What this category's data looks like
The data in this category consists mainly of admissions plans, filing rules, and admission rosters, with fields including institution-major groups, planned numbers, filing ratios, candidate numbers, scores, dates, and signatures. The admissions plan is the baseline for filing ratios and planned numbers, the filing rules are the basis for ranking and tie-breaking, and the admission roster is the evidence of results and signatures. Sources are mostly provincial admissions office documents, institutional admissions regulations, and internal approval documents, and the update cadence changes with batch collection, plan adjustments, and rule supplements. When the materials enter the knowledge base retrieval node, each plays a different role: the admissions plan corresponds to quantity and date verification, the filing rules correspond to rule clauses and applicable batches, and the admission roster corresponds to the list and signature receipts. The name, quantity, and date of the same record may not match across two materials; when versions are not synchronized or signature receipts are missing, cited snippets and original paragraphs are difficult to trace back to each other.
Configuration and implementation checks
This step is executed by the citation output of the knowledge base retrieval node, which outputs an array of cited content, each item containing the original snippet and source identifier, and the length may be 0. The "version ledger and change records" are completed subsequently by AI nodes, code nodes, or manual work.
| Object | Check | Verification action |
|---|---|---|
| Associated knowledge base | The selected knowledge bases use the same vector model | In the node, check the selected collections and their vector models, and confirm that the three types of materials—admissions plans, filing rules, and admission rosters—belong to the same vector model |
| Cited content output | The number of returned items matches the hit materials; empty is also normal | Run a test with a question that has no matching materials and see that the output is empty while the chain continues; run a test with a question that has original-text support and see that the number of returned items matches the hit materials, and check the length of the citation array in responseData (non-streaming module results returned with detailed output enabled) and the hit records in quoteList |
| Citation-to-original correspondence | Each citation can be traced back to the original paragraph | In the conversation, open the citation markers and compare them with the source documents to check the quantities and dates in the admissions plan, the clauses and batches in the filing rules, and the list and signatures in the admission roster |
| Search parameters | Consistent with the values set in the knowledge base retrieval section | Open the search parameters panel and check item by item; when the citation limit is involved, check the tokens of the cited text, record the array item count separately |
Three common errors
- Phenomenon: The name, quantity, and date of the same record do not match between the admissions plan and the admission roster. Cause: The two materials are maintained by different stages and were not compared using the same field definitions.
- Phenomenon: The filing rules have been revised, but citations still point to the old clauses. Cause: Versions are not synchronized, and the knowledge base was not replaced or appended as the file was updated.
- Phenomenon: The admission roster lacks signatures or receipts, and the cited snippet cannot be traced back to the complete evidence. Cause: The signature page was not ingested together with the main text, and citations cover only the main-text paragraphs.
How to verify the setup
Acceptance for this feature: ask a question with original-text support, check whether the citation array and markers can be traced back to the original text, and check the number of citations in responseData and the hit records in quoteList; then ask a question with no matching materials and confirm that the array is empty while the chain still runs through.
Business test: change one piece of content and see whether the version number and citations are updated in sync.
- Check whether the vector models of the selected knowledge bases are consistent, and confirm that the three types of materials—admissions plans, filing rules, and admission rosters—belong to the same vector model.
- Open the search parameters panel and check item by item; when the citation limit is involved, check the tokens of the cited text, record the array item count separately; the number of citations is the array length, and
quoteMaxTokenis the citation content budget, and the two are not compared for equality.