Multi-Turn Conversation and Prompts for Academic Misconduct Reports and Investigations: Trace Exceptions and Review Results

Research Misconduct: review conversation for exception review. Check source versions, input mappings and the values received by downstream nodes.

What this category's data looks like

This page addresses abnormal traceback and manual review in the investigation and handling of academic misconduct reports. The materials come from three types of original records: report materials, investigation reports, and expert appraisal opinions. Report materials usually contain the reported person, incident description, evidence list, and submission date; investigation reports contain the investigation process, established facts, handling recommendations, and the investigator's signature; expert appraisal opinions contain the appraisal matters, appraisal conclusions, the appraiser's signature, and the issuance date. When the materials enter the multi-turn dialogue and prompts feature, report materials serve as the source of abnormal clues, investigation reports serve as the carrier of process and conclusions, and expert appraisal opinions serve as the basis for professional judgment; together, the three support the generation of the abnormality list and review conclusions. The update cadence advances with the investigation stages, and the field characteristics center on names, quantities, dates, signatures, and receipts, which need to be compared item by item.

Configuration and implementation checks

This step is executed by the memory turns and prompt configuration in the app settings, which output the historical message range referenced and the answer constraints. The "abnormality list and review conclusions" are then completed by the model. Memory turns control how many previous turns of dialogue can be referenced at most when answering; if the value is too small, the preceding text cannot be used, and if the value is too large, irrelevant content is brought in and consumption increases.

ObjectCheckVerification action
Memory turnsMatches the range of preceding text required by the questionCheck the current value in the app settings, and run a trial with two consecutive follow-up questions to observe whether the second turn carries forward the references from the first turn
Report materialsNames, quantities, and dates are consistent item by item with the investigation report and expert appraisal opinionsExtract the names, quantities, and dates of the same record from the two materials, and check whether the signatures and receipts are complete
Investigation reportThe version is synchronized with the expert appraisal opinions, and the signatures and receipts are completeCheck whether the report version number is synchronized with the appraisal opinion version, and inspect the investigator's signature and receipt date
Expert appraisal opinionsThe appraisal matters correspond to the facts established in the investigation report, and the signature and date are traceableCompare the appraisal conclusions with the report conclusions, check the appraiser's signature and issuance date, and confirm that there are no missing items

Three common errors

  • Phenomenon: The names and quantities of the same record do not match between the report materials and the investigation report. Cause: The two were entered separately, without field-level comparison and receipt verification.
  • Phenomenon: The investigation report and expert appraisal opinions are not synchronized in version, and an old conclusion is cited. Cause: After the version was updated, the related materials were not written back, and there was no synchronization confirmation.
  • Phenomenon: The abnormality list lacks the reviewer's signature or receipt. Cause: The review step was not put in writing, and the process node did not retain written evidence.

How to verify the setup

Feature acceptance: Ask two consecutive follow-up questions and see whether the second question carries forward the references from the first question; at the same time, check whether the values of chatId, messages, and variables in the historical message range are consistent with the answer constraints.

Business test: Create an abnormal record, see whether its source material can be traced, and check the names, quantities, dates, and signatures of the report materials, investigation report, and expert appraisal opinions.

  • Check the memory turns value in the app settings, run a trial with two consecutive questions, and compare the usage records of the two runs.
  • Extract the abnormal record and compare item by item the names, quantities, dates, signatures, and receipts of the report materials, investigation report, and expert appraisal opinions.