What this category's data looks like
For the kindergarten health care and meal management scenario, daily materials include food procurement orders and acceptance sheets, recipes and quantified recipes, retention sample registration and retention sample batch labels, disinfection period records, absence and cause registration, and shift handover records. These materials are generated daily or per meal, with an update cadence of daily per meal, and their field characteristics center on food name and specification, retention sample time and batch, disinfection period, and absence cause classification. When entering this function, procurement orders and recipes are used to check category consistency, retention sample registration and batch labels are used to trace the source of exceptions, disinfection records are used to fill period gaps, and absence registration is used to classify causes. The exception list and review conclusions are answered by the model based on these materials, the original records serve as the basis for tracing, and the reviewer's signature serves as the condition for the conclusion to hold.
Configuration and implementation checks
This part is executed by the memory turns and prompt configuration in the application settings, and the output references the historical message range and answer constraints. The "exception list and review conclusions" are then completed by the model. Memory turns control how many previous dialogue turns can be referenced at most when answering; if the value is too small, the preceding text cannot be used, and if the value is too large, irrelevant content is brought in and consumption increases.
| Object | Check | Verification action |
|---|---|---|
| Memory turns | Matches the range of preceding text required by the question | Check the current value in the application settings, and run a trial with two consecutive questions to observe whether the second turn can carry forward the references from the first turn |
| Retention sample registration and retention sample batch labels | The retention sample time and batch match the batch on the document | Take one exception record, check the retention sample time, batch, and the batch on the procurement acceptance sheet, and confirm whether the source material can be traced |
| Disinfection period records | The disinfection period covers each step of the day | Check item by item according to the disinfection period, mark the missing periods, and form items pending review |
| Absence and cause registration | The absence cause classification is consistent with the recipe and procurement category | Compare with the recipe and procurement category, and check whether the absence cause classification matches the meal records of the day |
Three common errors
- Phenomenon: During review, it is found that the document and the retention sample batch do not match. Cause: Retention sample registration did not synchronize batch labels by meal, so tracing can only return to the acceptance sheet and cannot return to the specific retention sample.
- Phenomenon: The recipe and procurement category are inconsistent. Cause: After the quantified recipe was adjusted, the procurement order was not synchronized, a gap appeared in the category verification dimension, and the exception list lacks support from original records.
- Phenomenon: The disinfection record is missing periods. Cause: The shift handover record did not cover all disinfection periods, the review conclusion cannot be signed and confirmed period by period, and the disposal destination is unclear.
How to verify the setup
This function's acceptance: ask two consecutive follow-up questions and see whether the second question carries forward the references from the first question; check the referenced historical message range in messages and variables, and record the change in running consumption corresponding to chatId.
Business test: create one exception record, see whether the source material can be traced, and check whether the reviewer's signature and disposal destination are complete.
- Check the current value of memory turns in the application settings, and run a trial with two consecutive questions to confirm that the second turn carries forward the references from the first turn.
- Compare the usage records of two runs, confirm that consumption changes accordingly after adjusting the turns, and check the referenced historical message range.