What this category's data looks like
This page addresses the vocational school student internship management scenario, where materials come from three categories: internship agreements, internship plans, and internship logs. Agreements are signed by three parties: the school, the enterprise, and the student; plans are jointly confirmed by the supervising teacher and the academic affairs office; logs are filled in weekly by students and signed off by the mentor. In terms of update cadence, agreements are re-signed when positions change, plans are adjusted per semester, and logs are added week by week. Field characteristics center on names, quantities, dates, and signatures: names must be consistent with the student records database, quantities refer to internship weeks and position headcount, dates cover start and end dates as well as signature dates, and signatures include the student, mentor, supervising teacher, and unit receipt. When materials enter knowledge base retrieval, agreements and plans serve as evidence-type chunks, while logs serve as process-type chunks; each participates in recall as candidates, for subsequent nodes or humans to perform material verification and discrepancy localization.
Configuration and implementation checks
This step is executed by the knowledge base retrieval service (retrieval node and search test), outputting candidate chunks and similarity scores (after pattern, merge, re-rank, filter, and limit truncation). The "discrepancy list" is completed afterward by subsequent nodes or humans.
| Object | Check | Verification action |
|---|---|---|
| Internship agreement | Whether names, quantities, dates, and signatures are consistent with the ledger | In the search test, use synonymous questions to retrieve agreement chunks, record the number of candidates and scores, and verify the signing date and three-party signatures |
| Internship plan | Whether names, quantities, dates, and signatures are aligned with the agreement | Switch searchMode to compare semantic and full-text candidates, and check whether the plan weeks and start/end dates match the agreement |
| Internship log | Whether names, quantities, dates, and signatures are in sync with the plan | Compare the order of the top few entries before and after enabling usingReRank, and check whether the log weeks, mentor signature, and receipt are complete |
| Knowledge base retrieval service | Whether candidate chunks and similarity scores are reproducible | Fix similarity and limit and re-run, record the number of candidates and scores, and confirm that the referenced text tokens do not exceed the budget |
Three common errors
- Symptom: The same student's name or student ID does not match between the agreement and the log. Cause: When the two materials were entered, the student records database was not used as the baseline, and the name field was not uniformly mapped.
- Symptom: The plan states twelve weeks but the log records only ten weeks, so the quantities do not match. Cause: After the plan was adjusted, it was not synced to the log template, and the version was not synchronized, resulting in inconsistent week count definitions.
- Symptom: The agreement lacks the unit receipt or the log lacks the mentor signature. Cause: No signature and receipt validation was set up in the material collection step, and missing items entered retrieval without being identified.
How to verify the setup
Feature acceptance: Test once each with synonymously rewritten questions, compare candidates and scores, then toggle re-ranking on and off to observe the ordering of the top few entries; record the number of candidates and scores, and record the number of citations separately by array length.
Business test: Use one real record to find fields inconsistent with the ledger, locate the specific material and field, then hand it to the reviewer for confirmation and write-back.
- In the search test, switch
searchModeto compare differences in candidate chunks and scores across the semantic, full-text, and hybrid modes. - Fix
similarityandlimitand re-run the same question, and verify the number of candidates and the referenced text tokens budget.