What this category's data looks like
In the enrollment consultation conversion scenario for training institutions, business basis queries rely on three types of materials: enrollment brochures, consultation records, and trial class schedules. Enrollment brochures contain course names, class types, start dates, fee policies, and version numbers; consultation records contain parents' requests, communication dates, handlers, and receipts; trial class schedules contain trial times, teachers, classrooms, and sign-in signatures. These materials are updated by enrollment season, and new versions are generated when brochures are revised, consultation records are supplemented, or trial schedules are rescheduled. When materials enter this feature, brochures and trial schedules serve as searchable supporting materials, while consultation records serve as the basis source that can be traced back to the original text; together, the three support replying to parents or writing into proposals. In terms of field characteristics, the names, quantities, and dates of the same record may be inconsistent between brochures and consultation records. If versions are not synchronized or signed receipts are missing, basis entries cannot be traced back to the original text. Therefore, the indexing and retrieval stages must first ensure correct recall and merging.
Configuration and implementation checks
This step is executed by the indexing service and vector database, outputting vectors and index status (whether recall is possible and 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). The indexing service and vector database are only responsible for recalling vectors and merging them back to the original data by identifier; they do not generate glossaries, comparison drafts, or version ledgers.
| Object | Check | Verification action |
|---|---|---|
| Enrollment brochure | Names, class types, start dates, fee policies, and version numbers must match consultation records and trial schedules; only the current version can be used as a basis | Check the indexing model in the knowledge base settings to confirm it is consistent with the vector model actually used by the knowledge base; open the brochure data row and check whether the index field records multiple vector identifiers |
| Consultation record | Parents' requests, communication dates, handlers, and receipts must be complete; the names, quantities, and dates of the same record in brochures and trial schedules must match | Sample one consultation record data row and check whether the index field contains all vector identifiers; retrieve that record to see whether data in the same group is merged, with the score taking the highest score |
| Trial class schedule | Trial times, teachers, classrooms, and sign-in signatures must be complete; the version after rescheduling must be synchronized with consultation records | View the data scope covered by the rebuild task and confirm that after changing models, it is triggered only for data whose indexing has been completed; check the status value indexed and the rebuild task records |
Three common errors
- Symptom: duplicate entries appear in retrieval results. Inspect the following: the same group of data corresponds to multiple vector identifiers, and after recall it is not merged back to the original data by identifier, or the highest score is not taken during merging.
- Symptom: recall quality drops after changing models. Inspect the following: the rebuild task only covers completed data; data rows whose indexing is not completed do not participate in recall, so indexing must be completed before rebuilding.
- Symptom: a data row has content but cannot be retrieved. Inspect the following: the index field is missing vector identifiers, or consultation records and brochure versions are not synchronized, or signed receipts are missing, causing basis entries to be unable to trace back to the original text.
How to verify the setup
Acceptance for this feature: sample one data row and look at the index field to confirm that multiple vector identifiers for one piece of data are all present; then retrieve once and see whether the corresponding same group of original data is merged and whether the vector score takes the highest score.
Business test: ask a basis question that spans materials, see whether the answer gives a traceable source, and check the number of citations and the cited text tokens. quoteMaxToken is the citation content budget, and the number of array entries is recorded separately.
- Check the indexing model in the knowledge base settings to confirm it is consistent with the vector model actually used by the knowledge base; open the data row and check whether the index field records multiple vector identifiers and whether the status value is indexed.
- View the data scope covered by the rebuild task and confirm that after changing models, it is triggered only for data whose indexing has been completed; check the configurations related to
limit, and verify that cited text tokens and the number of array entries are recorded separately.