What this category's data looks like
This page addresses the enterprise compliance and safety training scenario, and the materials include three categories: compliance training courseware, safety operating procedures, and exam records. Courseware is updated quarterly and contains fields such as course name, applicable position, effective date, author, and approval signature; operating procedures are divided into volumes by equipment or process and contain version number, revision date, operating steps, and a receipt column; exam records are archived by session and contain the examinee, question version, score, and invigilator signature. When these three categories of materials enter the knowledge base retrieval and recall function, courseware and procedures mainly serve as candidate sources for semantic and full-text recall, while exam records are used to compare whether the names, quantities, and dates of the same session in the courseware and procedures are consistent, and to check whether version numbers are synchronized and whether signatures or receipts are missing.
Configuration and implementation checks
This step is performed by the knowledge base retrieval service (retrieval node and search test), which outputs candidate chunks and similarity scores (after pattern, merge, re-rank, filter, and upper-limit truncation). The "exception list and review conclusion" is completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Compliance training courseware | The course name, applicable position, effective date, and approval signature must be consistent with the exam records | Retrieve once each with synonymous questions, check the course name, date, and signature fields in the returned candidates, and record the number of candidates and scores |
| Safety operating procedures | The version number and revision date must be synchronized with the question version used in the exam records | Switch searchMode to semantic, full-text, and hybrid, compare whether the candidates include both types of sources at the same time, and check the version number and receipt column |
| Exam records | The name, quantity, and date of the same session must match across the two materials | Turn usingReRank on and off, observe changes in the ranking of the top few results, then filter by similarity and check the scores and key chunks of the original text |
| Retrieval node and search test | The number of candidates and scores are reproducible, and key chunks of the original text are still retained after upper-limit truncation | Record the number of candidates before and after limit truncation, count the tokens of the cited text separately, and do not compare it for equality with the array length |
Three common errors
- Symptom: The course name and date of the same session do not match between the courseware and the exam records. Cause: The two materials are updated separately without cross-comparison, and although retrieval candidates are hit, the fields are not checked one by one.
- Symptom: The version number of the operating procedure is inconsistent with the question version used in the exam records. Cause: The versions are not synchronized, and the old version of the procedure is still recalled from the knowledge base as a candidate source.
- Symptom: The review conclusion lacks a signature or receipt. Cause: The exception list does not include the original records, the reviewer cannot confirm the source, and the signature step is skipped.
How to verify the setup
Acceptance of this feature: test once each with synonymously rewritten questions, compare candidates and scores, then turn re-ranking on and off to observe the ranking of the top few results; switch searchMode to observe whether semantic and full-text candidates are merged, and record the number of items after filtering by similarity.
Business test: create an exception record, see whether the source material can be traced, and check whether the names, dates, versions, and signatures of the courseware, procedures, and exam records are consistent.
- In the knowledge base search test, switch
searchModeand compare the number of candidates and scores of the semantic, full-text, and hybrid modes to confirm the multi-path merged results. - Retrieve once each with
usingReRankturned on and off, compare the ranking of the top few results, then filter bysimilarityand record the number of cited items afterlimittruncation.