What this category's data looks like
Laboratory hazardous chemical procurement and ledger materials come from procurement application forms, usage registration forms, inventory ledgers, and supplier accompanying receipts. The update cadence is synchronized with batch and monthly inventory counts, and the fields include product name, specification, quantity, date, user, custodian, approver signature, and receipt number. When entering the knowledge base retrieval and recall function, the procurement application form plays the role of procurement basis, the usage registration form plays the role of usage flow, the inventory ledger plays the role of balance verification, and the receipt plays the role of external confirmation. The material version is based on the current controlled version, and historical versions are for comparison only. The name, quantity, and date of the same record in different materials must be able to match each other, and signatures and receipts must be complete. During retrieval, select semantic, full-text, or hybrid mode according to the material form, so that candidate snippets cover field-level basis and facilitate subsequent reference back to the original text.
Configuration and implementation checks
This step is performed 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). "Basis entries that can be traced back to the original text" are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Procurement application form | Product name, specification, quantity, application date, and approval signature must correspond to the inventory ledger and receipt | In the search test, use searchMode to switch between semantic, full-text, and hybrid, and check the number of candidates and scores |
| Usage registration form | User, usage quantity, usage date, and custodian signature must be consistent with the ledger outbound records | Compare the order of the top few items before and after enabling rerank, and use usingReRank to control secondary sorting |
| Inventory ledger | Balance quantity, inventory count date, and custodian signature must mutually corroborate the procurement application and usage registration | Check whether the candidate snippets after similarity filtering retain the key fields of the original text |
| Supplier receipt | Receipt number, arrival quantity, and date must match the procurement application form and inventory ledger | After limit trimming, record the number of cited entries and the cited text tokens separately |
Three common errors
- Phenomenon: The name, quantity, and date of the same record do not match between the procurement application form and the inventory ledger. Cause: During entry, synchronization was not performed according to a unified field standard, and after retrieval hits, cross-material comparison was lacking.
- Phenomenon: The usage registration form has been updated, but the inventory ledger is still the old version. Cause: The version was not synchronized, retrieval recalled old snippets, and candidates were inconsistent with the current materials.
- Phenomenon: The procurement application form lacks approval signature, and the receipt lacks custodian receipt confirmation. Cause: Signatures or receipts were not entered into the knowledge base with the materials, and retrieval could not recall complete basis.
How to verify the setup
This function's acceptance: test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ordering of the top few items; check the actual behavior of searchMode, similarity, usingReRank, and limit.
Business test: ask a cross-material basis question and see whether the answer provides a traceable source.
- In the knowledge base search test, switch
searchModeand record the number of candidates and scores under semantic, full-text, and hybrid modes respectively. - Enable and disable rerank, compare the order of the top few candidates, and check whether the cited text after
similarityfiltering retains the key snippets of the original text.