What this category's data looks like
The data for this category mainly comes from teachers' daily observation records, parent request logs, class notices, and commitment item tables, with an update cadence that follows daily drop-off/pick-up and weekly feedback. Core materials in this category include: draftable materials (such as descriptions of children's performance in kindergarten, activity notices, and reply drafts responding to parent requests) and handover records (including responsible person, version, date, and signature field). When these materials enter the multi-turn dialogue and prompt feature, each plays a different role: historical dialogue provides prior references and context scope, prompts constrain the answer tone and field format, and the memory turn count determines how many turns of chat are referenced. Draftable materials need complete fields, a clear responsible person, and consistent version and date; handover records need to be able to be signed off by the next stage. Common inconsistencies are concentrated in commitment items and response times, descriptions of children's performance not matching observation records, missing classification of parent requests, and unconfirmed notice delivery.
Configuration and implementation checks
This part is executed by the memory turn count and prompt configuration in the application settings, outputting the referenced historical message range and answer constraints. Draftable materials and handover records are then completed by the model. The memory turn count controls how many previous turns of dialogue are referenced at most when answering; too small a value fails to use prior context, while too large a value brings in irrelevant content and increases consumption.
| Object | Check | Verification action |
|---|---|---|
| Memory turn count | Matches the prior context range needed by the question, and can cover references to commitment items and response times | Check the current value in the application settings, run a trial with two consecutive follow-up questions, and observe whether prior context is correctly referenced in messages |
| Draftable materials | Complete fields, clear responsible person, consistent version and date, and descriptions of children's performance matching observation records | Draft one according to the template, and check whether commitment items and response times, classification of parent requests, and notice delivery are missing |
| Handover records | Clear responsible person, consistent version and date, and able to be signed off by the next stage | Complete one handover, and check whether chatId and variables are updated in sync with the material version |
| Consecutive follow-up questions | The second turn correctly inherits the references from the first turn and does not answer off-topic | Run a trial with two consecutive question turns, and confirm that the referent of the second question is consistent with the first question |
Three common errors
- Phenomenon: Commitment items are not responded to in the draft, and the handover record lacks response times. Cause: The prompt does not constrain the commitment and response time fields, and insufficient historical turns cause prior commitments to be omitted.
- Phenomenon: Descriptions of children's performance do not match observation records, and parent requests are not classified. Cause: When drafting materials, the source of observation records is not bound, and the prompt does not require output by request classification.
- Phenomenon: When the second question uses a reference, it answers off-topic, and responses become slower with increased consumption. Cause: The memory turn count is too small to use prior context, or too large and brings in irrelevant content and increases consumption.
How to verify the setup
Acceptance for this feature: with two consecutive follow-up questions, see whether the second question inherits the references from the first question, and check the referenced historical message range and answer constraints in messages.
Business test: draft one according to the template and complete one handover, and see whether missing items are blocked.
- Check the current value of the memory turn count in the application settings, run a trial with two consecutive follow-up questions, and confirm that the second question's references are correct.
- Compare the usage records of two runs, and confirm that the consumption change after adjusting the turn count matches the prior context range needed by the question.