What this category's data looks like
Material verification and discrepancy localization for primary and secondary school admissions organizations revolve around three types of materials: admission plans, application material checklists, and admission lists. Admission plans are published annually by the education authority and include admission scope, time nodes, and admission rules; application material checklists are organized by schools or registration points by batch and include student name, ID number, address, guardian information, and signature field; admission lists are generated during the admission stage and include name, admission category, date, and receipt. The three types of materials update at different rhythms: plans update by academic year, checklists roll by application batch, and lists are written back by admission batch. In terms of field characteristics, name, quantity, date, and signature are the four places most prone to inconsistency across materials for the same record. When materials enter the knowledge base retrieval function, each serves as a source of candidate chunks for subsequent nodes or manual comparison and discrepancy localization.
Configuration and implementation checks
This step is performed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, rerank, filter, and limit trimming). The discrepancy list is then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Admission plan | The application dates, batch quantities, and signature field in the plan must be consistent with the corresponding fields in the application material checklist and admission list | In the search test, use the name and date of the same record to search separately, switch searchMode to compare the number of candidates and scores, then enable usingReRank to see the ranking of the top few results |
| Application material checklist | The student name, ID number, address, guardian information, and receipt signature in the checklist must be consistent with the same record in the admission plan and admission list | Use name plus ID number to construct synonymous questions and test each once, record the number of candidates, check the similarity score and cited text tokens, and confirm that key chunks have not been trimmed by limit |
| Admission list | The name, admission category, date, and receipt in the list must be consistent with the same record in the admission plan and application material checklist | For the same record, switch to searching by date and admission category, compare the order of the top few results before and after enabling rerank, and check item by item whether the returned citations cover the signature and receipt fields |
Three common errors
- Phenomenon: The name or ID number of the same student does not match between the application material checklist and the admission list. Cause: The two entries come from different input sources, and no field-level comparison was performed for the same object.
- Phenomenon: The batch dates in the admission plan are inconsistent with the dates in the checklist and list. Cause: After the plan version was updated, the checklist and list were not written back in sync.
- Phenomenon: The application material checklist lacks the guardian signature or receipt, but the student has already been included in the admission list. Cause: The material verification step did not validate the signature and receipt as independent fields.
How to verify the setup
Acceptance of this function: test each synonymous rewritten question once, compare candidate chunks and similarity scores, then toggle rerank on and off to see whether the ranking of the top few results changes, and record the number of candidates.
Business test: use one real record to find fields inconsistent with the ledger, hand them to a reviewer for confirmation, and write back.
- In the search test, switch
searchMode, search separately by the name, quantity, and date of the same record, and record the number of candidates and thesimilarityscore. - Compare the ranking of the top few results before and after enabling
usingReRank, check the cited text tokens and array length, and confirm thatlimittrimming has not lost signature and receipt chunks.