What this category's data looks like
External service materials of primary and secondary school admissions organizations mainly consist of admission plans, application material checklists, and admission lists. An admission plan contains batches, dates, and signature fields; an application material checklist contains items, quantities, and receipts; an admission list contains names, numbers, and dates. The three types of materials are drafted by different roles, their update cadence changes as batches progress, and fields often differ in names, quantities, and dates. When materials enter knowledge base retrieval, the admission plan and application material checklist serve as the main sources for semantic and full-text recall, while the admission list is used to verify whether names and numbers are consistent; during retrieval, each material only provides candidate snippets and does not directly produce external documents.
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 pattern, merge, rerank, filter, and limit trimming). The externally available entry point and authorization scope are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Admission plan | Whether batches, dates, and signature fields are complete, and whether the dates in the plan are consistent with the application material checklist | In the search test, switch searchMode to semantic, full-text, or hybrid, compare candidate snippets and scores, and check the plan dates against the checklist dates |
| Application material checklist | Whether items, quantities, and receipts correspond to the plan; if signatures or receipts are missing, it is not visible | Test once each with synonymous questions, see whether candidates contain key snippets from the checklist, and record the number of candidates and scores |
| Admission list | Whether names, numbers, and dates are consistent with the plan batch, and whether the same record's name matches across materials | Turn usingReRank on or off and compare the ranking of the top few results; adjust similarity to see whether key snippets are still retained among the filtered candidates |
| External entry point and authorization scope | Whether visitor identity matches the visible scope, and whether unauthorized content is invisible | Access once with a restricted identity, see whether the visible scope matches the setting, and check the number of citations after limit trimming |
Three common errors
- Symptom: the name and quantity of the same record do not match between the admission plan and the application material checklist. Cause: the two materials are drafted by different roles, versions are not synchronized, and retrieval only returns their respective candidate snippets.
- Symptom: the admission list is inconsistent with the plan batch dates, and signatures or receipts are missing. Cause: materials update at different cadences, old versions still participate in recall in the knowledge base, and similarity filtering and limit trimming are not applied.
- Symptom: there are many candidates but the ones ranked first are all irrelevant, and the answer lacks key paragraphs. Cause: the retrieval mode does not match the material form, rerank is not enabled, and key snippets are truncated after citation limit trimming.
How to verify the setup
Acceptance of this feature: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank to inspect the ranking of the top few results; check the impact of searchMode, usingReRank, similarity, and limit on the number of candidates and scores.
Business test: access once with a restricted identity, see whether the visible scope matches the setting and whether unauthorized content is invisible, and record the number of citations and the tokens of the cited text.
- In the knowledge base search test, switch
searchMode, compare candidate snippets and scores across semantic, full-text, and hybrid modes, and check whether multi-path merging includes both types of results at the same time. - Turn
usingReRankon or off to compare the ranking of the top few results, adjustsimilarityto see whether key snippets from the original text are retained among the filtered candidates, and adjustlimitto record the number of citations and separately check the tokens of the cited text.