AI-103 Practice Question 35: Tool Input Accuracy Evaluator
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 4 · question 5 of 10
A booking agent is being hardened before release. Trace review shows a consistent pattern: the agent reliably picks the correct tool for each request and never calls a tool it doesn't need, and the tools themselves return HTTP 200. The failures come from the arguments: dates passed in the wrong format, an optional field populated with a value the schema doesn't allow, and occasionally a required parameter left out entirely.
Which built-in agent evaluator targets this failure directly?
Show answer and explanation
Answer:
Tool Input Accuracy is the process evaluator that "measures if all tool call parameters are correct across six strict criteria: groundedness, type compliance, format compliance, required parameters, no unexpected parameters, and value appropriateness." Those six map one for one onto the reported symptoms: a malformed date is format compliance, a disallowed value is value appropriateness, a missing field is required parameters. It takes query, response and tool_definitions, and returns a binary pass/fail.
- A is wrong because Tool Selection measures whether the agent chose the correct tools without selecting unnecessary ones, the one thing the traces show it already does well. - B is wrong because Tool Call Success detects technical errors and exceptions from the tool. These calls return 200; the data going in is what is wrong. - C is wrong because Task Adherence is a *system* evaluator asking whether the agent's actions follow the rules and constraints in its instructions. It works at the level of the whole task, not the schema of an individual argument.
Reference: Agent evaluators for generative AI