What this category's data looks like
This page addresses research funding budget adjustment and reimbursement scenarios. The materials come from three categories: budget adjustment applications, reimbursement forms, and contract invoices. The update cadence advances with the budget cycle and reimbursement batches, and the fields include name, quantity, date, signature, and receipt. Budget adjustment applications carry the adjustment reason and quota changes, reimbursement forms carry expense details and the approval flow, and contract invoices carry transaction evidence and invoicing information. When the three enter the document parsing and chunking feature, each is split into retrievable data rows and bound to a unique data identifier; the name, quantity, and date of the same record across multiple materials must remain consistent. The version ledger and change records are completed by retrieval and subsequent processing, ensuring that the current version is unique, old versions are traceable, and references stay in sync.
Configuration and implementation checks
Parsing and training tasks (server-side tasks) only output data rows and index status (parsed, indexing, indexed). The version ledger and change records must be continued by retrieval and subsequent processing. After parsing is complete, parsed chunks are immediately persisted as visible data rows and bound to a unique data identifier. The index transitions through the three states of parsed, indexing, and completed. Data pending indexing does not participate in recall. Changing the model only rebuilds data whose indexing is already completed.
| Object | Check | Verification action |
|---|---|---|
| Budget adjustment application | The name, quantity, date, and signature must match the reimbursement form and contract invoice item by item; a missing signature or receipt is considered incomplete | Open the data row and check whether the status is parsed / indexing / indexed, and verify the file record and the data row separately |
| Reimbursement form | When the name, quantity, and date of the same record do not match across the three materials, the current version prevails and the change is recorded | Verify the data rows covered by the rebuild task, and confirm that only data whose indexing is already completed is covered |
| Contract invoice | When the version is not in sync, the old version is traceable but cannot be recalled, and references must point to the current version | After checking the chunkSize and chunkSplitter settings, verify the tokens and the array count of the reference text and record them separately |
| Data row | It is in one of the three states and is consistent with what the interface displays; failed training data can be directly edited or deleted | Verify trainingType and the index status transitions, and confirm that data pending indexing does not participate in recall |
Three common errors
- Symptom: newly uploaded materials cannot be found. Cause: the data row is in the pending-indexing state, and the retrieval layer naturally isolates it so it is not recalled; you must wait for indexing to complete.
- Symptom: the name, quantity, and date of the same record do not match between two materials. Cause: the version is not in sync, the old version is traceable but references still point to the old version, and the signature or receipt is missing.
- Symptom: empty data rows are left after training fails. Cause: the initial training covers data that is pending indexing and being indexed, while rebuild only targets data whose indexing is already completed; the two cannot be interchanged.
How to verify the setup
This feature's acceptance: upload one material, observe the data row status transitions and the coverage scope of the rebuild task, confirm that the three states parsed / indexing / indexed are consistent with what the interface displays, and that rebuild only covers data whose indexing is already completed.
Business test: change one piece of content, see whether the version number and references are updated in sync, and verify the data rows and index status corresponding to the three identifiers chunkSize, chunkSplitter, and trainingType.
- Open the data row and check whether the status is parsed / indexing / indexed, and verify the file record and the data row separately
- View the data rows covered by the rebuild task, and confirm that only data whose indexing is already completed is covered, and that data pending indexing does not participate in recall