What this category's data looks like
This page focuses on the kindergarten home–school co-education communication category. The materials come from daily class records, parent messages, and notification receipts, and the update cadence rolls with daily communication. Field characteristics include committed items and response times, descriptions of children's performance, classification of parent requests, and notification delivery status. When these materials enter the multi-turn dialogue and prompt features, each plays a different role: committed items and response times are used to determine whether a commitment has gone unanswered; descriptions of children's performance are used to compare whether the statements match the observation records; classification of parent requests is used to check whether feedback has been categorized; notification delivery is used to confirm whether information has reached the recipient. Exception tracing and manual review are carried out around these materials, forming an exception list and review conclusions, which are handed to the person in charge for handling.
Configuration and implementation checks
This part is executed by the memory turns and prompt configuration in the app settings, and outputs the referenced historical message range and answer constraints. The "exception list and review conclusions" are then completed by the model. Memory turns control how many previous turns of dialogue are referenced at most when answering. If the value is too small, prior context cannot be used; if the value is too large, irrelevant content is brought in and consumption increases.
| Object | Check | Verification action |
|---|---|---|
| Committed items and response times | Whether the commitment received a response within the agreed time | In the app settings, check the memory turns, run a trial with two consecutive question turns, check whether the second turn can carry forward the references from the first turn, and inspect the commitment and response records in messages |
| Descriptions of children's performance | Whether the statements match the observation records | Compare the descriptions in messages with the original observation records, confirm whether there are inconsistencies, and record chatId for traceability |
| Classification of parent requests | Whether the feedback has been categorized | Check whether the request classification field in variables is complete; items that are not categorized are added to the exception list |
| Notification delivery | Whether the notification has been delivered and a receipt is retained | Check the notification delivery records and receipts; those missing receipts are added to the exception list and handed to the person in charge for handling |
Three common errors
- Symptom: a commitment goes unanswered. Cause: the memory turns are too small, the prior commitment is not referenced, and the answer does not carry forward the earlier agreement, causing the committed item to be lost in subsequent dialogue.
- Symptom: statements do not match the observation records. Cause: the prompt does not constrain the basis for the answer, the model references irrelevant historical messages, and brings inaccurate descriptions into the answer.
- Symptom: feedback is not categorized. Cause: the prompt does not require output by request classification, the classification information is scattered after multiple turns of dialogue, and the exception list lacks the classification field.
How to verify the setup
Feature acceptance: ask two consecutive follow-up questions, see whether the second question carries forward the references from the first question, and check the referenced historical message range and answer constraints in messages.
Business test: create an exception record, see whether the source material can be traced, and confirm that the exception list and review conclusions are supported by original records and signed by the reviewer.
- In the app settings, check the current value of memory turns and match it with the prior context range required by the question.
- Run a trial with two consecutive question turns, check whether the second turn correctly carries forward the references from the first turn, and compare the usage records of the two runs.