What this category's data looks like
In the self-taught examination registration record and graduation review scenario, materials mainly consist of registration archives, course qualification certificates, and graduation application forms, sourced from archived documents of provincial examination offices and host institutions, with updates rolling according to examination periods and graduation review batches. Registration archives record candidate identity and enrollment status, course qualification certificates record individual subject scores and dates of attainment, and graduation application forms carry completion status of the professional plan and signed receipts. When materials enter the vector model and indexing stage, their roles differ: registration archives and course qualification certificates are written into the business database as source data, while graduation application forms participate in retrieval and recall as review evidence. Field characteristics center on name, examination registration number, course code, score, date, and signature field; the same record often shows mismatched names, quantities, and dates across two materials, with versions not synchronized or missing signed receipts, requiring a version ledger and change records to fix the current version as unique, old versions as traceable, and references as synchronously updated.
Configuration and implementation checks
This stage is executed by the indexing service and vector database, outputting vectors and index status (whether recall is possible, whether data in the same group is merged). The "version ledger and change records" are subsequently completed by downstream AI nodes, code nodes, or manual work (retrieval and reference trimming only take supporting materials). The indexing model remains consistent with the vector model actually used by the knowledge base; the index field of a data row retains multiple vector identifiers for one piece of data; after changing models, rebuild is triggered only for data that has already been indexed.
| Object | Check | Verification action |
|---|---|---|
| Registration archive | Whether fields, dates, and signatures align item by item with the course qualification certificate and graduation application form | Open one registration archive data row, check whether the multiple vector identifiers in the index field are complete, and confirm the status is indexed |
| Course qualification certificate | Whether the name, quantity, and date of the same record are mismatched across the two materials | Search the course code, see whether data in the same group is merged, and check the tokens budget of limit in the reference text |
| Graduation application form | Whether it is still referenced as the current version when the signature or receipt is missing | Check the data range covered by the rebuild task record, confirm that old versions are traceable and references are synchronized, and record the array length separately |
| Version ledger and change records | Whether the current version is unique, old versions are traceable, and references are synchronized | After changing one piece of content, check whether the version number and references are synchronously updated, and inspect the reference budget of limit |
Three common errors
- Symptom: duplicate entries appear in retrieval results. Inspect the following: the name, quantity, and date of the same record are mismatched across the two materials, and vector identifiers are not merged by the same data group.
- Symptom: recall quality drops after changing models. Inspect the following: the rebuild task record only covers data that has already been indexed, and unsynchronized old versions are still referenced.
- Symptom: a data row has content but cannot be retrieved. Inspect the following: the signature or receipt is missing, the graduation application form has not entered the current version, and reference synchronization is interrupted.
How to verify the setup
This feature's acceptance: sample one data row and inspect the index field, confirm that multiple vector identifiers are complete and the status is indexed, then search once to see whether the same group is merged, with the vector score taking the highest score.
Business test: change one piece of content and see whether the version number and references are synchronously updated; the reference count is the array length, quoteMaxToken is the reference content budget, the two are not compared for equality, check the tokens of the reference text, and record the array length separately.
- In the knowledge base settings, check that the indexing model is consistent with the vector model actually used, and view the data range covered by the rebuild task record.
- Open one data row and check the index field, confirm the reference text tokens budget corresponding to
limit, and record the array length.