What this category's data looks like
Regular higher education institution admissions enrollment and filing involves three types of materials: enrollment plans, filing rules, and admission rosters. The sources are official texts published by provincial admissions and examination institutions and university admissions offices, and the update cadence changes on a rolling basis with batch collection, plan adjustments, rule revisions, and roster receipts. Enrollment plans contain institution professional group, planned number, subject selection requirements, batch, and date fields; filing rules contain filing ratio, tie-breaking ranking, application unit, and effective date; admission rosters contain candidate identifiers, major, admission status, and signed receipts. When these three types of materials enter knowledge base retrieval and recall, they respectively serve as corpus sources for candidate chunks, used to compare whether the name, quantity, date, and signature of the same record are consistent, supporting the generation of version ledgers and change records.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, rerank, filter, and limit trimming). The version ledger and change records are completed by subsequent nodes or manually. Retrieval mode, multi-path merge, rerank, and similarity filtering jointly determine whether the candidates cover the fields and dates of the three types of materials.
| Object | Check | Verification action |
|---|---|---|
| Enrollment plan | Whether the fields, dates, and signatures are consistent with the current batch, and whether the planned number and professional group name match | In the search test, switch searchMode, check the planned number, subject selection requirements, and effective date in the candidate chunks, and record the number of candidates and scores |
| Filing rules | Whether the filing ratio, tie-breaking ranking, and application unit correspond to the same effective version, and whether the old version is traceable | Compare the order of the first few items before and after enabling usingReRank, and check whether the dates and signed receipts of the rule clauses are complete |
| Admission roster | Whether the candidate identifiers, major, and admission status are consistent with the plan and rule records, and whether receipts are missing | Adjust similarity and limit, observe the tokens and array count of the cited text after trimming, and record them separately |
| Version ledger and change records | Whether the current version is unique, the old version is traceable, and citations are updated synchronously | After changing one piece of content, rerun retrieval and check whether the version number and cited chunks change synchronously |
Three common errors
- Phenomenon: The major name and admission date of the same candidate in two rosters do not match. Cause: The candidate chunks come from different batch versions and were not filtered by effective date.
- Phenomenon: The enrollment plan has been adjusted, but the filing rules still cite old version clauses. Cause: The version was not synchronized, the old version was not marked invalid, and retrieval still recalls old chunks.
- Phenomenon: The admission roster lacks signature or receipt fields. Cause: The scanned materials are incomplete, the candidate chunks do not cover the signature page, and the cited content lacks key paragraphs.
How to verify the setup
Acceptance of this feature: Test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ranking of the first few items, and record changes in the number of candidates and scores.
Business test: Change one piece of content, see whether the version number and citations are updated synchronously, and check that the current version is unique, the old version is traceable, and citations are synchronized.
- In the search test, switch
searchMode, compare the number of candidates and scores of the semantic, full-text, and hybrid modes, and confirm that the material form matches. - Compare the order of the first few items before and after enabling
usingReRank, and adjustsimilarityandlimit, recording the cited text tokens and array count separately.