What this category's data looks like
Admissions enrollment and filing materials for general higher education institutions mainly consist of three categories: enrollment plans, filing rules, and admission rosters. The sources are documents issued by provincial admissions offices and receipts from institutions, and the update cadence follows batch progression with daily additions. Enrollment plans provide major groups, planned numbers, selected examination subjects, and remarks; filing rules provide filing ratios, tie-breaking sorting, batch time nodes, and signature fields; admission rosters provide candidate numbers, filing scores, admitted majors, and receipt status. When these three categories of materials enter the knowledge base retrieval service, each assumes the role of a source of candidate snippets: plans and rules are used to compare field definitions, and rosters are used to verify record counts and totals. Field characteristics are concentrated in dates, quantities, signatures, and version numbers. The same record often fails to match in name, quantity, and date across two materials. When versions are not synchronized or signed receipts are missing, retrieval can only return candidate snippets and similarity scores, and cannot directly provide statistical tables or authorization results.
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 "progress table and statistical definition notes" are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Enrollment plan | Material row: fields include major group, planned number, selected examination subjects, date, and signature; check whether the name and quantity of the same record are consistent across the two materials | In the knowledge base search test, test once each with synonymous questions, switch searchMode to compare the number of candidates and scores, and record whether the planned number field is complete |
| Filing rules | Material row: fields include filing ratio, tie-breaking sorting, batch nodes, and signature fields; check whether the version number and date are synchronized | Test once each with usingReRank enabled and disabled, and observe changes in the ordering of the top few results; check whether the original text of tie-breaking sorting is still retained after similarity filtering |
| Admission roster | Material row: fields include candidate number, filing score, admitted major, and receipt; check whether the record count and total match | Use one week of records to verify the total, and see whether the definitions and exceptions are clearly stated; after limit trimming, record the tokens of the quoted text and the number of array entries separately |
| Progress table and statistical definition notes | Finished row: record counts and statistical counts match, definitions are written in the header, and exceptions are listed separately | Check fields, dates, and signatures item by item against the three materials: enrollment plan, filing rules, and admission roster; mark entries missing signatures or receipts separately |
Three common errors
- Phenomenon: the major name for the same candidate number does not match between the roster and the plan. Cause: the versions of the two materials are not synchronized, and retrieval only returns candidate snippets without cross-material field alignment.
- Phenomenon: the date and batch node in the filing rules differ by one day in the progress table. Cause: after the rules file was updated, it was not synchronized to the roster, and when candidate scores are similar, the older version snippet ranks higher.
- Phenomenon: the roster total does not match the record count in the progress table. Cause: entries missing signatures or receipts were not included in the total, key snippets were trimmed after similarity filtering, and the definitions were not written in the header.
How to verify the setup
Acceptance of this feature: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank to observe the ordering of the top few results; check the returns of searchMode, similarity, usingReRank, and limit, record the number of quoted entries separately by array length, and record quoted text tokens separately.
Business test: use one week of records to verify the total, see whether the definitions and exceptions are clearly stated, and compare fields, dates, and signatures item by item against the enrollment plan, filing rules, and admission roster.
- In the knowledge base search test, retrieve once each with two phrasings of the same question, record the number of candidates and scores, and after switching
searchMode, compare whether semantic and full-text results appear simultaneously. - With
usingReRankenabled and disabled, observe the order of the top few results once each, check whether key passages are retained aftersimilarityfiltering, and afterlimittrimming register the quoted text tokens and the number of array entries separately.