What this category's data looks like
This page centers on the primary and secondary school mental health screening and referral scenario. Material sources include psychological screening plans, interview records, referral forms, and multilingual reference materials, with an update cadence synchronized with semester screening batches and referral receipts. Psychological screening plans provide scale dimensions and decision thresholds, interview records carry process descriptions and observation points, referral forms record destination, date, and signature, and multilingual materials carry terminology translations and source-text back-references. After these materials enter parsing and training tasks, they first go through document parsing and chunking to form retrievable data rows, then flow through the three index states, for retrieval and subsequent processing to generate a unified terminology glossary and comparison draft. At the field level, attention must be paid to names, quantities, dates, signatures, and receipts, ensuring the same term is consistent throughout the whole text and that translations can back-reference the source text.
Configuration and implementation checks
This step is executed by parsing and training tasks (server-side tasks), outputting data rows and index states (parsed, indexing, indexed). The "unified terminology glossary and comparison draft" is then completed 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 flows through the three states, data pending indexing does not participate in recall, and changing models only rebuilds data that has completed indexing.
| Object | Check | Verification action |
|---|---|---|
| Psychological screening plan | Fields contain scale dimensions, decision thresholds, date, and signature; the name and quantity in the same record must match those in the interview record | Check whether the name, quantity, and date in the screening plan and interview record are consistent, check whether the signature field and receipt are complete, and confirm that the data row is in one of parsed / indexing / indexed |
| Interview record | Fields contain process description, observation points, date, and signature; terminology translations must be able to back-reference the source text | Extract three terms and check whether their spelling is consistent throughout the whole text, check whether the name, quantity, and date of the record in the two materials fail to match, and inspect the chunk boundaries corresponding to chunkSize and chunkSplitter |
| Referral form | Fields contain destination, date, signature, and receipt; the version must be synchronized with the screening plan | Check whether the versions of the referral form and screening plan are synchronized, check for missing signatures or receipts, and confirm the training mode and index state corresponding to trainingType |
| Data row | Is in one of the three states and consistent with the interface display; the identities of the file record and the data row are kept separate | Open the data row to check its state, check the file record and the data row separately, and check whether the data rows covered by the rebuild task contain only data that has completed indexing |
Three common errors
- Symptom: the name, quantity, and date of the same record fail to match between the screening plan and the interview record. Cause: multilingual material terminology is not unified, and a whole-text consistency check is not performed after parsing and chunking.
- Symptom: the referral form version is not synchronized, and the old-version receipt is missing. Cause: the material update cadence is out of step with the screening batch, and the signature and receipt fields are not included in the check.
- Symptom: newly uploaded materials cannot be found. Cause: the data row is in the pending-indexing state and has not entered completed indexing, so the retrieval layer naturally isolates it and it is not recalled.
How to verify the setup
This feature's acceptance: upload one material, observe the state transitions of the data row and the coverage scope of the rebuild task, confirm that the state only changes among parsed / indexing / indexed, that the rebuild only covers data that has completed indexing, and check the chunking and training mode corresponding to chunkSize, chunkSplitter, and trainingType.
Business test: extract three terms and check whether their spelling is consistent throughout the whole text, confirm that translations can back-reference the source text, that the name, quantity, and date of the same record match between the two materials, and that signatures and receipts are complete.
- Open the data row to check its state, confirm it is consistent with the interface display, and check the file record and data row identifiers separately.
- View the data rows covered by the rebuild task, confirm that they contain only data that has completed indexing, and that data pending indexing does not participate in the rebuild.