRF, which is a fusion process and has no on/off switch. Image retrieval requires the corresponding image input and model capability, and re-ranking requires a model that is enabled according to its settings. The optional paths shown in the documentation flow do not all execute on every request.
Hypothetical drill sample: an importer's consignment, where the contract states "transaction term CIF, total price USD 120,000, shipment period March 2025". The commercial invoice states "CIF, USD 120,000, shipped March 2025". The draft sentence to be assessed is "The transaction term, total price and shipment period in the contract and the invoice for this consignment are consistent, so the comparison conclusion can be submitted." This sample serves only to illustrate the checking method.
Configuration and implementation checks
The table below is built around this page's sample and lists documented settings together with implementation checks. The numeric values are hypothetical starting points for this trial, used for comparison against another set of values; they are not official defaults.
| Configuration item | Hypothetical starting point for this trial | Reason for the choice and comparison method |
|---|---|---|
| Associated knowledge bases | Associate only the contract and invoice bases that use the same vector model | Check the vector model identifier first, exclude non-matching bases, then associate |
| Search mode | Run semantic retrieval first, then run full-text retrieval | Use the same question and compare the retrieved content on the two paths |
| Citation limit | Hypothetical starting point of 2000 tokens | Test content completeness with materials of different lengths, then adjust |
| Minimum relevance | Hypothetical starting point of 0.5 | Label relevant and irrelevant samples first, then compare results as the threshold changes |
| Question optimisation | Enabled for follow-up questions that contain references | Compare the original question with the optimised question and check whether the subject was altered |
| Result re-ranking | Enabled when candidates are numerous | Compare ordering before and after re-ranking and confirm the model is enabled per its settings |
Three common errors
- Treating minimum relevance as a permission or entity filter. Input a query unrelated to the contract and compare whether the returned content can still include material belonging to another party. The observation point is that relevance affects ordering and filtering only and does not isolate business entities.
- Forcing knowledge bases with different vector models into one association. Input two bases whose vector model identifiers differ and compare whether the association action is permitted. The observation point is that the identifier is checked and non-matching bases are excluded; no cross-model retrieval is run to compare missed recalls.
- Treating a pending review request as granted authorisation. Input a draft sentence stating "the comparison conclusion has been submitted" and compare whether the source text records approval or authorisation. The observation point is that a pending review neither demonstrates a lack of operating permission nor equals approval.
How to verify the setup
- Check the draft sentence against the sample source text. Input the contract and invoice source text together with the draft sentence, and compare whether the transaction term, total price and shipment period share the same party, the same period and the same field with consistent values. All three must be consistent for the condition to hold; if any item is not recorded, log it as unsupported.
- Check downstream handling with a paper empty result. Input an explicit empty citation array [] and compare whether the downstream design continues to run on an empty result. Observe whether the path still executes with no content; do not claim that a fictional rule retrieval necessarily returns [].
- Check question optimisation with a follow-up containing a reference. Input "what about the transaction term for that consignment" together with the preceding text, and compare whether the optimised question keeps the correct subject and object. If the subject is altered or the object drifts, the condition does not hold; when optimisation is not enabled, record that this path was not executed.
This page covers only general trade import declaration / contract and transaction material comparison / knowledge base retrieval and recall. In operational handover, the authorisation status of materials, actions and requests must not be inferred from retrieval or from a form. The same vector model requirement applies to the associated knowledge bases and does not extend to business registration sites or other external systems. Any system error or automatic business action not stated in the source text must not be asserted.