What this category's data looks like
In the teacher qualification examination and certification business, materials such as written and interview registration forms, Mandarin proficiency certificates, and medical examination forms each play different roles after entering the knowledge base: the registration form provides exam subjects and dates, the Mandarin proficiency certificate provides the level and number, and the medical examination form provides the conclusion and signature. The material sources are the current documents issued by local certification agencies and examination institutes, and the update pace changes with certification batches and policy adjustments. Field characteristics are concentrated in names, quantities, dates, and signatures. The same record often has mismatched names, quantities, and dates across two materials, versions are not synchronized, and signatures or receipts are missing. When these materials enter the vector model and index function, they are first split into data rows, and each row can generate multiple vector identifiers for subsequent reference back to the basis entries by original text.
Configuration and implementation checks
This step is executed by the index service and vector database, and outputs vectors and index status (whether recall is possible, whether data in the same group is merged). "Basis entries that can be traced back to the original text" are then completed by subsequent AI nodes, code nodes, or manual work (retrieval and citation trimming only take supporting materials).
| Object | Check | Verification action |
|---|---|---|
| Index model | Consistent with the vector model actually used by the knowledge base | Check the index model in the knowledge base settings, confirm it is consistent with the current vector model, and verify the values of defaultToken and maxToken |
| Index field of the data row | Multiple vector identifiers of one data item are all present | Open a data row and check the index field to confirm that multiple vector identifiers are complete and the status is indexed |
| Rebuild | After changing the model, trigger only for data that has already been indexed | Check the rebuild task records and confirm that the coverage is limited to data that has already been indexed |
| Written and interview registration form | Fields, dates, and signatures can match the Mandarin proficiency certificate and medical examination form | Check the names, quantities, and dates of the registration form, Mandarin proficiency certificate, and medical examination form, and check whether signatures or receipts are complete |
Three common errors
- Symptom: duplicate entries appear in retrieval results. Inspect the following: the same group of data corresponds to multiple vector identifiers, and when merging back to the original data, they are not merged by the highest score.
- Symptom: recall quality drops after changing the model. Inspect the following: the rebuild task records only cover data that has already been indexed, and unsynchronized versions cause the index model to be inconsistent with the vector model.
- Symptom: a data row has content but cannot be retrieved. Inspect the following: the index field is missing vector identifiers, or the material version is not synchronized and signatures or receipts are missing, causing the data row to fail to complete indexing.
How to verify the setup
Acceptance for this feature: sample a data row and check the index field to confirm that multiple vector identifiers are complete and the status is indexed, then retrieve once to see whether data in the same group is merged and whether the score takes the highest value; check whether defaultToken and maxToken match the index model.
Business test: ask a basis question across materials and see whether the answer gives a traceable source, and check the number of citations and the tokens of the cited text. quoteMaxToken is the citation content budget, and the number of array entries is recorded separately.
- Check the index model in the knowledge base settings, confirm it is consistent with the vector model actually used by the knowledge base, and view the data range covered by the rebuild task records.
- Open a data row and check the index field to confirm that multiple vector identifiers are complete; during retrieval, confirm that data in the same group is merged and the score takes the highest value, with status indexed.