| Chat Session ID |
Unique identifier for the session — helps group related prompts/responses. |
CSID001, CSID002, CSID003, CSID004, CSID005 |
| User Prompt |
The exact natural language input from the user. |
Show yield for Batch 662, List vet issues from last week, Show cost per kg by cut, Export all batch data, Compare yield vs target |
| User |
The name or ID of the user who submitted the prompt. |
John Smith, Alice Ray, Mike Johnson, Priya Patel, Sam Chen |
| Question Type |
General category of the user's question type (used for routing and reporting). |
B/H Yield, Cost Summary, Batch Lookup, Compliance, Financial Variance |
| Tags / Context |
Relevant tags used for classification, routing, or training purposes. |
Yield, Batch 662, Liver Fluke, Angus Scheme, Skellingthorpe |
| Tool Used |
Tool or function used to generate the answer (internal system reference). |
Gold_Yield_Query_2, Batch_Kill_Info, HealthCheck_VetTool, Cost_Summary_Calc, Yield_Trend_Tool |
| Query/Output |
Expected output format returned to user (e.g., table, chart, sentence). |
Batch Summary Table, Line Chart, CSV Export, Text Summary, Pivot View |
| Query Detail Level |
Level of data aggregation the output should reflect. |
Batch-Level Summary, Product-Level Breakdown, Scheme Overview, Weekly Rollup, Month-on-Month |
| Fields Used |
List of fields required to answer the question (expected). |
Batch No, Input Weight, Output Total, Yield %, Cost per kg |
| Fields Accessed |
Fields the system actually accessed to try and answer the question. |
Batch No, Output Total, Yield %, Scheme, Product Code |
| Field Coverage |
Whether the fields accessed matched the required ones (Full / Partial / None). |
Full, Partial, None |
| Model Used |
AI model used to generate the response. |
GPT-4 Mini, GPT-4 Turbo, Claude 2, GPT-3.5, OpenAI Functions |
| AI Response |
The raw response from the model. |
I cannot find complete data for 662, Yield for 662 is 73.4%, Missing field for Output Total, Invalid batch reference, Data not yet approved |
| Confidence |
Confidence level reported or assigned to the model’s output. |
34%, 72%, 95%, 88%, 56% |
| Correct Output? |
Was the model’s response correct? Yes / No. |
Yes, No |
| User Feedback |
User’s feedback comment (free text) on result quality or issues. |
Is the data not approved yet?, Output missing again?, Why is the table empty?, This looks wrong, Seems outdated |
| Resolved Intent |
Clarified intent behind the prompt after review. |
Batch Yield Lookup, Scheme Comparison, Vet Flag Summary, Cost Analysis, Mapping Error Flag |
| Root Cause |
Identified cause of failed or poor output (e.g. mismatch, missing data). |
Tool unavailable, Data missing, Misunderstood intent, Mapping error, Vector mismatch |
| Permission Issue |
Was there a permissions-related issue? Yes / No. |
Yes, No |
| Tool Status |
Operational status of the tool at query time. |
Operational, Unavailable, Error, Restarting, Idle |
| Training Action Type |
Type of training improvement action taken. |
Alias Added, Mapping Rule Added, Prompt Updated, Model Re-trained, Tool Fix Required |
| Training Prompt Notes |
Additional explanation of what was done or changed in training. |
Clarified alias for batch lookup, Updated mapping rule, Model confused Yield vs Output, Added fallback, Trimmed stop words |
| Status |
Training outcome status (accepted, rejected, etc.) |
Used in Training, Not Reviewed, Accepted, Rejected, For Review |
| Action Taken |
Actual changes or actions performed to improve outcome. |
Prompt updated, alias added, Re-trained with feedback, Adjusted query logic, No action taken |