What the documents look like
For product questions in General import declarations, this page examines FastGPT hybrid retrieval. Compare the material received with the sources used in the answer. The examples are hypothetical.
Constraints at the hybrid retrieval step
The proposed import invoice states “standard-model dark-blue goods.” Mark the original wording for model and colour, then inspect whether fused candidates keep those details attached to the same product. Retain the requested product in the complete question and observe whether a similar model passage moves ahead during reranking. The answer should trace its description to the invoice, without extending that sentence into performance or market-admission claims.
Configuration and implementation checks
| Item or implementation check | Input or applicable condition | Traceable check |
|---|---|---|
| Fused candidates | Complete standard-model dark-blue description | Keep model and colour attached to the same item. |
| Complete question | One hundred invoice pieces versus ninety-nine packing-list pieces | Retain total quantities and both file identities. |
| Minimum relevance | Inspect filtering during semantic retrieval or reranking | Check units and batch scope after filtering. |
| Citation token limit | Original quantities and their total-piece meaning | Do not silently standardize the two values. |
| Missing-source retest | Invoice supplies no market-admission evidence | Add the applicable source and inspect its citation. |
Documented conditions
| Documented item / implementation check | Condition |
|---|---|
Hybrid retrieval | Vector and full-text results use RRF fusion. |
| Reranking | Reranking reorders candidates using the full question. |
Minimum relevance | Filtering applies to semantic search or reranking. |
Citation limit | Citation content is limited by tokens. |
Question optimization | Compare subjects and restrictions after optimization. |
Three common errors
One hundred pieces and ninety-nine pieces occur in the proposed commercial invoice and packing list respectively. A quantity check needs the product and total-piece wording from both sources, beyond the side most similar to the question. Preserve order and provenance, then inspect whether citations retain the meaning of total quantity. Batch and version correspondence still require confirmation. Different values raise a review point; ranking cannot establish a short shipment or a customs-declaration error. Retain both quantity statements with their units, without silently normalizing ninety-nine into one hundred. Even a short final response should leave both retrieved values traceable. Similar surrounding product text cannot erase a numerical difference between the supplied sources.
How to verify the setup
The available invoice contains no evidence about the specified market-admission standard. Product name, colour and quantity cannot replace it. Use this missing-source question to inspect whether reranking encourages a general product description to be presented as conformity evidence. After obtaining the relevant standard and applicable material, repeat the question and inspect citation of the added evidence. Validate minimum relevance only under its documented conditions, without converting rank or score into an established admission outcome.
| Material context | Scope |
|---|---|
| Document types | Import Declaration Form, Commercial Invoice, Packing List, Bill of Lading and Contract |
| Candidate sources | International Trade Single Window, Customs Tariff, Freight Forwarder Arrival Notice and Foreign Exchange Payment Slip |
| Technical baseline | FastGPT v4.17.2 |