What this category's data looks like
This page addresses kindergarten operation behavior and quality self-assessment. Materials include self-assessment reports, issue lists, and rectification plans. Self-assessment reports are formed by academic year or semester, with fields covering operation conditions, safety and hygiene, care and education, faculty and staff, and internal management, along with dates and signatures; issue lists are formed by instance, with fields covering issue description, responsible department, rectification deadline, and names and quantities of supporting materials; rectification plans are formed by batch, with fields covering measures, completion time, and receipts. When the three enter knowledge base retrieval and recall, the self-assessment report serves as the primary evidence source, the issue list as the anomaly clue source, and the rectification plan as the closed-loop supporting evidence source. Retrieval results are presented as candidate snippets and scores for subsequent nodes or manual tracing and comparison.
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 limit trimming). The "anomaly list and review conclusion" is completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Self-assessment report | Whether fields, dates, and signatures are complete, and whether the name and quantity of the same record match the issue list | Retrieve the record name in the search test, record the number of candidates and scores, and verify the original date and signature positions |
| Issue list | Whether the issue description, responsible department, and rectification deadline correspond to the rectification plan, and whether receipts are missing | Retrieve once each with synonymous questions, compare candidates and scores, and check whether different expressions of the same record are recalled |
| Rectification plan | Whether the completion time and receipts are synchronized with the self-assessment report version, and whether the quantities match | Switch searchMode to hybrid, enable usingReRank to compare the ranking of the top few, and verify candidates under similarity and limit |
| Knowledge base retrieval service | Whether candidate snippets and scores are returned after mode, merge, rerank, filter, and limit trimming | Switch between semantic, full-text, and hybrid modes in the search test, record the number of candidates and scores, and verify the tokens of the cited text |
Three common errors
- Symptom: The name and quantity of the same record do not match between the self-assessment report and the issue list. Cause: Naming and measurement conventions were not unified during entry, and no original text comparison was performed after retrieval hits.
- Symptom: The rectification plan version is not synchronized, and receipts are missing. Cause: Materials were not re-ingested after updates, retrieval returned old-version candidates, and manual review did not verify signatures and receipts.
- Symptom: There are many candidates, but the top-ranked ones are irrelevant. Cause: The retrieval mode does not match the material form, rerank was not enabled, and key snippets were squeezed out after similarity filtering and citation limit trimming.
How to verify the setup
Acceptance for this feature: Test once each with synonymously rewritten questions, compare the number of candidates and scores, then toggle rerank on and off to observe the ranking of the top few, and verify the returns under searchMode, similarity, limit, and usingReRank.
Business test: Create an anomaly record, see whether the source material can be traced, and verify the dates and signatures of the self-assessment report, issue list, and rectification plan.
- In the knowledge base search test, switch between semantic, full-text, and hybrid modes, record the number of candidates and scores, verify the tokens of the cited text, and record the array length separately.
- Retrieve the same record using different expressions, run once with rerank enabled and once disabled, compare the order of the top few, and confirm that key original snippets are still retained after similarity filtering and citation limit trimming.