What this category's data looks like
This category covers pre-departure training and overseas safety reminders. Materials include pre-departure training courseware, safety reminder checklists, and accommodation contact forms. Sources are the institution's academic affairs and overseas affairs teams, updated by semester or project batch. The courseware contains schedule, discipline, and emergency contact fields; the safety checklist contains risk items, handling steps, and a receipt column; the accommodation contact form contains address, landlord or dormitory phone, signature, and date. After the materials enter hybrid retrieval, the courseware and checklist serve the role of basis provenance, while the accommodation form serves the role of verifiable contact fields; the rerank model sorts candidates, and citation trimming and answer generation produce basis items that can be traced back to the original text accordingly.
Configuration and implementation checks
The rerank model in the retrieval service only outputs sorted candidates and rerank scores; basis items that can be traced back to the original text must be produced next by citation trimming and answer generation. Hybrid retrieval runs both the vector and full-text paths and merges the results. After merging, the scope is large and similarity filtering cannot be applied directly; typically the rerank model is then used to sort and filter by rerank score.
| Object | Check | Verification action |
|---|---|---|
| Pre-departure training courseware | Fields include schedule, discipline clauses, and emergency contact; the date and signature columns must be consistent with the safety reminder checklist | In the search parameters, check the two-path switches of searchMode, confirm that usingReRank is already used for sorting and filtering, and compare the number of candidates and scores before and after reranking |
| Safety reminder checklist | Risk items, handling steps, and the receipt column must match the accommodation contact form in name, quantity, and date | Check the number of candidates and scores by similarity and limit, and verify whether the name, quantity, and date of the same record are consistent across the two materials |
| Accommodation contact form | Address, landlord or dormitory phone, signature, and date must be the current version and the receipt must be complete | Compare the ordering and scores before and after reranking, confirm that candidates come from both the vector and full-text paths, and list separately any items missing a signature or receipt |
| Hybrid retrieval and rerank pipeline | The two-path results have been merged, and the rerank model is bound and used for sorting and filtering | Check the two-path switches of searchMode and usingReRank, record the number of candidates and rerank scores, and do not compare the number of citations with quoteMaxToken |
Three common errors
- Symptom: the name, quantity, and date of the same record do not match between the courseware and the safety checklist. Cause: the two materials are updated separately, and no cross-material field verification was performed.
- Symptom: the accommodation contact form shows an old address or old phone number. Cause: versions were not synchronized, and old entries remain in the candidates after the new batch of materials replaced them.
- Symptom: the safety reminder checklist is missing a signature or receipt. Cause: the receipt column was not archived together with the materials, and retrieval still returns incomplete entries.
How to verify the setup
Acceptance for this feature: compare the order and scores of the first three items before and after enabling reranking, and confirm that usingReRank takes effect and that the number of candidates and rerank scores can be recorded.
Business test: ask a cross-material basis question and see whether the answer provides traceable provenance.
- In the search parameters, check the two-path switches of
searchModeandlimit, and confirm that both the vector and full-text paths are enabled and merged. - Compare the ordering and scores before and after reranking, check the number of candidates and scores by
similarity, and verify whether the basis items can be traced back to the original text.