What this category's data looks like
When material drafting and responsibility handover for peer review organizations enter the knowledge base retrieval and recall function, they usually contain three types of materials: review invitations, expert review comments, and author revision notes. Review invitations record the proposed experts, manuscript titles, invitation dates, and reply status; expert review comments record the review conclusion, comment body, submission date, and signature; author revision notes record point-by-point responses, revision dates, and receipts. In the retrieval step, the three types of materials each serve as sources of candidate fragments to support subsequent drafting and handover. The update rhythm follows manuscript circulation: each version is generated when an invitation is sent, comments are submitted, and revisions are returned. Field characteristics center on names, quantities, dates, version numbers, and signature fields; during retrieval, key fragments of the original text must be retained to facilitate checking whether the same record is consistent across different materials.
Configuration and implementation checks
This step is performed by the knowledge base retrieval service (retrieval node and search test), outputting candidate fragments and similarity scores (after pattern, merge, rerank, filter, and limit trimming). The material drafts and handover records that can be issued are completed by subsequent nodes or manually.
| Object | Check | Verification action |
|---|---|---|
| Review invitation | Fields include proposed experts, manuscript title, invitation date, and reply status; signature or receipt is complete | In the search test, retrieve once each with synonymous questions, compare the number of candidates and scores, and check whether the names, quantities, and dates are consistent with the expert review comments |
| Expert review comments | Fields include review conclusion, comment body, submission date, and signature; the version is synchronized with the review invitation | Switch searchMode to semantic, full-text, or hybrid, see whether the returned candidates include both types of recall results at the same time, then check the signature field |
| Author revision notes | Fields include point-by-point responses, revision date, and receipt; the same record's name and quantity match across the two materials | Compare the ranking of the first few items before and after enabling usingReRank, and check the tokens and array count of the quoted text after similarity filtering |
| Citation limit trimming | After trimming, key fragments of the original text are still retained, and the array count is recorded separately | Check the number and content completeness of returned citations after limit trimming, and confirm that missing items can be blocked |
Three common errors
- Symptom: The same record's name and quantity do not match between the review invitation and the expert review comments. Cause: Candidate fragments come from different recall channels, and drafting begins without verifying hits and associations.
- Symptom: The author revision notes and expert review comments versions are not synchronized, and the dates contradict each other. Cause: After multi-path merging, no rerank secondary sorting was performed, so old-version fragments rank at the front.
- Symptom: The handover record lacks a signature or receipt, and the next step cannot sign for acceptance. Cause: After similarity filtering and citation limit trimming, key fragments are truncated and fields are not completed.
How to verify the setup
Acceptance for this function: test once each with synonymously rewritten questions, compare candidates and scores, then toggle rerank to observe the ranking of the first few items; check the changes in candidate count and scores under searchMode, usingReRank, similarity, and limit.
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
searchModeand compare the candidate fragments and scores returned by semantic, full-text, and hybrid modes. - Enable and disable
usingReRank, compare the ranking of the first few items; then check the tokens and array count of the quoted text aftersimilarityfiltering andlimittrimming.