What this category's data looks like
In the research ethics review scenario of material drafting and responsibility handover, common materials include ethics review applications, informed consent form templates, risk assessment forms, and accompanying handover records. Material sources are mostly institutional template libraries, historical review files, and regulatory entries, with update cadence following institutional revisions and template version changes. Field characteristics center on subject names, sample or subject counts, study start and end dates, version numbers, and signature and receipt fields. When entering the knowledge base retrieval service, the ethics review application assumes the role of the primary retrieval object, the informed consent form template assumes the role of clause comparison, the risk assessment form assumes the role of risk item verification, and the handover record assumes the role of tracing version and signing status. The retrieval service only produces candidate snippets and similarity scores; material drafts and handover records are generated by subsequent nodes or manually based on these, and completeness of fields and clarity of responsible persons must be determined in the next stage.
Configuration and implementation checks
This stage is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate snippets and similarity scores (after mode, merge, re-rank, filter, and limit trimming). "Signable material drafts and handover records" are then completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Ethics review application | Name, count, date, and signature fields must be consistent with the informed consent form template | In the search test, retrieve once each with synonymous questions, verify whether candidates cover the application fields after switching searchMode, and record the number of candidates and scores |
| Informed consent form template | Version number and signing receipt must be synchronized with the risk assessment form | Compare the top few rankings before and after enabling usingReRank, check whether template clause snippets are recalled, and verify the key original text snippets retained after similarity filtering |
| Risk assessment form | Risk items and subject counts must match the ethics review application | Use hybrid retrieval to recall risk items, check whether returned candidates contain both semantic and full-text results, and record the number of candidates and scores |
| Handover record | Responsible person, version, and date must be consistent with the materials handed over | Retrieve handover record snippets, verify whether signature and receipt fields appear in candidates; the citation limit only checks citation text tokens, and array item counts are recorded separately |
Three common errors
- Symptom: The same subject name does not match across two materials. Cause: Candidates come from different recall channels and are not merged and verified, with name and count fields each forming separate segments.
- Symptom: Template versions are not synchronized, and old-version clauses are still recalled. Cause: The knowledge base is not updated with institutional revisions, and retrieval is not filtered by version.
- Symptom: The handover record lacks a signature or receipt. Cause: Retrieval only returns candidate snippets and scores; signing status must be completed by subsequent nodes or manually in the material draft.
How to verify the setup
Feature acceptance: Test once each with synonymously rewritten questions, compare candidates and scores, then toggle re-ranking to observe the top few rankings.
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
searchMode, compare the number of candidates and scores across semantic, full-text, and hybrid modes, and record the retained snippets aftersimilarityfiltering. - Toggle
usingReRankto compare the order of the top few, and verify the tokens of the citation text and the array item count afterlimittrimming, confirming that key passages are still within the candidates.