What this category's data looks like
This page addresses the enterprise independent skill talent evaluation scenario. Materials include evaluation filing forms, evaluation specifications, candidate rosters, progress tables, and statistical caliber descriptions. The filing forms and specifications come from the evaluation agency's internal archives, the rosters are submitted by the declaring units, and the progress tables are updated with weekly or monthly reports. Field characteristics center on names, quantities, dates, and signatures: the filing forms record the evaluation occupation and level, the specifications record the assessment method and scoring caliber, the rosters record candidate identity and batch, and the progress tables record the number of records at each stage. When the materials enter the knowledge base retrieval service, each serves as a candidate fragment source, recalled by the retrieval node and search test according to the mode, for subsequent nodes or humans to generate progress tables and statistical caliber descriptions accordingly. Retrieval itself does not produce statistical tables or authorization results.
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 limit trimming). The "progress tables and statistical caliber descriptions" are completed subsequently by later nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Evaluation filing form | Names, dates, and signatures must be consistent with the evaluation specifications; the names and dates of the same record must match across the two materials | In the search test, search once each with the filing form's occupation name and the specification title, verify the name, date, and signature fields in the candidate fragments, and record the number of candidates and the similarity score |
| Candidate roster | Quantities and batches must be consistent with the progress table; entries missing signatures or receipts should be marked separately | Search separately with the roster batch number and the progress table batch number, compare the quantities and dates in the candidate fragments, confirm whether the signature or receipt fields appear, and record the number of candidates and the score |
| Evaluation specification | The version must be synchronized with the filing form; the assessment method and scoring caliber must be written in the header | Switch searchMode to semantic, full-text, and hybrid and test each once, compare the specification version number with the filing form version number, and see whether the candidates include both semantic and full-text results |
| Progress table and statistical caliber description | The record count and statistical count must match; the caliber must be written in the header, and exceptions must be clearly stated | Verify the total with one week's records, compare the ordering of the top few entries before and after enabling usingReRank in the search test, and check the tokens and array length of the cited text after similarity filtering and limit trimming |
Three common errors
- Symptom: The name and quantity of the same candidate do not match across the two materials. Cause: The filing form and roster versions are not synchronized, and the retrieval candidates hit only one of them; when verifying, you must first look at hits and associations.
- Symptom: The progress table date and the evaluation specification date do not match. Cause: The specification was revised but the filing form was not backfilled, and the candidate fragments come from the old version; you must verify the version number and signature receipt.
- Symptom: There are many candidates but the ones ranked at the front are all irrelevant. Cause: The same question phrased differently cannot be recalled, or rerank is not enabled; you must compare the order of the top few entries before and after enabling it in the search test.
How to verify the setup
Acceptance of this feature: Test once each with synonym-rewritten questions, compare the candidates and similarity scores, then toggle rerank on and off to see the ordering of the top few entries; switch searchMode to semantic, full-text, and hybrid, record the number of candidates and scores, and confirm that after limit trimming the tokens and array length of the cited text are recorded separately.
Business test: Verify the total with one week's records, see whether the caliber and exceptions are clearly stated, and confirm that the record count and statistical count match and the caliber is written in the header.
- In the knowledge base search test, search once each with the filing form, roster, and specification titles, and record the number of candidates and the
similarityscore. - Compare the ordering of the top few entries before and after enabling
usingReRank, verify the tokens and array length of the cited text, and confirm thatlimittrimming has not dropped key fragments.