What this category's data looks like
This category's data covers three types of materials: employment agreements, graduation destination registration forms, and employer certifications. Sources are departmental submissions, employment center consolidation, and employer receipts, with update cadence changing on a rolling basis with signing batches, registration windows, and supplementary receipt submissions. Employment agreements focus on employer name, position, signing date, and signatures and seals of both parties; graduation destination registration forms focus on destination category, employer location, report certificate information, and personal confirmation; employer certifications focus on hiring intent, onboarding time, and employer contact receipt. After entering the knowledge base retrieval, the three types of materials each serve as candidate fragment sources, recalled and compared by field, date, and signature dimensions, to support the generation of version ledgers and change records, without directly producing statistical 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 pattern, merge, rerank, filter, and upper-limit trimming). The "version ledger and change records" are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Employment agreement | Whether employer name, position, signing date, and signatures and seals of both parties are complete and consistent | In the search test, switch searchMode and check whether candidate fragments cover the signature and date fields |
| Graduation destination registration form | Whether destination category, employer location, report certificate information, and personal confirmation date match the agreement | Compare semantic and full-text candidates to see whether the name and date of the same record are hit at the same time |
| Employer certification | Whether hiring intent, onboarding time, and employer contact receipt are missing items | Compare the ranking of the top few items before and after enabling usingReRank, and check whether the receipt field is still in the leading positions |
| Version ledger and change records | The current version is unique, old versions are traceable, and references are synchronized | After filtering by similarity, record the number of candidate items and scores, and check whether old-version fragments have been trimmed |
Three common errors
- Symptom: The employer name and signing date of the same record do not match between the employment agreement and the graduation destination registration form. Cause: The two materials are submitted separately and have not been merged and checked against the version ledger.
- Symptom: An employer certification missing a signature or receipt is still recalled as a valid candidate. Cause: Filtering only looks at similarity and does not validate by signature and receipt fields.
- Symptom: After changing one piece of content, the version number is not synchronized, and references still point to the old version. Cause: Change records are not written back to the knowledge base, and retrieval recall returns unupdated fragments.
How to verify the setup
Acceptance of this feature: test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank on and off to see the ranking of the top few items.
Business test: change one piece of content and see whether the version number and references are updated synchronously.
- In the search test, retrieve the same record in semantic, full-text, and hybrid modes respectively, record the number of candidate items and scores, and check whether semantic and full-text results are both included.
- Retrieve once each with rerank enabled and disabled, compare the order of the top few items, confirm that key signature and date fragments have not been trimmed, and separately record the number of items in the reference array and the reference text tokens.