What this category's data looks like
Graduate admissions interviews and selection use interview plans, candidate score tables and adjustment lists. The plans provide the selection rules; score tables and adjustment lists provide candidate identifiers, names, initial and interview scores, totals, ranks, dates and signatures. Materials are updated by admissions batch. Compare the same candidate across the three sources and retain the version used in each conversation. Memory settings determine which earlier messages reach the model, while the published entry and its authentication settings determine access to these materials.
Configuration and implementation checks
Conversation memory and prompt settings control the historical message range and answer instructions. Publishing and authentication settings control the external entry and access scope. Verify those controls with a restricted account before checking what materials reach the conversation.
| Object | Check | Verification action |
|---|---|---|
| Re-examination plan | Names, counts and dates match the candidate score table and adjustment list | Check the configured memory turns and use two consecutive questions to compare plan clauses with the list records |
| Candidate score table | The same record's name, count and date agree across the plan and adjustment list | Use messages and business variables for two follow-up questions, referring back to the score field in the second question |
| Adjustment list | Versions, signatures and acknowledgments correspond to the plan and score table | Use chatId to distinguish sessions and compare the three materials' versions and signatures |
| External entry and access scope | The visitor identity matches the configured access scope | Access through the published entry with a restricted identity and compare the visible results with the configured scope |
Three common errors
- Symptom: the same candidate has different names or counts in the score table and adjustment list. Check the separately entered records against their names, counts and dates.
- Symptom: the plan has changed while the list still refers to an older version. Update the source versions used by the conversation and verify the plan reference.
- Symptom: the adjustment list lacks signatures or acknowledgments. Complete its filing evidence, and check the published entry's authentication and access scope separately.
How to verify the setup
Run two consecutive questions and check whether the second question refers back to the intended context. Inspect the historical message range and prompt instructions used by that run. Separately access the published entry with a restricted identity and compare the returned materials with the configured access scope.
- Check that the configured memory turns match the preceding context needed by the question.
- Record the two questions, messages, business variables and chatId, then compare the inputs, outputs and usage records of the two runs.