Field Name Developer Description Example Values
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 Is the data not approved yet?, Output missing again?, Why is the table empty?, This looks wrong, Seems outdated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