What this category's data looks like
Common materials for social evaluation of vocational skill level certification include the evaluation implementation plan, candidate registration form, and assessor dispatch form. The evaluation implementation plan is usually updated by batch or occupational direction, and contains the occupation name, level, assessment modules, schedule, and signature/seal page; the candidate registration form is summarized by batch, and contains name, ID number, declared level, years of work experience, and signature field; the assessor dispatch form is generated by session, and contains assessor name, certificate number, dispatch date, session, and receipt. After these three types of materials enter the knowledge base, they respectively take on the roles of rule basis, personnel qualification basis, and on-site execution basis: the implementation plan provides the evaluation framework and process requirements, the registration form provides the candidate's individual conditions and declaration records, and the dispatch form provides the assessor's attendance and signed receipt. The key point of field alignment is whether the same record can be mutually corroborated in name, quantity, date, and signature, and whether the version is synchronized to the current batch.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, re-rank, filter, and limit trimming). The basis entries that can be traced back to the original text are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Evaluation implementation plan | Fields include occupation name, level, assessment modules, date, and signature/seal page; the version must be the current batch | In the search test, use synonymous questions to retrieve, check whether the candidate chunks trace back to the original clauses of the implementation plan, and record the number of candidate chunks and scores |
| Candidate registration form | Fields include name, declared level, years of work experience, and signature field; the name and date of the same record must be consistent with the implementation plan batch | Switch searchMode to hybrid, compare semantic and full-text candidates, and check whether the registration form chunks can match the implementation plan clauses |
| Assessor dispatch form | Fields include assessor name, certificate number, dispatch date, session, and receipt; missing signature or receipt means inconsistency | Compare the order of the top few results before and after enabling usingReRank, and check whether the dispatch form chunks retain the date and signature fields |
| Retrieval node | Candidates are returned after multi-path merge, re-rank, similarity filtering, and limit trimming; the number of chunks and scores can be recorded | Adjust similarity and limit, observe changes in the number of candidate chunks and scores, and confirm that the tokens of the cited text and the number of array entries are recorded separately |
Three common errors
- Symptom: the same candidate name is written differently in the registration form and the implementation plan. Cause: the name convention was not unified during data entry, so after cross-material retrieval hits, it cannot trace back to the same record.
- Symptom: the dispatch form date does not match the implementation plan batch date. Cause: the version was not synchronized, and old-version materials are still in the knowledge base participating in recall.
- Symptom: the assessor dispatch form is missing a signature or receipt. Cause: the scanned copy is incomplete, and the candidate chunks returned by retrieval cannot prove that the on-site execution basis is complete.
How to verify the setup
Feature acceptance: test once each with synonymously rewritten questions, compare candidates and scores, then toggle re-rank on and off to check the ordering of the top few results.
Business test: ask a cross-material basis question and see whether the answer gives a traceable source.
- In the knowledge base search test, switch
searchModeand compare the number of candidate chunks and scores across semantic, full-text, and hybrid modes. - Adjust
similarityandlimit, check the tokens of the cited text and the number of array entries, and record the ranking differences before and after enablingusingReRank.