What this category's data looks like
This page addresses the enterprise new employee onboarding training scenario, and the materials include three types: onboarding training plans, course sign-in sheets, and job descriptions. Onboarding training plans are updated by batch, with fields including course name, training date, instructor, participant list, and approval signature; course sign-in sheets are generated by session, with fields including sign-in date, signer, position, and signature or receipt; job descriptions are published by position version, with fields including position name, responsibility items, effective date, and version number. When these three types of materials enter hybrid retrieval, each plays a different role: onboarding training plans provide training arrangements and basis items, course sign-in sheets provide attendance and signature evidence, and job descriptions provide position responsibilities and version benchmarks. In terms of update cadence, training plans are adjusted by batch, sign-in sheets are archived by session, and job descriptions are re-versioned as positions change. Field characteristics are concentrated in names, quantities, dates, signatures, and version numbers, and these fields determine whether a basis item can be traced back to the original text and whether it is judged to be the current version.
Configuration and implementation checks
This step is executed by the rerank model in the retrieval service, which outputs the sorted candidates and rerank scores. "Basis items that can be traced back to the original text" are then completed by quote clipping and answer generation. Hybrid retrieval simultaneously runs the vector and full-text paths and merges the results; after merging, the scope is large and similarity filtering cannot be applied directly, so the rerank model is usually then used to sort and filter by rerank score.
| Object | Check | Verification action |
|---|---|---|
| Onboarding training plan | Training date, course name, number of participants, and approval signature must match the sign-in sheet item by item, and it must be the current batch | In the search parameters, check whether searchMode enables both the vector and full-text paths at the same time, and see whether candidates come from both paths |
| Course sign-in sheet | Sign-in date, signer, position, and signature or receipt must be consistent with the training plan for the same session; items missing a signature or receipt cannot be used as a basis | Compare the ordering and scores before and after reranking, and check that usingReRank is bound and used for sorting and filtering |
| Job description | Position name, responsibility items, effective date, and version number must be the current version, and the name and date of the same record in the two materials must not conflict | Adjust limit to observe changes in the number of candidates and scores, and confirm that the top few are relevant to cross-material questions |
| Rerank model | The sorted candidates and rerank scores must be able to support filtering by score, and the top few should be more relevant than the direct hybrid retrieval results | Compare the order and rerank scores of the top three before and after enabling reranking, and record the number of candidates and scores |
Three common errors
- Symptom: The course name or number of participants for the same record does not match between the training plan and the sign-in sheet. Cause: The two materials are entered separately, and field verification was not performed by the same session and the same batch.
- Symptom: The responsibility items referenced in the job description are inconsistent with the current version. Cause: The version was not synchronized, and the old job description is still left among the candidates and was not filtered by effective date.
- Symptom: A sign-in sheet item lacks a signature or receipt but is still output as a basis. Cause: Quote clipping did not verify the signature field, and answer generation did not trace back to the original text.
How to verify the setup
This feature acceptance: compare the order and rerank scores of the top three before and after enabling reranking, check the number of candidates and scores, and confirm that usingReRank takes effect and that after adjusting limit the top few are more relevant.
Business test: ask a cross-material basis question and see whether the answer gives a traceable source; the source must come from the institution’s materials, be the current version, and the item must match the original text.
- In the search parameters, check that
searchModeenables both the vector and full-text paths at the same time, see whether candidates come from both paths, and compare the ordering and scores before and after reranking. - Verify item by item the names, quantities, dates, signatures, and version numbers of the training plan, sign-in sheet, and job description by batch and session, and record the tokens and array item count of the quoted text.