What this category's data looks like
In the compulsory education dropout prevention and school retention scenario, materials fall mainly into three categories: dropout risk ledgers, return-to-school persuasion records, and home visit records. Ledgers are updated on a rolling basis by semester, with fields including student name, grade, risk level, guardian, responsible teacher, date, and signature; persuasion records are generated per instance, including persuasion time, participants, result, and receipt; home visit records include home visit date, teacher, parent signature, and image description. When these three categories of materials enter the knowledge base retrieval service, each participates in recall as an independent fragment source: ledgers provide the student and responsibility attribution, persuasion records provide the process and result, and home visit records provide supporting evidence and signature status. During material drafting and responsibility handover, the retrieval service first recalls candidate fragments, and based on these, names, quantities, dates, versions, and signatures are compared for consistency, with missing items intercepted by subsequent nodes.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after pattern, merge, re-rank, filter, and upper-limit trimming). The signable material drafts and handover records are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Dropout risk ledger | Fields complete, responsible person clear, version and date consistent | In the search test, search once each using different phrasings for the same student, record the number of candidates and scores, and check whether the name, grade, risk level, responsible teacher, date, and signature fields are hit at the same time |
| Persuasion record | Persuasion time, participants, result, and receipt complete | Switch searchMode to semantic, full-text, and hybrid and test each once, check whether the name, quantity, and date of the same persuasion record match across the two materials, and whether candidates missing a receipt are filtered out |
| Home visit record | Home visit date, teacher, parent signature, and image description consistent | Turn usingReRank on and off and compare the order of the first few entries, check the signature and date; then adjust similarity and limit to see whether the number of citations and text completeness retain key fragments |
Three common errors
- Symptom: The name spelling or quantity for the same student does not match between the ledger and the persuasion record. Cause: Candidate fragments come from different materials, and drafting was based on them without first verifying hits and associations.
- Symptom: The home visit record version is not synchronized, and an old date is mixed with a new signature in the same draft. Cause: After retrieval returned candidates, the version and date were not checked, and the handover proceeded directly.
- Symptom: A persuasion record is missing a parent signature or receipt, yet the draft is still issued. Cause: Similarity filtering and citation upper-limit trimming only affect cited content and do not replace signature and receipt verification.
How to verify the setup
Acceptance for this feature: test once each with synonym-rewritten questions, compare candidates and scores, then turn re-ranking on and off to observe the ordering of the first few entries; record the number of candidates and scores under searchMode, similarity, limit, and usingReRank, record the number of citations separately by array length, and verify the tokens of the cited text separately.
Business test: draft one according to the template and complete one handover, and see whether missing items are intercepted.
- In the search test, search the ledger, persuasion records, and home visit records once each with synonymous questions, record the number of candidates and scores, and check the name, quantity, date, version, and signature fields.
- Run once each with re-ranking on and off, adjust similarity and the citation upper limit, check the cited text tokens and array length, and confirm that candidates missing a receipt or signature are intercepted.