What this category's data looks like
This category's data revolves around four types of materials: correspondence between observation records and area plans, material placement quantities, achievement of activity objectives, and reflection-to-next-week plan. Observation records come from on-site shorthand notes by the lead teacher, area plans come from the grade group's pre-week drafting, material placement lists come from the caregiver's requisition registration, and reflection-to-next-week plans come from weekend teaching and research consolidation. The update rhythm is same-day supplementary recording after the activity, finalization before the week, and rolling revision on weekends. Field characteristics include activity name, area, observation subject, material name and quantity, objective items, responsible person, version number, date, and handover signature slot. When entering the knowledge base retrieval and recall function, observation records and area plans serve as the main material for semantic recall, material placement lists and handover records serve as supplements for full-text recall, and reflection-to-next-week plans serve as context for reranking and filtering; each plays the role of a candidate source and does not directly generate a signable finished product.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate snippets and similarity scores (after mode, merge, rerank, filter, and upper-limit trimming). Signable material drafts and handover records are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Corresponding materials between observation records and area plans | Whether the observation subject, area, and objective items can be matched item by item, and whether the date and version number are consistent | In the search test, use synonymously rewritten questions to search once each, record the number of candidates and scores, and check whether the returned snippets cover the observation subject and objective items; switch searchMode to compare candidate differences among semantic, full-text, and hybrid options |
| Material placement quantity list | Whether the material names and quantities match the actual usage in the observation records, and whether shortages are marked | Use material name plus area as the search term, and check whether candidate snippets contain both semantic and full-text results; verify tokens for quoted text, record the number of array items separately, and when an upper limit is involved, check whether the key original snippets are retained after limit trimming |
| Achievement of activity objectives and reflection-to-next-week plan | Whether the objective items are consistent with the observation conclusions, and whether the reflection items have been transferred into the next week's plan and the plan has been updated | Search once before and after enabling usingReRank, and compare whether the top few rankings are closer to the objective items; use similarity to filter low-scoring candidates, and confirm that snippets where the plan has not been updated or the objectives do not match the observations can be recalled |
Three common errors
- Phenomenon: The quantities in the material placement list do not match the actual usage in the observation records, and shortages are only discovered at handover. Cause: The list was not updated in sync with the supplementary recording after the activity, and retrieval only recalled old-version snippets.
- Phenomenon: The activity objective items are inconsistent with the observation conclusions, and the draft is still returned after signing. Cause: Objectives and observations belong to different sources and were not merged and reranked in the same retrieval.
- Phenomenon: The reflection-to-next-week plan has not been updated, and the handover record still references last week's version. Cause: The plan update rhythm and the retrieval recall timing are offset, and candidate snippets do not cover the latest revision.
How to verify the setup
This function's acceptance: test once each with synonymously rewritten questions, compare the number of candidates and similarity scores, then toggle reranking to inspect the top few rankings; switch searchMode and check the impact of similarity, limit, and usingReRank on candidates.
Business test: draft one according to the template and complete one handover, and see whether missing items are blocked.
- In the knowledge base search test, switch retrieval modes, record the number of candidates and scores, and check whether semantic and full-text results appear at the same time.
- Compare the order of the top few items before and after enabling reranking, check the quoted text tokens and the number of array items, and confirm that key snippets have not been trimmed off.