What this category's data looks like
This category's data mainly consists of three types of materials for credit management of primary and secondary school teacher training: training notices, credit recognition forms, and participant lists. Sources are documents issued by education administrative departments, school summaries, and individual teacher submissions. The update cadence advances with training batches and semester recognition cycles. Field characteristics center on course name, number of class hours, recognition date, signature, and receipt. Training notices define the project and time; credit recognition forms carry class hours and recognition conclusions; participant lists provide the basis for personnel and attendance. When these three types of materials enter knowledge base retrieval and recall, they respectively serve the roles of rule basis, result evidence, and personnel scope, providing searchable candidate snippets and cited content for authorized access and external services.
Configuration and implementation checks
The knowledge base retrieval service (retrieval node and search test) only outputs candidate snippets and similarity scores (after mode, merging, reranking, filtering, and upper-limit trimming). The externally available entry point and authorization scope must be handled by subsequent nodes or manually. The retrieval mode is selected among semantic, full-text, or hybrid; multiple candidate paths are merged and then subjected to secondary ranking; finally, trimming is performed according to similarity and citation upper limits.
| Object | Check | Verification action |
|---|---|---|
| Training notice | Course name, training date, and number of class hours must match the credit recognition form item by item, and signatures and receipts must be complete | In the knowledge base search test, switch searchMode, search using the course name and date in the notice, check the candidate snippets and scores, and see whether key fields are still retained after similarity filtering |
| Credit recognition form | Recognized class hours, recognition date, and the number of people in the participant list must match; the version must be the latest and signatures must not be missing | Initiate retrieval using the course name and date in the recognition form, compare the order of the top few results before and after enabling usingReRank, record the number of candidates and scores, then check whether the cited text after limit trimming contains the recognition conclusion |
| Participant list | Personnel names, attendance dates, and notice batch must be consistent, and receipts and signatures must not be missing | Use a restricted identity to search for names and batches in the list, check whether the returned candidates fall within the authorized scope, record the number of candidates and scores, and confirm that unauthorized content is not visible |
Three common errors
- Symptom: the name, quantity, or date of the same course do not match across two materials. Cause: the notice and recognition form are maintained separately, lacking unified field verification and a closed loop for signatures and receipts.
- Symptom: the knowledge base still contains an old version of the recognition form, but retrieval returns the old conclusion. Cause: versions are not synchronized, and there is no release confirmation between the update cadence and retrieval recall.
- Symptom: there are many candidates, but the top-ranked ones are irrelevant, and the answer lacks key passages. Cause: the retrieval mode does not match the material form, and reranking and similarity filtering have not been adjusted as needed.
How to verify the setup
Acceptance for this feature: test once each with synonym-rewritten questions, compare candidates and similarity scores, then toggle reranking on and off to observe the order of the top few results; at the same time, record the number of candidates and scores, and check the tokens budget of the cited text.
Business test: access once with a restricted identity, see whether the visible scope matches the setting, and confirm that unauthorized content is not visible.
- In the knowledge base search test, switch
searchMode, use semantic, full-text, and hybrid retrieval respectively for the same course name, and record the number of candidates and scores. - Enable and disable
usingReRankand retrieve once each, compare the order of the top few results, and check whether the cited text retains key fields aftersimilarityandlimittrimming.