How much of a tool call and of a tool result may reach the conversation.
A result is the evidence the agent gathered, and what a judge reads of it is
short: command names, URLs and ids, which stand at the start of the output.
A call carries what the agent wrote (a file, a report, a test), which is the
work under judgement, so it keeps far more room, and there are few such
calls in a run.
How much of a tool call and of a tool result may reach the conversation.
A result is the evidence the agent gathered, and what a judge reads of it is short: command names, URLs and ids, which stand at the start of the output. A call carries what the agent wrote (a file, a report, a test), which is the work under judgement, so it keeps far more room, and there are few such calls in a run.