What this category's data looks like
Enterprise independent skill talent evaluation involves three types of materials: enterprise evaluation filing forms, evaluation specifications, and candidate rosters. The filing form is submitted by the enterprise and updated with each batch, containing the evaluation occupation, level, date, and signature fields; the evaluation specification is maintained by the technical department, containing assessment modules, question counts, and score allocations; the roster is compiled by examination affairs, containing candidate names, ID numbers, declared levels, and receipts. The field definitions of the three types of materials are not fully consistent: names have abbreviations and full names, quantities have expected and actual examinees, and dates have filing dates and assessment dates. After the materials enter the knowledge base retrieval service, the filing form and specification serve as basis fragments, and the roster serves as object fragments; each is segmented and participates in recall, for subsequent nodes or manual generation of a discrepancy list.
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 discrepancy list is then completed by subsequent nodes or manually. Retrieval returns candidates and cited content, not statistical results.
| Object | Check | Verification action |
|---|---|---|
| Enterprise evaluation filing form | Name, date, and signature fields need to be compared item by item with the evaluation specification and roster | In the search test, recall using the filing form's occupation name, and check whether the date and signature fields in the candidate fragments are complete |
| Evaluation specification | Assessment modules, question counts, and score allocations need to be consistent with the declared level in the filing form | Switch searchMode to hybrid, and check whether the specification fragments and filing form fragments both enter the candidates |
| Candidate roster | Candidate names, declared levels, and receipts need to correspond to the filing form batch | Recall using the roster names, compare candidate scores, and see whether the names and quantities of the same record match across the two materials |
| Retrieval parameters | The number of candidates and scores need to be reproducible, and the order before and after reranking can be compared | Record the number of candidates, toggle usingReRank to compare the ordering of the top few, and check the tokens of the cited text after similarity filtering |
Three common errors
- Symptom: The same candidate's name or declared level does not match between the roster and the filing form. Cause: The two material versions are not synchronized, or an abbreviation was used during entry.
- Symptom: The filing form date is inconsistent with the evaluation specification assessment date, and the signature field is blank. Cause: Materials are submitted in batches, and receipts are not updated with the batch.
- Symptom: There are many candidates but the ones ranked first are all irrelevant, and key passages are missing. Cause: The retrieval mode does not match the material form, or the similarity filter and cited upper-limit trimming are too tight.
How to verify the setup
This feature acceptance: test once each with synonym-rewritten questions, compare candidates and scores, then toggle reranking to see the ordering of the top few.
Business test: use one real record to find fields inconsistent with the ledger.
- In the search test, switch
searchModeto semantic, full-text, or hybrid, and compare the number of candidates and scores for the same question. - Toggle
usingReRank, check the tokens of the cited text aftersimilarityfiltering, and record the array length separately.