What this category's data looks like
This category's data includes daily life schedules, morning and afternoon check records, and child health observation records. Sources are mostly filled in by class teachers and childcare workers on duty, updated daily and archived weekly. The daily life schedule provides time slots and activity names; morning and afternoon check records carry entry and midday temperature measurements, physical signs, and abnormal descriptions; child health observation records supplement details such as water intake, meals, naps, and bowel movements. After these three types of materials enter the material verification and discrepancy localization function, the schedule serves as the baseline object, while morning and afternoon check records and health observation records serve as comparison objects. Fields center on child name, activity or check item name, quantity, date, and signature column, used to determine whether the name, quantity, and date of the same object are consistent across the two materials.
Configuration and implementation checks
This part is executed by application settings (response limit, etc.), which output the length of a single answer and the range of context that can be accommodated. The discrepancy list must then be produced by the model; the entry point and authorization are handled by the publishing service and file permissions, and the business output is still answered by the model based on this.
| Object | Check | Verification action |
|---|---|---|
| Daily life schedule | Check whether time slots and activity names, quantities, dates, and signature columns match the ledger item by item | In application settings, check the response limit switch and slider position, confirm that long tables can be output completely, and check the single-answer length corresponding to maxResponse |
| Morning and afternoon check records | Check whether child names, check item names, quantities, dates, and signatures correspond to the schedule | Ask a question that requires a long answer and look at the ending, then compare whether the context range covered by maxContext can hold both materials |
| Child health observation records | Check whether water intake, meal, and nap entry names, quantities, dates, and receipts are consistent with the morning and afternoon check records | Check the tokens of the quoted text, record the number of array entries separately, and treat quoteMaxToken only as the budget for quoted content |
| Discrepancy list | Check whether inconsistent fields land on the specific material, specific row, and specific column | Hand it to the reviewer for confirmation and write-back; before write-back, check once more whether the ending of the answer is complete |
Three common errors
- Symptom: the same child's name is written differently in the two materials, and the dates do not match either. Cause: the two materials were filled in by different shifts, without unified naming and date conventions.
- Symptom: the schedule has been replaced with a new version, but the morning and afternoon check records are still filled in according to the old version's entries. Cause: the version was not synchronized, the old table continued to be used, and discrepancy localization landed on expired fields.
- Symptom: the health observation record lacks a signature or receipt, so the responsible person cannot be confirmed during verification. Cause: the person on duty missed the signature, and the signature column was not checked before archiving.
How to verify the setup
This function's acceptance: ask a question that requires a long answer, check whether the ending is complete, and confirm that the response limit slider position matches the answer length required by the page.
Business test: use one real record to find fields inconsistent with the ledger, verify names, quantities, dates, and signatures item by item, form a discrepancy list, and hand it to the reviewer for confirmation and write-back.
- In application settings, check the response limit switch and slider position, compare usage across two runs, and confirm the change in single-run usage after adjusting the limit.
- Check the tokens of the quoted text and the number of array entries, record them separately, and do not compare the number of quoted entries with
quoteMaxTokenfor equality.