What this category's data looks like
This page addresses the scheduling and class-consumption scenarios of training institutions. Materials come from three types of business documents: class schedules, class-consumption records, and schedule-change applications, along with the accompanying version ledgers and change records. Class schedules are updated weekly or monthly, with fields including course name, class date, time slot, classroom, teacher, and class; class-consumption records are generated one by one, with fields including student, course, consumption date, number of class hours, and signature; schedule-change applications are submitted one by one, with fields including original session, target session, application date, and approval receipt. When materials enter knowledge base retrieval and recall, class schedules play the role of the current scheduling basis, class-consumption records play the role of actual consumption evidence, schedule-change applications play the role of change source, and version ledgers and change records play the role of old-version traceability and reference synchronization. In terms of update rhythm, class schedules change with scheduling adjustments, class-consumption records increase as classes occur, and schedule-change applications change as approvals circulate. All three types of materials must retain old versions and mark the current version.
Configuration and implementation checks
This step is performed 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). Version ledgers and change records are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Class schedule | Course name, class date, time slot, classroom, teacher, and class are consistent across all materials, and the current version is unique | In the knowledge base search test, test once each with synonymous rewritten questions, switch searchMode to compare the number of candidates and scores, and check the date and class fields in the returned snippets |
| Class-consumption record | Student, course, consumption date, number of class hours, and signature are complete, old versions are traceable, and references are synchronized | Toggle usingReRank to compare the ordering of the top few results, and check whether the number of class hours and signature receipt in the candidate snippets correspond to the current version ledger |
| Schedule-change application | Original session, target session, application date, and approval receipt match, and references are synchronized after changes | Adjust similarity to observe changes in the number of candidates and scores, and check whether the original session and target session in the returned snippets are consistent with the change records |
| Version ledger and change records | After changing one piece of content, the version number and references are updated synchronously, and old versions are traceable | After changing one class schedule date, rerun retrieval, record the number of returned candidates and scores, and check whether the reference text points to the current version |
Three common errors
- Symptom: The number of class hours for the same class-consumption record does not match between the class schedule and the class-consumption materials. Cause: The two materials are maintained separately without synchronization, and the version ledger does not record the change.
- Symptom: A schedule-change application has been approved, but the class schedule still shows the original session, and the reference snippet points to the old version. Cause: The version is not synchronized, and the old version is not marked as invalid.
- Symptom: A class-consumption record lacks a signature or receipt, but the candidate snippet is still recalled. Cause: The material fields are incomplete, and similarity filtering does not intercept it.
How to verify the setup
Acceptance for this feature: test once each with synonymous rewritten questions, compare the number of candidates and scores, then toggle rerank to see the ordering of the top few results.
Business test: change one piece of content and see whether the version number and references are updated synchronously.
- In the knowledge base search test, switch
searchMode, record the number of candidates and scores, and check whether the returned snippets contain both semantic and full-text results. - Adjust
similarityandlimit, look at the tokens and array length of the reference text, record the length separately, and do not compare the length as equal to the reference budget.