What this category's data looks like
For the pre-departure training and overseas safety reminders category, materials include pre-departure training courseware, safety reminder checklists, and accommodation contact forms. Sources are the project team and overseas service providers, and the update cadence rolls forward by batch and by weekly/monthly milestones. The courseware carries training key points and progress markers, the safety reminder checklist carries risk items and acknowledgment requirements, and the accommodation contact form carries address, contact person, dates, and signature. When these three types of materials enter knowledge base retrieval and recall, the courseware and checklist serve as the main sources for semantic and full-text recall, while the contact form serves as an auxiliary source for field verification and exception notes; the retrieval step is only responsible for returning candidate chunks and similarity scores, and the progress table and statistical scope notes are generated by subsequent nodes or manually based on these. Names, quantities, dates, and signatures in the materials need to be aligned item by item on the business side.
Configuration and implementation checks
The knowledge base retrieval service (retrieval node and search test) only outputs candidate chunks and similarity scores (after mode, merge, rerank, filter, and limit trimming), and the progress table and statistical scope notes must be done next by subsequent nodes or manually. One retrieval consists of question optimization, semantic/full-text/hybrid recall, multi-path merge, rerank secondary sorting, similarity filtering, and quote limit trimming, and only then is the quoted content handed to the model obtained.
| Object | Check | Verification action |
|---|---|---|
| Pre-departure training courseware | Fields include training batch, courseware version, date, and signature column; the name and quantity of the same record in the courseware and the safety reminder checklist must be consistent | In the knowledge base search test, test once each with synonymous questions, record the number of candidates and scores, and verify whether the courseware version and signature column are synchronized |
| Safety reminder checklist | Fields include risk items, acknowledgment requirements, date, and version number; missing signature or acknowledgment is regarded as inconsistent | Switch searchMode to semantic, full-text, or hybrid, compare whether the candidates simultaneously contain both types of recall results, and verify the acknowledgment and version |
| Accommodation contact form | Fields include address, contact person, check-in date, and signature; the date of the same record in the contact form and the courseware must be aligned | Compare the top few rankings before and after enabling usingReRank, record the number of candidates and scores, and verify whether the date and signature have missing items |
| Progress table and statistical scope notes | The number of records and the statistical count must match, the scope must be written in the header, and exceptions must be noted one by one | Use one week of records to verify the total, check the tokens and array length of the quoted text after similarity filtering and record them separately, and ensure key chunks are still retained after limit trimming |
Three common errors
- Phenomenon: The name, quantity, and date of the same record in the courseware and the safety reminder checklist do not match. Cause: Material versions are not synchronized, the field scope is not written in the header, and verification only compares totals rather than details.
- Phenomenon: The accommodation contact form lacks signatures or acknowledgments, and there is a discrepancy between the statistical total and the number of records. Cause: The acknowledgment collection and signature steps are not included in the progress table, and exceptions are not noted one by one.
- Phenomenon: There are many candidates, but the ones ranked at the front are all irrelevant, and the answer lacks key paragraphs. Cause: The retrieval mode does not match the material form, rerank is not enabled, or the similarity filtering threshold and quote limit trimming are not adjusted as needed.
How to verify the setup
Acceptance for this feature: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ranking of the top few; record the number of candidates and scores, verify the effects of similarity, searchMode, usingReRank, and limit, record the number of quotes separately by array length, and verify the tokens of the quoted text separately.
Business test: use one week of records to verify the total, check whether the scope and exceptions are clearly written, and check whether the fields, dates, and signatures of the courseware, checklist, and contact form are aligned.
- In the knowledge base search test, switch retrieval modes, compare whether semantic and full-text candidates appear simultaneously, and record the number of candidates and scores.
- Compare the order of the top few before and after enabling rerank, verify the tokens and array length of the quoted text, and check whether key chunks are retained after similarity filtering and quote limit trimming.