What this category's data looks like
Kindergarten home-school co-education communication materials mainly come from class notices, child observation records, parent request registrations, and commitment response ledgers, with an update cadence that follows daily class-leading and shift-handover nodes. When these materials enter the knowledge base retrieval and recall function, notices and observation records serve as the source of candidate snippets, parent requests and commitment items serve as the source of classification and time fields, and handover forms serve as the source of version and connection status. Category check dimensions include commitment items and response times, child performance descriptions, parent request classification, and notice delivery; common inconsistencies are unresponded commitments, descriptions that do not match observation records, and unclassified feedback. In deployment, operations, and application handover, the materials respectively correspond to deployment status, operation records, and task records, and the three must remain consistent with version and connection before handover to operations or the next shift.
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). The handover-ready deployment status and operation records are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Commitment items and response times | Whether the commitment received a response within the agreed time, and whether the response record corresponds to the request classification | In the knowledge base search test, test once each with synonymous questions, record the number of candidates and scores, and check whether similarity and the commitment time field are hit at the same time |
| Child performance descriptions | Whether the description is consistent with the original text of the observation record, and whether there is any wording that does not match the observation record | Switch searchMode to semantic, full-text, or hybrid, compare whether the candidate snippets retain the key sentences of the observation record, and check the number of returned candidates |
| Parent request classification | Whether requests are classified into existing categories, and whether unclassified feedback is marked separately | Enable and disable usingReRank, compare the ordering of the first few items, and check whether classification labels are returned together with the candidate snippets |
| Notice delivery and handover form | Whether notice delivery records, deployment status, operation records, and task records are consistent with version and connection | Check item by item once according to the handover form, and record missing items; the quote limit only checks the tokens of the quoted text, the number of array items is recorded separately, and limit and quoteMaxToken are not compared |
Three common errors
- Symptom: a commitment item has no response record in the handover form. Cause: the request classification is not associated with the commitment time field, and retrieval candidates only return notice snippets.
- Symptom: the child performance description is inconsistent with the original text of the observation record. Cause: the order of candidate snippets changes after reranking, and no manual review of the original text was performed for verification.
- Symptom: parent feedback is unclassified, and the next shift cannot check by category. Cause: operation records only record the task name, not the feedback classification and response status.
How to verify the setup
This function's acceptance: test once each with synonymously rewritten questions, compare candidates and scores, then toggle reranking to observe the ordering of the first few items; switch searchMode and record the number of candidates and scores, and enable and disable usingReRank to compare ordering changes.
Business test: check item by item once according to the handover form to see which record is missing; only when version, connection, and task records are consistent may it be handed over to operations or the next shift.
- In the knowledge base search test, test once each with synonymous questions, record the number of candidates and scores, and check whether under
similarityandsearchModethe candidates simultaneously include semantic and full-text results. - Enable and disable
usingReRankto compare the ordering of the first few items, check version, connection, and task records item by item according to the handover form, mark missing items, and then hand over to operations or the next shift.