What this category's data looks like
This page addresses graduate training process management. Materials include training plans, proposal reports, and mid-term assessment forms, sourced from versions and receipts submitted at each training stage, with update cadence following semesters and assessment milestones. Training plans define courses and research directions, proposal reports record topic selection and justification, and mid-term assessment forms summarize progress and signatures. When these three types of materials enter knowledge base retrieval and recall, each plays a different role: training plans provide the training framework and schedule, proposal reports provide the basis for topic selection and review comments, and mid-term assessment forms provide stage conclusions and signature status. Field characteristics center on names, quantities, dates, and signatures. The same record often shows mismatched names, quantities, and dates across different materials, unsynchronized versions, and missing signatures or receipts. During retrieval, the retrieval mode must be selected according to material form to ensure candidate snippets cover key fields, facilitating subsequent nodes or manual generation of externally available entry points and authorization scopes.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate snippets and similarity scores (after mode, merge, rerank, filter, and upper-limit trimming). Externally available entry points and authorization scopes are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Training plan | Fields, dates, and signatures must be consistent with the proposal report and mid-term assessment form; names and quantities must not appear in two sets | In the search test, switch searchMode to semantic, full-text, or hybrid, compare candidate snippets and scores, and check whether date and signature fields are hit simultaneously |
| Proposal report | Topic name, justification date, and review signature must align with the training plan and mid-term assessment form; when versions are not synchronized, the latest receipt prevails | Test once each with synonym-rewritten questions, observe the similarity score and number of candidates, and confirm key passages are not trimmed |
| Mid-term assessment form | Stage conclusions, quantities, and signature receipts must be consistent with the training plan and proposal report; when signatures or receipts are missing, it must not be visible | Compare the order of the top few entries before and after enabling usingReRank, record the number of candidates and scores, and verify that cited text tokens and array counts are recorded separately |
| Knowledge base retrieval service | Candidate snippets and similarity scores are output after mode, merge, rerank, filter, and upper-limit trimming, and do not directly generate authorization results | In the search test, adjust limit to observe changes in the number of candidates, and confirm that the citation upper limit only trims cited text tokens without changing source associations |
Three common errors
- Symptom: The same record has mismatched names, quantities, and dates between the training plan and the proposal report. Cause: versions are not synchronized, retrieval hits old-version snippets, and no field-level verification is performed.
- Symptom: The mid-term assessment form is missing signatures or receipts but is still recalled as a candidate. Cause: material status is not updated with the version, and similarity filtering does not cover signature fields.
- 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, rerank is not enabled, or the citation upper limit trims excessively.
How to verify the setup
Acceptance for this feature: test once each with synonym-rewritten questions, compare candidate snippets and similarity scores, then toggle usingReRank to inspect the ordering of the top few entries, and record the number of candidates and scores.
Business test: access once with a restricted identity, check whether the visible scope matches the setting, ensure unauthorized access is not visible, and open entry points for parents or superiors.
- In the knowledge base search test, switch
searchModeto semantic, full-text, or hybrid, compare candidate snippets and scores, and confirm key fields are hit simultaneously. - Compare the order of the top few entries before and after enabling
usingReRank, record the number of candidates and cited text tokens, record the array count separately, and verify thatlimittrimming does not change source associations.