What this category's data looks like
Kindergarten health care and meal management materials include quantified recipes, ingredient procurement orders, sample retention records, disinfection records, absence cause registrations, and parent notification letters. Their sources are the health room, the canteen, and class teachers. They are updated daily and summarized weekly. Fields include ingredient name and specification, sample retention time and batch, disinfection period, and absence cause classification. When entering parsing and training tasks, recipes and procurement orders support terminology extraction and comparison, sample retention and disinfection records provide period and batch fields, absence registrations provide a classification vocabulary, and multilingual materials additionally include comparison drafts, which are used to generate a unified terminology list and comparison drafts.
Configuration and implementation checks
This stage is executed by parsing and training tasks (server-side tasks), outputting data rows and index statuses (parsed, indexing, indexed). "Unified terminology list and comparison drafts" are then completed by retrieval and subsequent processing.
| Object | Check | Verification action |
|---|---|---|
| Quantified recipes and procurement orders | Whether ingredient names and specifications correspond item by item, and whether recipe categories and procurement categories are consistent | Sample three ingredient terms, check their wording throughout the document and the specification field, and check whether chunk boundaries under chunkSize and chunkSplitter cut through terms |
| Sample retention records and disinfection records | Whether sample retention time and batch can be traced back to the document, and whether disinfection periods are missing segments | Compare item by item by batch number and period field, and confirm that the same term is consistent throughout the document and that the translated name can be traced back to the source text |
| Absence cause registrations and multilingual comparison drafts | Whether absence cause classifications are unified, and whether comparison draft terms can be traced back to the source text | Sample three cause terms to check whether their wording is consistent throughout the document, and check the data row status under the processing mode corresponding to trainingType |
Three common errors
- Phenomenon: documents and sample retention batches do not match. Cause: sample retention records are stored by batch, procurement orders are stored by ingredient name, and parsing chunking does not bind the batch field and ingredient term in the same data row.
- Phenomenon: recipe and procurement categories are inconsistent. Cause: recipes use abbreviations, procurement orders use full specification names, the terminology list is not unified, wording differs across segments after chunking, and retrieved hit fragments cannot be traced back to the source text.
- Phenomenon: disinfection records are missing periods. Cause: the disinfection period field is truncated during parsing, data rows are in parsed or indexing status, data pending indexing does not participate in recall, and the retrieval layer is naturally isolated.
How to verify the setup
Acceptance for this feature: upload one material and observe the data row status transitions and the coverage scope of the rebuild task. The status should pass through parsed, indexing, and indexed in sequence. Rebuilding covers only data that has completed indexing; data pending indexing does not participate in rebuilding. At the same time, check the chunking and processing modes corresponding to the three identifiers chunkSize, chunkSplitter, and trainingType.
Business test: sample three terms to check whether their wording is consistent throughout the document.
- Open the data row to check the status, confirm that it is in one of the three states and consistent with the interface display, and separately check the identifiers of the file record and the data row.
- View the data rows covered by the rebuild task, confirm that it targets only data that has completed indexing, and that the initial training covers data pending indexing and data being indexed.