What this category's data looks like
This category's data looks like this: materials for laboratory hazardous chemical procurement and ledgers include three types: procurement applications, issuance registrations, and inventory ledgers. Procurement applications record the product name, specification, quantity, application date, and approval signature; issuance registrations record the recipient, issued quantity, issuance date, and custodian signature; inventory ledgers record the balance quantity, inbound and outbound dates, and review signature. The material update rhythm is per-transaction registration and archiving by batch, and the version is based on the signed receipt. When entering knowledge base retrieval and recall, procurement applications and issuance registrations serve as the main sources for semantic recall and full-text recall, while inventory ledgers serve as a supplementary source for checking quantities and dates. Together, the three types of materials support missing-item identification and follow-up questions.
Configuration and implementation checks
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 trimming). Intake records and the completed follow-up questions are then handled by subsequent nodes or manually. The retrieval mode switches among semantic, full-text, or hybrid, matching the material form.
| Object | Check | Verification action |
|---|---|---|
| Procurement application | Whether the product name, specification, quantity, application date, and approval signature are complete, and whether they are consistent with the name, quantity, and date in the issuance registration | In the search test, test once each with synonymous questions, and switch searchMode to compare the number of candidates and scores |
| Issuance registration | Whether the recipient, issued quantity, issuance date, and custodian signature are complete, and whether they match the inbound and outbound dates in the inventory ledger | Compare the order of the top few items before and after enabling rerank, and check the impact of usingReRank on candidate ranking |
| Inventory ledger | Whether the balance quantity, inbound and outbound dates, and review signature are complete, and whether the version is synchronized with the procurement application | Check the number and content completeness of returned references, and see whether key snippets are still retained after limit trimming |
| Knowledge base retrieval service | Candidate snippets and similarity scores are output after mode, merge, rerank, filter, and limit trimming | Record the number of candidates and scores, separately record the length of the reference array, and do not compare the number of items as equal to the reference content budget |
Three common errors
- Symptom: the name, quantity, and date of the same record do not match between the procurement application and the issuance registration. Cause: the two materials are registered separately, without field-level verification.
- Symptom: the inventory ledger version is not synchronized and still uses the old batch balance quantity. Cause: the update rhythm is inconsistent, and the signed receipt is not archived with the batch.
- Symptom: the issuance registration lacks the custodian signature or receipt, and the missing item is not identified. Cause: the intake step answers before asking follow-up questions, and the material basis is insufficient.
How to verify the setup
Acceptance of this function: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank on and off to see the ranking of the top few items.
Business test: ask a question missing key information and see whether it asks a follow-up question before answering.
- In the knowledge base search test, switch
searchModeand compare the number of candidates and scores under the semantic, full-text, and hybrid modes. - Record the number of candidates and scores, separately record the length of the reference array, and check whether the tokens of the reference text after
limittrimming retain key snippets.