Import and export permits: retrieval for product questions

Review retrieval for product questions in Import and export permits. Check document scope, failure paths and source evidence through controlled tests.

What the documents look like

Permit handling for import and export takes place before declaration, and it determines whether a shipment may lawfully cross the border. The materials handled in this task include automatic import licences, dual-use item licences, contracts and product descriptions. These materials come from the competent commerce authority system, customs commodity codes, supplier materials and customer statements of use. They are updated irregularly, on a per-shipment basis.

Knowledge base retrieval and recall turns these materials into quotable reference content for product selection questions. Data quality, chunk completeness and the clarity of the search terms jointly affect recall precision. Permit questions frequently carry numbers, models and proper nouns, so the behaviour of semantic recall and full-text recall needs to be checked separately against the actual materials at hand.

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 using different vector models cannot be associated together. Search operates on the whole knowledge base; a collection is used only for classification management and does not change search behaviour. Reference content is output as an array, and its length can be zero, so the output path still completes when nothing is found.

Search modes include semantic search, full-text search and hybrid search. Semantic search relies on vector distance. Full-text search relies on keyword hits and suits numbers, models and proper nouns. Hybrid search merges the two paths through RRF and usually needs to be paired with reranking, filtering by the rerank score. The minimum relevance value takes effect only under semantic search or when result reranking is used; it is a filter threshold and does not perform permission or entity isolation. Question optimisation performs coreference resolution and question expansion and adds one model call before retrieval.

Hypothetical paper exercise: on 10 March 2026 a trade compliance role assembles the materials for one shipment. The source text reads: "This shipment consists of ball screws for CNC machine tools, commodity code 8483.40, and the supplier product description states the use as industrial transmission. The contract was signed on 28 February 2026, and no automatic import licence application has been submitted." The draft sentence to be judged is: "The goods have obtained an automatic import licence and can be declared directly." This exercise is used to check whether the retrieved materials can support that conclusion.

Configuration and implementation checks

The settings below relate to this exercise and the primary task. Where a value appears, it is a starting point for this hypothetical trial, not an official default.

Configuration itemHypothetical starting point for this trialReason and comparison method
Associated knowledge basesAssociate only one permit materials base using a single vector modelCheck the vector model identifier of each base first, exclude those that do not match, and do not run cross-model retrieval
Search modeHybrid searchThe exercise contains a commodity code and proper nouns and also needs semantic matching; compare retrieved content across the three modes using the test question
Reference limitMeasured in tokens; start at a length that can hold one product description plus a contract summaryTest content completeness with materials of differing length, then adjust the limit
Minimum relevanceStart at the middle value of the 0-1 rangeLabel relevant and irrelevant samples first, then compare whether filtered reference content covers the exercise materials
Result rerankingEnabled as configured, with a rerank model availableReranking produces a 0-1 score; filter by that score and compare ordering changes
Question optimisationEnabled for multi-turn follow-ups or short questionsCompare the original question with the optimised question and check whether the subject was altered

Three common errors

  1. Treating a retrieval hit as an authorisation result. In the exercise the materials state only that the contract was signed and that no application has been submitted; retrieving that content does not mean the licence has been obtained. Check by entering the draft sentence together with the source text, comparing whether subject, object and status agree between them. A draft stating "has obtained" against a source stating "no application has been submitted" is a mismatch.
  2. Treating minimum relevance as a permission or entity filter. That value only filters low-relevance results and does not exclude unrelated business or unauthorised records. Check by entering a question unrelated to permits and comparing whether the returned reference content still contains permit materials; if it does, the setting does not perform an isolation function.
  3. Associating knowledge bases that use different vector models. Check by reading the vector model identifier of each base first, excluding those that do not match, and recording the actual observation without inferring a runtime result from it.

How to verify the setup

  1. Enter the exercise source text and the draft sentence, then compare whether the retrieved reference content includes the status "no automatic import licence application has been submitted". If the reference content supports "has obtained", the check is not satisfied and the evidence gap must be recorded.
  2. Enter a precise query containing only the commodity code, then compare the retrieved content of full-text search and hybrid search and observe whether the number-type question hits the corresponding material. If it does not hit, record it as pending verification rather than as a negative finding.
  3. Enter a question unrelated to permits and compare whether the returned reference content has length zero or contains only low-relevance fragments, using the explicit paper input [] to check the downstream design for handling an empty result. Do not claim that a fictional rule search necessarily returns an empty array.

This page addresses only import and export permit handling, product materials and selection questions, and knowledge base retrieval and recall. At operational handover, the authorisation status of materials, operations and requests must be kept clear; retrieval and forms must not be treated as authorisation execution.