What this category's data looks like
Enterprise compliance and safety training involves three types of materials: compliance training courseware, safety operating procedures, and exam records. Courseware is usually updated annually or quarterly, with fields including course name, applicable position, version number, and release date; operating procedures are divided into volumes by equipment or operation type, with fields including procedure number, effective date, revision record, and approval signature; exam records are generated by session, with fields including personnel name, department, exam paper version, score, invigilation signature, and receipt. When the three types of materials enter knowledge base retrieval and recall, their roles differ: courseware and procedures serve as the source of candidate fragments to be retrieved, while exam records serve as the original evidence for verifying progress and statistical criteria. In terms of update cadence, courseware is revised as regulations change, procedures are re-versioned as equipment changes, and exam records are archived weekly or monthly. Field characteristics require that names, quantities, dates, and signatures can be located in the candidate fragments returned by retrieval, so that subsequent nodes or humans can use them to generate progress tables and statistical criteria descriptions.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after mode, merge, rerank, filter, and upper-limit trimming). The progress table and statistical criteria description are then completed by subsequent nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Compliance training courseware | Fields include course name, applicable position, version number, and release date; whether the name and version date of the same courseware are consistent across two materials | In the knowledge base search test, switch searchMode, search using a combination of courseware name and position, and compare the version number and release date in the candidate fragments |
| Safety operating procedures | Fields include procedure number, effective date, revision record, and approval signature; whether the record is still recalled when the signature or receipt is missing | Search by procedure number, check whether the candidate fragments contain the signature column and receipt field, and record the number of candidate items and similarity scores |
| Exam records | Fields include personnel name, department, exam paper version, score, and invigilation signature; whether the quantity and date of the same session match across two materials | Search using a combination of session and department, verify the scores and signatures in the returned candidates, and compare the ranking of the top few items before and after enabling usingReRank |
| Progress table and statistical criteria description | Record counts and statistical counts match, and the criteria are written in the header; whether exceptional cases are listed separately | Verify the totals using one week of records, check whether the criteria and exceptions are clearly written, and confirm that the reference text tokens retained after similarity filtering and the number of array items after limit trimming are recorded separately |
Three common errors
- Phenomenon: The name, quantity, and date of the same exam record do not match between the progress table and the statistical description. Cause: The two materials are taken from different recall channels, and hits and associations have not been verified.
- Phenomenon: The version of compliance training courseware is not synchronized, and both the old and new versions appear in the candidate fragments. Cause: The courseware version in the knowledge base has not been updated according to the release date, and retrieval still recalls the old version.
- Phenomenon: Safety operating procedures lack signatures or receipts but are still counted as completed during statistics. Cause: The candidate fragments returned by retrieval do not include the signature column, and subsequent nodes or humans have not verified the original evidence.
How to verify the setup
Acceptance for this feature: Test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank to observe the ranking of the top few items; switch searchMode to observe the differences in candidates among semantic, full-text, and hybrid modes, and record the number of candidates and scores after similarity filtering.
Business testing: Verify the totals using one week of records, and check whether the criteria and exceptions are clearly written; check whether the headers of the progress table and statistical criteria description specify the record counts and statistical counts, and whether exceptional cases are listed separately.
- In the knowledge base search test, retrieve using two phrasings of the same question separately, compare the candidate fragments and similarity scores, and confirm whether referent resolution and question expansion are effective.
- Enable and disable
usingReRank, compare the ranking of the top few candidates; then adjustlimitandsimilarity, record the reference text tokens and the number of array items separately, and verify the totals in the progress table and statistical criteria description.