What this category's data looks like
This page addresses the scenario of annual progress and acceptance of research projects, covering three types of materials: annual progress reports, mid-term review materials, and acceptance reports. The sources are submissions by project teams and archiving by management departments, and the update cadence advances according to annual, mid-term, and acceptance milestones. When materials enter this feature, the annual progress report plays the role of recording phased tasks and indicator completion, the mid-term review materials play the role of milestone comparison and rectification explanations, and the acceptance report plays the role of confirming final results and signed receipts. Field characteristics are concentrated in task names, indicator quantities, start and end dates, signatures, and receipts. The same record may have different naming conventions, different quantity calibers, and contradictory dates across different materials. When versions are not synchronized, old-version data is mixed in, and missing signatures or receipts result in incomplete materials. These characteristics determine that material verification requires first completing document parsing and chunking, then landing on data rows and index status, to provide a verifiable draft for subsequent retrieval and discrepancy localization.
Configuration and implementation checks
This step is executed by parsing and training tasks (server-side tasks), outputting data rows and index status (parsed, indexing, indexed). The "discrepancy list" is then completed by retrieval and subsequent processing. After parsing is complete, parsing chunks are immediately persisted as visible data rows and bound to a unique data identifier. The index transitions through the three states parsed / indexing / indexed. Data pending indexing does not participate in recall, and changing models only rebuilds data that has completed indexing.
| Object | Check | Verification action |
|---|---|---|
| Annual progress report | Task names, indicator quantities, start and end dates, and signed receipts must be aligned item by item with the ledger | After upload, check the data row status transitions, inspect chunk granularity according to chunkSize and chunkSplitter, and compare name, quantity, and date fields |
| Mid-term review materials | The dates, quantities, and signatures of milestone tasks and rectification explanations must be consistent with the annual progress report | Separately check file records and data rows, confirm the training mode according to trainingType, and locate field discrepancies for the same record across the two materials |
| Acceptance report | Final result names, quantities, acceptance dates, and receipts must connect with the mid-term review materials | Check whether the data rows covered by the rebuild task contain only data that has completed indexing, and verify fields with missing signatures or receipts |
| Data rows and index status | Is in one of the three states parsed / indexing / indexed and consistent with the interface display | Open the data rows to check status, confirm that data pending indexing does not participate in recall, and that the rebuild scope covers only data that has completed indexing |
Three common errors
- Symptom: newly uploaded materials cannot be found. Cause: successful parsing does not mean retrieval will necessarily hit; the data rows are still in parsed or indexing status, and data pending indexing is naturally isolated at the retrieval layer, so verification must wait until indexing is complete.
- Symptom: the name, quantity, and date of the same record do not match across two materials. Cause: the annual progress report and mid-term review materials are not version-synchronized, old-version data is mixed in, and after parsing they land in different data rows; during discrepancy localization, trace back by unique data identifier.
- Symptom: training failure leaves empty data rows, or missing signatures and receipts are not discovered. Cause: failed training data supports editing or deletion, but when the material itself lacks signatures or receipts, parsing and chunking still produce data rows, which must be confirmed against the ledger during the review step.
How to verify the setup
Feature acceptance: upload one material, observe the data row status transitions and the coverage scope of the rebuild task, confirm that data rows go through the three states parsed / indexing / indexed, that rebuild targets only data that has completed indexing, and that initial training covers data pending indexing; at the same time, check whether the records of chunkSize, chunkSplitter, and trainingType on the data rows and tasks are consistent.
Business test: use one real record to find fields inconsistent with the ledger, compare names, quantities, dates, and signed receipts across the annual progress report, mid-term review materials, and acceptance report, output a discrepancy list for the reviewer to confirm and write back.
- Open the data rows to check status, confirm that the identities of file records and data rows are distinguishable, and check file records and data rows separately.
- Check the data rows covered by the rebuild task, confirm that only data that has completed indexing is covered, and that data pending indexing does not participate in rebuild.