What this category's data looks like
In the enterprise sales product training scenario, product manuals, sales scripts, and channel training schedules are the three core types of materials. Product manuals carry product names, specification parameters, version numbers, and release dates, and are the baseline source for sales presentations; sales scripts record recommendation logic, objection handling, and closing actions, with fields including sales script number, applicable product, revision date, and author signature; channel training schedules list course names, start dates, instructors, participating channels, and receipt status. The three types of materials update at different rhythms: manuals follow product versions, sales scripts are revised quarterly, and schedules roll by training batch. When materials enter the knowledge base retrieval and recall function, manuals and sales scripts mainly serve as candidate sources for semantic and full-text recall, while schedules are used to verify training batches and receipts. Retrieval only provides candidate snippets and similarity scores; the exception list and review conclusions are generated by subsequent nodes or humans based on these.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate snippets and similarity scores (after mode, merge, rerank, filter, and limit truncation). The "exception list and review conclusions" are then completed by subsequent nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Product manual | Fields, dates, and signatures: product name, specification quantity, and version release date must be cross-checkable against sales scripts and schedules | In the search test, use the same question to switch searchMode between semantic, full-text, and hybrid, and record the number of candidates and scores |
| Sales script | Whether the name, quantity, and date of the same record match across the two materials, and whether versions are synchronized | Retrieve once each with synonym-rewritten questions, and compare the ranking of the top few results with usingReRank enabled and disabled |
| Channel training schedule | Whether schedule records missing signatures or receipts can be retrieved and traced back to the source material | Check whether the returned candidates include both semantic and full-text sources, and record the number of candidates after similarity filtering |
| Retrieval node and search test | Whether candidate snippets and similarity scores have undergone mode, merge, rerank, filter, and limit truncation | Verify the tokens of the cited text after limit truncation, and record the array count separately |
Three common errors
- Symptom: The name, quantity, and date of the same product do not match across the two materials. Cause: The manual and sales script are revised separately, without field-level cross-checking.
- Symptom: The sales script has been updated, but the schedule still references the old version, and the training batch is out of sync with the manual version. Cause: Versions are not synchronized, and there is no unified revision record.
- Symptom: The schedule lacks instructor signatures or training receipts, and exception records cannot be traced back to the source material. Cause: Receipt collection is incomplete and not bound to the schedule.
How to verify the setup
Acceptance of this feature: Test once each with synonym-rewritten questions, compare the number of candidates and scores, then toggle rerank to observe the ranking of the top few results, and record the differences in the values of searchMode, similarity, and usingReRank.
Business test: Create an exception record, see whether it can be traced back to the source material, and verify whether the fields, dates, and signatures of the product manual, sales script, and channel training schedule are consistent.
- In the knowledge base search test, switch retrieval modes, record the number of candidates and scores, and check whether the returned candidates include both semantic and full-text sources.
- Compare the order of the top few results before and after enabling rerank, check the tokens of the cited text after similarity filtering and citation limit truncation, and record the array count separately.