What this category's data looks like
For this category, materials come from employment agreements, graduation destination registration forms, employer certifications, and receipts, updated by batch. Employment agreements carry the signing parties, position, date, and signature; graduation destination registration forms carry the destination category, employer name, quantity, and registration date; employer certifications carry acceptance opinions and stamped receipts. Multilingual materials and terminology processing require the same term to be consistent throughout the entire text and the translated name to be traceable back to the original text. Therefore, after materials enter document parsing and chunking, each has a different role: agreements and registration forms provide fields and dates, certifications provide signatures and receipts, and the terminology glossary and comparison draft are consolidated during retrieval and subsequent processing stages.
Configuration and implementation checks
Parsing and training tasks (server-side tasks) only output data rows and index status (parsed, indexing, indexed). The unified terminology glossary and comparison draft must be handled next by retrieval and subsequent processing. After parsing is completed, parsed chunks are immediately persisted as visible data rows and bound to a unique data identifier. The index transitions through three states, and data pending indexing does not participate in recall.
| Object | Check | Verification action |
|---|---|---|
| Employment agreement | Fields, dates, and signatures are complete, and the name, quantity, and date match the same record in the registration form | Open the data row to check the status, verify the file record and the data row separately, and confirm that after chunking by chunkSize and chunkSplitter, fields are not misaligned across chunks |
| Graduation destination registration form | The destination category, employer name, quantity, and registration date can be traced back to the original text of the agreement, and versions are synchronized | Sample three terms to check whether their spelling is consistent throughout the entire text, and verify whether trainingType and the index status are parsed / indexing / indexed |
| Employer certification | Acceptance opinions, stamp, and receipt are complete, and the name, quantity, and date match the same record | Verify whether signatures and receipts are persisted along with the materials, and check whether the data rows covered by the rebuild task contain only data with completed indexing |
| Unified terminology glossary and comparison draft | The same term is consistent throughout the entire text, and the translated name can be traced back to the original text | After chunking by chunkSize, sample three terms, and verify their spelling throughout the entire text and the index status |
Three common errors
- Symptom: The name, quantity, and date of the same record do not match between the agreement and the registration form. Cause: Multilingual material terminology is not unified, fields are misaligned across chunks after parsing and chunking, and the index status is not synchronized.
- Symptom: Material versions are not synchronized, and the old registration form is still recalled. Cause: When changing models, only data with completed indexing is rebuilt, and data pending indexing does not participate in the rebuild chain.
- Symptom: The employer certification lacks a signature or receipt. Cause: Successful parsing does not mean retrieval will definitely hit; the index status must also be checked, and failed training may leave empty data rows.
How to verify the setup
Acceptance for this feature: Upload one material, observe the data row status transitions and the coverage scope of the rebuild task, confirm the status is parsed / indexing / indexed, and verify the three identifiers chunkSize, chunkSplitter, and trainingType.
Business test: Sample three terms to check whether their spelling is consistent throughout the entire text, confirm the translated name can be traced back to the original text, and record the number of citations and the tokens of the cited text separately.
- Open the data row to check the status, verify the file record and the data row separately, and confirm that the unique data identifier is not mixed up.
- View the data rows covered by the rebuild task, confirm that only data with completed indexing is covered, and that data pending indexing does not participate in recall.