What this category's data looks like
Kindergarten play activity organization involves materials such as observation records, area plans, material placement lists, activity objective tables, and reflection-to-next-week plans. Observation records are generated per activity session, area plans are updated weekly, material placement quantities are registered before the activity, activity objective achievement is marked by teachers after the activity, and reflection-to-next-week plans are summarized on weekends. When these materials enter the knowledge base retrieval service, each plays a different role: observation records and area plans are used to compare whether objectives and behaviors are consistent, material placement lists are used to check whether quantities meet the plan, activity objective tables are used to determine whether achievement descriptions match observations, and reflection-to-next-week plans are used to track whether plans are updated with reflections. During exception tracing, it is necessary to trace back from the exception list to the original records, area plan versions, and material registrations; after the reviewer signs for confirmation, it is handed over to the responsible person for disposition.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, rerank, filter, and limit truncation). The exception list and review conclusions are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Observation records and area plans | Whether the observed behaviors correspond to the objectives in the area plan | Use the same activity objective with two different question phrasings for retrieval, compare candidate chunks and similarity scores, and check the observation records and area plan versions |
| Material placement quantity | Whether the registered material quantity meets the placement requirements in the area plan | Switch searchMode to hybrid in the search test, check whether candidates include both material list and area plan chunks, and record the number of candidates and scores |
| Activity objective achievement | Whether the objective achievement description is consistent with the behaviors in the observation records | Compare the order of the top candidates before and after enabling usingReRank, and check whether the objective table and observation records point to the same activity session |
| Reflection-to-next-week plan | Whether the reflection content has been transferred to the next week's plan and updated the corresponding area | Retrieve once each with synonymous question sentences, check whether candidate chunks cover both the reflection and the next week's plan, record the number of candidates and scores, and check the plan update date |
Three common errors
- Phenomenon: The material pointed to by the exception list cannot find corresponding chunks in the knowledge base. Cause: Insufficient materials, observation records or area plans are not fully ingested, retrieval candidates are empty or similarity scores are low.
- Phenomenon: The review conclusion is inconsistent with the observation record description. Cause: Objectives and observations do not match, and there is a discrepancy between the achievement status marked in the activity objective table and the behavior description in the observation records.
- Phenomenon: During exception tracing, it is found that the area plan has not been updated. Cause: The plan has not been updated, the reflection-to-next-week plan has not synchronously modified the area arrangement, and retrieval still returns old version chunks.
How to verify the setup
Acceptance of this feature: Test once each with synonymously rewritten question sentences, and compare candidate chunks and similarity scores; then toggle rerank on and off, see whether the order of the top candidates changes, and record the number of candidates and scores.
Business test: Create an exception record, see whether the source material can be traced, check whether the observation records, area plans, and material registrations correspond, and after the reviewer signs, hand it over to the responsible person for disposition.
- In the knowledge base search test, switch
searchMode, compare candidate chunks and similarity scores under semantic, full-text, and hybrid modes, and confirm that the material form matches the mode. - Adjust
usingReRankandsimilarity, observe the order of the top candidates and the filtering results, record the number of cited chunks and cited text tokens separately, and do not compare the array length withlimitfor equality.