What this category's data looks like
For the deployment, operations, and application handover of kindergarten play activity organization, materials come from observation records, area plans, material placement lists, activity objectives, and reflections converted into next week's plans. When these materials enter the knowledge base retrieval and recall function, each plays a different role: observation records and area plans are the main sources for semantic recall, material placement quantities and activity objectives participate as structured fields in filtering and reranking, and reflections converted into next week's plans are used for question optimization and expansion. The update cadence advances weekly or by activity cycle, and field characteristics include activity name, area, material quantity, objective description, observation points, and reflection conclusions. During retrieval, these fragments are first recalled as candidates, then undergo merging, reranking, filtering, and limit trimming, ultimately forming reference content that can be handed to subsequent nodes or humans, used to generate handover-ready deployment status and operation records.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after mode, merging, reranking, filtering, and limit trimming). The handover-ready deployment status and operation records are then completed by subsequent nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Observation records and area plans | Whether the behavior descriptions in the observation records correspond to the activity objectives in the area plans, and whether the material quantities meet what the plan requires | In the knowledge base search test, retrieve once each using different phrasings of the same question, view the candidate fragments and similarity scores, and confirm whether the observation records and area plans are recalled at the same time |
| Material placement list | Whether the material placement quantities are consistent with what is needed to achieve the activity objectives, and whether there is insufficient material causing the objectives to be unachievable | Switch searchMode to hybrid retrieval, compare whether both the semantic and full-text candidates contain fragments related to material quantities, and check whether the similarity scores rank near the top |
| Reflections converted into next week's plans | Whether the reflection conclusions have been updated into next week's plans, and whether retrieval still returns old-version fragments when the plans have not been updated | Turn usingReRank on and off and retrieve once each, compare the order of the top few candidates, and confirm whether the reflection and next week's plan fragments are correctly reranked |
| Reference limit trimming | Whether key fragments of the original text are still retained after trimming, and whether the number of reference items and the tokens of the reference text are recorded separately | Check the number of items in the returned reference array, record the limit value separately, and verify whether the tokens of the reference text exceed the budget, without comparing the two as equal |
Three common errors
- Symptom: changing the phrasing of the same question makes the observation records unrecallable. Cause: question optimization does not cover references and expansion, or the retrieval mode does not match the material form.
- Symptom: there are many candidates, but the ones ranked at the front are all irrelevant. Cause: reranking was not enabled after hybrid retrieval, or similarity filtering and reference limit trimming were set improperly.
- Symptom: the answer lacks key paragraphs, materials are insufficient, or objectives and observations do not match. Cause: material placement quantities were not updated synchronously, and reflections converted into next week's plans were not updated, causing the recalled fragments to be inconsistent with the current plan.
How to verify the setup
Acceptance for this feature: test once each with synonym-rewritten question sentences, compare candidates and scores, then turn reranking on and off to inspect the ordering of the top few items.
Business test: check item by item against the handover form once, and see which record is missing.
- In the knowledge base search test, retrieve once each for the same question using two phrasings, record the number of candidates and the similarity scores, and confirm whether the observation records and area plans are hit at the same time.
- Switch
searchModeto semantic, full-text, and hybrid, turnusingReRankon and off respectively, compare the order of the top few candidates, and separately record the number of array items corresponding tolimit.