What this category's data looks like
In the scenario of the establishment and annual reports of Chinese-foreign cooperatively-run schools, materials come from the establishment application, cooperation agreement, articles of association, as well as annual report receipts and signature pages. The update cadence rolls with application batches and annual submissions, and the field characteristics are concentrated in names, quantities, dates, signatures, and version numbers. The establishment application carries the institution's basic information and declared school-running scale, the cooperation agreement carries the names of both parties, the cooperation term, capital contribution, and agreed number of people, and the articles of association carry the governance structure, decision-making procedures, and signing requirements. When the three types of materials enter the verification function, each serves as an independent comparison source, used to locate field discrepancies for the same record across different texts and to form a discrepancy list that can be handed to a reviewer for confirmation, then written back to the ledger.
Configuration and implementation checks
This part is executed by application settings (response limit, etc.), which output the length of a single answer and the range of context that can be accommodated. The discrepancy list is then completed by the model, and entry and authorization are handled by the publishing service and file permissions.
| Object | Check | Verification action |
|---|---|---|
| Establishment application | Whether the institution name, declared school-running scale quantity, application date, and signature field are consistent with the ledger | Compare name, quantity, date, and signature field by field, record inconsistent items, and check the answer length and context range corresponding to maxResponse and maxContext |
| Cooperation agreement | Whether the names of both cooperating parties, cooperation term, capital contribution, and agreed number of people are consistent with the application | Check name, quantity, date, and signature, mark places where versions are not synchronized, record the number of citations separately, and quoteMaxToken only verifies the tokens of the cited text |
| Articles of association | Whether the governance structure, decision-making procedures, and signing requirements are consistent with the agreement and application | Check whether signatures and receipts are complete, locate missing fields, and confirm that the material status is parsed |
| maxContext and maxResponse | Whether the length of a single answer and the range of context that can be accommodated match long-text verification | Check the switch and slider position in application settings, and observe whether the ending is complete |
Three common errors
- Phenomenon: the same institution name is written differently in the application and the agreement. Cause: the two materials were drafted separately, without field-level alignment and version synchronization.
- Phenomenon: the declared school-running scale quantity does not match between the two materials. Cause: after application, the agreement was revised but not written back, and the ledger still retained the old value.
- Phenomenon: the articles of association lack a signature or receipt. Cause: the signing process was not closed-loop, and the scanned copy and original version were not synchronized and archived.
How to verify the setup
Acceptance of this function: ask a question that requires a long answer, see whether the ending is complete, confirm the response limit and truncation behavior, and check the context and citation budgets corresponding to the three identifiers maxContext, maxResponse, and quoteMaxToken.
Business test: use one real record to find a field that is inconsistent with the ledger, hand it to a reviewer for confirmation, and write it back.
- Check the response limit switch and slider position in application settings, and confirm that the length of a single answer matches the page requirements.
- Compare usage between two runs, confirm the change in single-run usage after adjusting the limit, and record the number of citations and cited text tokens.