What this category's data looks like
This page addresses the enterprise sales product training scenario, with materials including product manuals, sales response scripts, and channel training schedules. Product manuals are updated quarterly, recording product features, parameters, and version numbers, with fields including name, quantity, date, and signature; sales response scripts are updated monthly, recording response statements and source references, with fields including product name, applicable version, and confirmer; channel training schedules are updated weekly, recording course topics, instructors, and attendance receipts, with fields including date, signature, and version. When these three types of materials enter the vector model and indexing, product manuals provide facts, sales response scripts provide citable statements, and channel training schedules provide time and personnel basis. Material versions must be synchronized; the name, quantity, and date of the same record in the manual and the script must be consistent. If a signature or receipt is missing, it cannot be used as a basis for a response.
Configuration and implementation checks
This step is executed by the indexing service and the vector database, outputting 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 |
|---|---|---|
| Product manual | Name, quantity, date, signature are consistent with the current version, and it can be traced back to the original text | Open a data row, check whether the index field records multiple vector identifiers, and confirm the status is indexed |
| Sales response script | Product name, applicable version, and confirmer signature match the manual record, and the date has not expired | In the knowledge base settings, check that the indexing model is consistent with the vector model actually used by the script materials |
| Channel training schedule | Course date, instructor signature, attendance receipt, and version number are complete, with no missing items | Check the rebuild task records to confirm that after changing the model, only data that has already been indexed is triggered |
| Citation entry | The entry can be traced back to the original text, the array count is recorded separately, and the citation text tokens do not exceed the budget | Sample a basis entry and check limit, and the citation content budget |
Three common errors
- Symptom: The name, quantity, and date of the same product do not match between the manual and the script. Inspect the following: The versions of the two materials are not synchronized.
- Symptom: Duplicate entries appear in retrieval results, and the same record is recalled multiple times. Inspect the following: The index field of the data row does not fully record multiple vector identifiers, or the rebuild task does not cover all data.
- Symptom: The channel training schedule lacks a signature or receipt, and the basis entry cannot be traced back to the original text. Inspect the following: The signature and receipt fields were not checked before the materials were stored.
How to verify the setup
Acceptance for this feature: sample a data row to inspect the index field, confirm that multiple vector identifiers are all present, then retrieve once to see whether the same group is merged and whether the vector score takes the highest score; for status values, only indexed and rebuild task records are recognized.
Business test: ask a basis question that spans materials, see whether the answer provides a traceable source, and check limit, and citation text tokens.
- In the knowledge base settings, check that the indexing model is consistent with the vector model actually used by the knowledge base, and open a data row to check whether the index field records multiple vector identifiers.
- Check the rebuild task records to confirm that after changing the model, only data that has already been indexed is triggered, and check whether the highest score is taken after merging data in the same group.