What new event type must be handled when combining tool use with streaming?	The input JSON event.
What are the two important properties contained within an input JSON event?	partial JSON and snapshot.
What does the partial JSON property represent?	A piece of JSON representing a part of an argument that Claude wants to send into a tool.
What does the snapshot property represent?	A cumulative sum of all the different partial JSON pieces received so far.
What is the primary issue with standard tool streaming when generating tool call arguments?	A significant delay followed by a single, large chunk of text appearing.
Why does the delay occur in standard tool streaming?	The Anthropic API buffers chunks to perform a validation step against the provided JSON schema.
How does the API perform JSON validation during tool generation?	It waits until a complete key-value pair is generated (e.g., the closing quote of a value) and validates that pair in isolation.
What API feature can be used to eliminate the delay in tool streaming?	Fine-grained tool calling.
What is the core function of fine-grained tool calling?	It disables the JSON validation step on the API.
What critical development practice must be implemented when using fine-grained tool calling?	Robust error handling, as the API may send invalid JSON.
What was the original requirement for the `meta` field in the JSON schema?	It had to be an object containing a `word count` (number) and a `review` (string).
What is the current data type of the `meta` field?	It is now a string.
How easily can tool streaming be integrated into an existing stringing pipeline?	It can be easily added.
What causes a delay when generating a large top-level key-value pair using default tool streaming?	The API performs a validation step.
What is the primary benefit of enabling fine-grained tool streaming?	It provides a more classic streaming experience.
What is the trade-off when using fine-grained tool streaming?	It incurs the cost of the validation step.