What this category's data looks like
This category's data includes review invitations, expert review comments, and author revision notes; each document carries fields, dates, and signatures. A review invitation records the manuscript title, the submission time for review, expert information, and the receipt; expert review comments record the review conclusion, comment items, and the signing date; author revision notes record point-by-point responses, revision locations, and the submission date. Documents enter the knowledge base in batches by manuscript version, and the update cadence follows review rounds and revised manuscript submissions. When entering the knowledge base retrieval and recall function, review invitations, expert review comments, and author revision notes each participate in recall as independent fragment sources, for subsequent nodes or humans to trace anomalies accordingly and generate an anomaly list and review conclusion.
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, merge, rerank, filter, and limit trimming). The anomaly list and review conclusion are then completed by subsequent nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Review invitation | Whether the fields, dates, and signatures correspond to the expert review comments | In the search test, switch searchMode to see whether candidates include both semantic and full-text results, and verify the invitation receipt against the signing date of the review comments |
| Expert review comments | Whether the name, quantity, and date of the same record are consistent between the review comments and the author revision notes | Compare the order of the first few items before and after enabling rerank, verify the number of comment items against the number of response items in the revision notes, and check for missing signatures or receipts |
| Author revision notes | Whether the version is synchronized with the current review round, and whether the revision locations and dates are traceable | Test once each with synonym-rewritten questions, compare candidates and similarity scores, and check whether the cited text after limit trimming retains the key passages |
| Retrieval node and search test | Whether the candidate fragments and similarity scores are sufficient to support anomaly tracing | Record the number of candidates and scores, separately check the tokens of the cited text, and confirm the usingReRank status and ranking changes |
Three common errors
- Symptom: The name, quantity, and date of the same record do not match between the review invitation and the expert review comments. Cause: The two documents were not synchronized by the same manuscript version, and retrieval hit fragments from different rounds.
- Symptom: The author revision notes lack a signature or receipt, and the review conclusion has no original record to support it. Cause: The signing fields were not validated when the documents entered the knowledge base, and the candidate fragments did not cover the signature page.
- Symptom: There are many candidates, but the ones ranked first are all irrelevant, and the answer lacks key passages. Cause: The retrieval mode does not match the document form, rerank is not enabled, or the similarity filter threshold is too broad.
How to verify the setup
This feature's acceptance: test once each with synonym-rewritten questions, compare candidates and scores, then toggle rerank on and off to observe the ranking of the first few items.
Business test: create an anomaly record and see whether the source document can be traced.
- In the knowledge base search test, switch
searchModeto compare the number of candidates andsimilarityscores across the semantic, full-text, and hybrid modes. - Enable and disable
usingReRankto compare the order of the first few items; record the number of citations afterlimittrimming, and separately check the tokens of the cited text.